At this point, 99% of code in our engineering team is written by AI. The common fear is that this leads to unmaintainable AI slop. Our experience is the exact opposite. AI actually allows us to deliver high-quality code at high speeds. As resource constrained engineers we used to always have to compromise between velocity and quality - now we feel we can have both.

The models have certainly become smarter, but the tooling has also become significantly better. This combination has transformed our entire development workflow. In fact, traditional Agile sprints aren’t a thing for us anymore. We have moved beyond the old rituals into a continuous stream of execution.

Here is what currently works for us, after much experimenting:

1. We prioritize planning with AI

Even though vibe coding with AI felt like we moved fast initially, we quickly realized that relying solely on iterative coding often meant babysitting the agent all day. Instead, we now shift the heavy lifting to the planning phase. Planning now takes precedence over coding.

Sometimes, we spend an entire day just on planning. We iterate the plans with multiple models and AI providers. We refine the architecture until we have the perfect specification. We write multiple detailed tickets with clear, atomic requirements in Linear.

Once the plan is solidified, the building step becomes trivial. We assign Linear tickets in bulk to Cursor agents. We sometimes dispatch tickets to 20 agents at a time to run in parallel. The AI handles the implementation. It writes the boilerplate, connects the APIs, and builds the UI components. This elevates our engineers from “coders” to “system architects” who manage queues and review logic rather than syntax.

2. We let AI find its own work

A proactive codebase is better than a reactive one. Waiting for bugs to be reported is inefficient. We let the AI play offense. We often use Cursor (also works with other coding tools such as Claude Code / Codex etc) to proactively scan our codebase for opportunities. It looks for tech debt, refactors, potential performance improvements, and even whole new product features that are implied but missing.

Once identified, we use Linear MCP to automatically generate backlog tickets based on these findings. Our most important task as an engineering team is then to prioritize these tickets to ensure the highest impactful work is done first.

Bonus: We also generate the agent prompt directly in the ticket description. This creates a self-sustaining loop where the ticket is ready to be assigned back to a Cursor agent immediately. This reduces the friction of starting new work to near zero.

3. We trust AI but always verify

Writing code with AI is incredibly easy and fast but speed without stability is basically racing towards AI slop and technical debt. Agents are trusted, but everything is verified. We use strict linters and rigorous type checking. A strict mandate of 80-100% minimum test coverage has been established as non-negotiable (a standard that has always been hard to prioritize and enforce in traditional engineering teams).

We rely heavily on pre-commit hooks and GitHub Actions. The workflow is binary: the generated code simply does not merge until it passes every single check. There are no exceptions for small fixes. If the coverage drops or the linter complains, the agent must fix it before a human ever reviews it. This preserves our team’s mental energy for high-value problem solving.

Bonus: Reliance solely on human review is risky, as we can suffer from fatigue (especially from the volumes of AI-generated code we need to review). Therefore every PR gets reviewed in depth by an AI code reviewer (we use CodeRabbit but Copilot and Cursor also support this) before a human looks at it. It analyzes the logic, checks for security vulnerabilities, and ensures consistency. We in fact also have a branch rule that all of CodeRabbit’s comments need to be resolved before a PR can get merged.

4. We codify hard earned lessons for AI

We don’t just hope the AI understands our style; we write it down. We guide the agents with strict, project-specific rules files (we like .cursorrules but there’s also AGENTS.md). For example, one of the key agent rules we have for each project is to commit and push after every meaningful change. This ensures we have a granular history for easy rollbacks when things go wrong (which they often do).

We treat these rules as living documents of our institutional knowledge. Each time we encounter a new issue or a repeating pattern of failure, we add a specific constraint to the rules to prevent it from happening again. Over time this produces a body of work that allows us to move even faster because we’re not bogged down by repeated mistakes of the past.

5. We ensure AI has a great Developer Experience

Investing in Developer Experience (DX) has always been key to productivity. With AI, it is the difference between a toy and a tool. A flaky environment confuses a human, but it breaks an agent.

We invest heavily in robust testing harnesses and puppeteering tools that allow the AI for example to control the browser directly or to easily spin up a docker environment. This enables the agent to do more than just write code; it can run the app, click through flows, verify the UI, and iterate until the solution actually works. When the DX is optimized for machines, the AI becomes exponentially more powerful because it can close its own feedback loop.

Of course, we are still learning, breaking things, and rebuilding them better. So while the above might still evolve, it’s just what currently works for us.

I’m curious to know what part of the AI-native transition is clicking for your engineering team, and where you think we might be getting it wrong. Let’s compare notes!

Europe often prides itself on being the “good boy” of global tech regulation, championing ethics, privacy, and accountability. But as the AI revolution accelerates, this virtue threatens to become a liability.

The landmark AI Act, crafted in good faith to protect citizens, now threatens to suffocate the very innovation it sought to guide. Its stringent requirements around copyright compliance, transparency, and risk mitigation are already creating an innovation chokehold. Startups, already scrambling to compete with well-funded U.S. and Chinese giants, now face a labyrinth of legal hurdles. The result? A growing exodus of talent, capital, and ambition, leaving Europe at risk of becoming a bystander in the defining technological shift of our era.

The stakes couldn’t be higher. As Apple, OpenAI, Meta and Anthropic increasingly withhold cutting-edge AI features from European users over regulatory fears, the U.S. solidifies its dominance in foundation models, while China races ahead with pragmatism (and minimal copyright scruples). Europe’s choice is stark: adapt or accept irrelevance.

The Problem: How the AI Act is Holding Back AI Innovation

The AI Act’s most controversial mandate, requiring developers to prove their models were trained on copyright-free or publicly licensed data, sounds reasonable in theory. But for startups, it’s a logistical and financial nightmare. Auditing massive datasets for copyright compliance is a herculean task, one that only deep-pocketed corporations can afford.

But that’s just the beginning. The Act’s broad definition of AI casts a wide regulatory net, potentially pulling even low-risk applications into its compliance regime. This overreach creates uncertainty for developers, making it difficult to determine what rules apply and discouraging innovation in areas that pose minimal risk.

The high compliance costs further stack the odds against startups. Conformity assessments, extensive documentation, and continuous monitoring (all mandatory under the Act) place an enormous financial burden on smaller companies. While industry giants can absorb these costs, startups with limited resources face a crushing disadvantage.

Then there’s the issue of delayed time-to-market. The EU’s extensive regulatory hurdles slow down AI development cycles, making it nearly impossible for agile startups to compete in a global landscape where speed is everything. In a field where first-mover advantage is critical, these bureaucratic delays could spell doom for European-born AI ventures.

Even Big Tech is wary, not just small startups. If the world’s largest and wealthiest tech companies find the AI Act too restrictive, what chance do startups have? Meta recently disabled its AI assistant, Meta AI, for European users, citing “regulatory uncertainty.” OpenAI quietly restricts access to voice cloning tools in the EU, and Google’s Gemini avoids entire markets over compliance fears. These aren’t small players struggling to keep up, these are trillion-dollar companies deciding that Europe’s rules are simply too difficult, too risky, or too costly to comply with.

Meanwhile, talent flees. European AI researchers increasingly decamp to Silicon Valley or Shenzhen, where funding flows freely and experimentation isn’t shackled by preemptive red tape and legal ambiguity. The Act’s vague classifications of AI risk levels leave developers second-guessing their own innovations, making compliance a constant guessing game rather than a clear roadmap.

In its attempt to lead on regulation, Europe may have written itself out of the AI race altogether.

The Geopolitical Stakes: US & China Are Pulling Ahead

The global race for artificial intelligence (AI) supremacy is intensifying, with the United States and China adopting distinct strategies that are propelling them ahead of Europe.

• United States: Private Sector Dominance and Deregulation: In the US, the AI landscape is driven by private sector innovation. Companies like OpenAI, Anthropic, and Google lead with advanced models such as GPT-4, Claude 3, and Gemini, respectively. The U.S. government's approach has been to minimize regulatory barriers, fostering an environment where AI research and development can thrive. In a significant move to bolster AI infrastructure, President Donald Trump recently announced the Stargate initiative, a joint venture involving OpenAI, SoftBank, and Oracle. This project plans to invest up to $500 billion in AI infrastructure, including the construction of data centers and energy generation facilities across the United States. The initiative underscores the administration's commitment to accelerating AI development and maintaining a leading position in the global AI race.

• China: State-Backed Ambitions and Rapid Implementation: China's approach to AI is characterized by substantial government support and a focus on rapid deployment. The Chinese government has committed significant funding to AI initiatives, including a $50 billion semiconductor fund to bolster its technological infrastructure. Companies like DeepSeek have emerged, developing state-of-the-art models efficiently and cost-effectively. This progress is facilitated by China's more lenient data regulations, allowing for swift integration of AI technologies across various sectors.

• Europe: Risk of Becoming a Consumer Rather Than a Creator: In contrast, Europe faces challenges that could relegate it to the role of an AI consumer rather than a creator. Despite having a large consumer market and abundant talent, the EU lags behind global leaders like China and the US. Recurrent issues include excessive focus on AI risks and stringent regulations that stifle innovation. Notably, groundbreaking AI tools, such as OpenAI's Operator, are inaccessible in the EU due to regulatory delays, hampering competitiveness. Without cultivating homegrown AI champions, Europe risks dependence on foreign technologies for critical applications, from healthcare algorithms to defense systems, thereby ceding economic power and strategic autonomy.

The Silver Lining: Europe’s Counteroffensive

Not all hope is lost. A new wave of investment and ambition is brewing, led by France.

• Macron’s €109 Billion Initiative: France is making AI a national priority under President Emmanuel Macron's leadership. He has announced a €109 billion investment plan to bolster the country's AI capabilities. This initiative includes substantial funding from international investors, such as €50 billion from the United Arab Emirates and €20 billion from Canadian firm Brookfield. A significant portion of these funds is allocated for developing advanced data centers, leveraging France's nuclear energy infrastructure to provide efficient and sustainable power solutions. This strategy not only enhances computational capacity but also positions France as a leader in eco-friendly AI development.

• The EU’s €200 Billion InvestAI Program: Complementing national efforts, the European Union has launched the InvestAI initiative, aiming to mobilize €200 billion for AI advancement. This program focuses on establishing AI gigafactories, large-scale facilities equipped with cutting-edge technology to train sophisticated AI models. The initiative seeks to foster collaboration among member states, reduce reliance on external tech giants, and ensure that Europe remains competitive in the rapidly evolving AI landscape.

• The EU-Inc Initiative: Streamlining Business Across Borders: In a bid to further enhance competitiveness, an online community led by Philipp Herkelmann has proposed the creation of "EU-Inc," a unified, digital-first private limited liability company structure for the European Union. This initiative aims to address the challenges of fragmented national regulations by establishing a standardized corporate framework governed by EU regulations. Key features include no minimum capital requirement, flexible share structures, and streamlined cross-border operations. The proposal seeks to make it easier for businesses to operate across EU member states, thereby fostering innovation and economic growth.

• Learning from China’s DeepSeek: Europe is also drawing inspiration from international success stories. For instance, China's DeepSeek has demonstrated that significant AI advancements can be achieved with leaner, more cost-effective models. This approach underscores the potential for European startups to innovate efficiently, provided they navigate the regulatory environment effectively.

Through these concerted efforts, Europe aims to not only catch up with but also lead in the global AI race, balancing innovation with ethical considerations and strategic autonomy.

A Call for Regulatory Balance: Learning from Mistakes

The AI Act’s rigid early framework needs urgent recalibration. France is already pushing for carveouts to protect startups, arguing that “innovation cannot thrive under fear.” Key fixes could include:

• Regulatory Sandboxes: The AI Act proposes the establishment of coordinated AI 'regulatory sandboxes' across EU member states. These sandboxes are controlled environments where businesses can develop, test, and validate AI systems under the supervision of national authorities before market deployment. This approach allows innovators to experiment with new technologies while ensuring compliance with regulatory standards. It also provides regulators with insights into emerging technologies, facilitating a balanced oversight mechanism.

• Public Data Commons: To alleviate the burden of sourcing high-quality, copyright-free data, the EU could establish publicly funded repositories. These public data commons would offer startups access to extensive datasets necessary for training AI models, thereby reducing costs and promoting innovation. Such initiatives would ensure that smaller enterprises have the resources to develop competitive AI solutions without the prohibitive expenses associated with data acquisition.

• Talent Fast Lanes: Attracting and retaining top AI talent is crucial for Europe's competitiveness. Implementing streamlined visa processes and offering tax incentives for AI professionals and researchers can make the EU a more appealing destination for global talent. These measures would help build a robust AI ecosystem, fostering collaboration and accelerating advancements within the region.

By adopting these strategies, the EU can strike a balance between maintaining ethical standards and promoting a dynamic, innovation-friendly environment for AI development.

The Opportunity: Europe’s Agentic AI Frontier

While the EU has fallen behind in foundational AI model development, the next wave of AI innovation is still up for grabs: Agentic AI, autonomous systems that act, reason, and collaborate. Unlike traditional AI models that require constant human input, these AI "agents" can execute tasks end-to-end, from managing supply chains to running customer service operations.

This shift toward Agentic AI is why I launched this newsletter. As an engineer and entrepreneur working at the intersection of AI and automation, I’ve seen firsthand how agentic systems can transform industries. While much of today’s AI conversation is dominated by large language models, the real opportunity lies in designing AI that doesn’t just generate text but acts autonomously to solve real-world problems. Europe, with its strengths in robotics, industrial automation, and ethical AI, is uniquely positioned to lead this transformation.

Why Agentic AI?

1. Lower Computational Demands: Unlike massive foundation models that require enormous GPU clusters, agentic AI is designed to be modular and efficient. These systems focus on task-specific reasoning rather than brute-force computation, making them cheaper to deploy and more accessible to startups.

2. Alignment with European Expertise: Europe has long been a leader in robotics, IoT, and industrial automation, all fields where agentic AI can thrive. Unlike the U.S., where AI is largely consumer-focused, Europe can build AI systems that enhance factories, optimize logistics, and power next-gen smart infrastructure.

3. Ethical Leadership
   The EU’s emphasis on trust, safety, and transparency is a competitive advantage. In a world increasingly wary of AI’s black-box decision-making, Europe can become the go-to supplier of responsible, explainable AI that companies and governments can trust.

Agentic AI is still at the frontier of innovation. While the U.S. dominates foundation models and China races ahead in scale, Europe can carve out its own leadership niche by pioneering intelligent, autonomous systems that power the real economy.

This is the mission behind Designing Agentic Systems, to explore how AI agents can reshape industries, unlock new entrepreneurial opportunities, and drive the next wave of innovation. Europe still has a shot at winning this race. But it needs to act—now.

Imagine you’re a marketing manager in charge of a direct mail campaign. Your goal is to target high-income households in Austin, Texas—a diverse city of nearly 1 million residents. You have access to a database of over a million households, each with detailed demographic data: income, age, housing type, and more.

Your challenge? Create a segment that includes at least 1,000 households but no more than 10,000, which also needs to align closely with your Ideal Customer Profile (ICP) to maximize relevance and impact.

This isn’t a simple task. With dozens of demographic filters and numerous possible configurations, the combinations quickly explode into billions. To make matters worse, these filters aren’t independent—certain income groups correlate with specific neighborhoods, and younger families are unlikely to live in retirement areas.

Even if you spent days manually researching and testing filters, you’d constantly question whether you’d found the best audience. What if a better combination exists? This uncertainty is what makes audience segmentation such a time-consuming and challenging problem.

Why Manual Segmentation Is Challenging

Segmentation is one of the hardest tasks in marketing because it requires balancing two competing forces: scale and precision. On one hand, you need a segment large enough to justify your campaign, but on the other, it must be narrow enough to target the right audience effectively.

If you were to tackle this task manually, here’s what the process might look like:

1. Research the Region: You’d start by googling demographic data about Austin. After hours of work, you might find that the median household income is $71,500, the city skews younger, and there’s a mix of renters in urban areas and homeowners in suburbs.

2. Propose Initial Filters: Based on your research, you’d guess that your target audience might be:

   • Age: 25–40

   • Income: $75,000–$150,000

   • Housing Type: Single-family homes or luxury apartments

   But these are just educated guesses. There’s no guarantee this segment will work, and the process of refining it could take hours—or even days.

3. Test and Iterate: You’d query your database repeatedly, tweaking filters to see how many households match. If one set of filters results in 50,000 households, it’s too large. If another results in 500, it’s too small. This trial-and-error process is slow and frustrating, and even after days of effort, you’d still wonder: Did I miss a better option? And even after finding the right number, you need to ensure it’s relevant to your Ideal Customer Profile (ICP).

How Agentic AI Can Tackle This Problem

Modern large language models come equipped with world knowledge—a built-in understanding of relationships between concepts, based on the vast data they’ve been trained on. This allows them to make intelligent guesses about demographic patterns, similar to how an experienced marketing manager would, but with far greater speed and scale.

For instance, an LLM might “know” that high-income households are concentrated in suburban areas or that younger professionals tend to rent apartments downtown. Instead of starting from scratch, it uses this knowledge to propose meaningful filter combinations right away.

But AI isn’t perfect—it can sometimes “hallucinate,” offering insights that sound plausible but aren’t grounded in reality. However, this challenge can be addressed by allowing the AI agent to autonomously conduct online research by querying sources like Google to validate and enhance its recommendations with real-world data.

By combining grounded insights with its world model, the AI agent avoids hallucinations and ensures its recommendations are tied to real-world data.

Iterative Refinement with Bayesian Inference

Once the AI agent proposes a set of filters, the next step is to test them against your database to see how many households match. This is where Bayesian inference comes into play—a statistical method that helps the AI agent refine its understanding based on new evidence.

Think of Bayesian inference as a feedback loop:

1. The AI agent starts with an initial “belief” about what filters will work (e.g., income $75,000–$150,000, age 25–40).

2. It tests these filters and gets a result (e.g., 5,000 households match).

3. It updates its belief based on this result, refining the filters to get closer to your goal.

This iterative process ensures the AI agent doesn’t waste time on irrelevant combinations. Instead, it continuously learns and adjusts, converging on the right segment far faster than any human could.

The Science Behind the Process

Agentic AI’s ability to tackle complex segmentation problems is grounded in well-established scientific principles:

• Reducing Uncertainty (Shannon Entropy): At the start, the AI agent faces high uncertainty with billions of possible filter combinations. Its goal is to systematically reduce this uncertainty by exploring the most promising options first.

• Systematic Exploration (Ergodic Theory): The AI agent ensures that, given enough iterations, it will explore all relevant possibilities. This doesn’t mean brute-forcing every combination—it means intelligently narrowing the search space to focus on areas most likely to succeed.

• Iterative Learning (Bayesian Inference): Each test provides new data, allowing the agent to update its strategy and refine its filters in real time. This dynamic learning process mirrors how humans learn from feedback but operates at a much faster scale.

Together, these principles explain why agentic AI can solve segmentation problems in minutes, not days.

The Result: Smarter Segmentation in Minutes

Let’s go back to our example of targeting Austin, Texas. Using agentic AI, the process might look like this:

1. The AI agent conducts online research to gather demographic insights about the city.

2. It uses its world model to propose an initial set of filters based on these insights.

3. It tests the filters against your database, refining them through iterative feedback loops until it finds a segment that meets your constraints.

In minutes, the AI agent delivers a segment like this:

• Age: 25–40

• Income: $75,000–$150,000

• Housing Type: Single-family homes or upscale apartments

• Segment Size: 3,200 households

What would have taken days of manual work—and still left you questioning the results—is now done in a fraction of the time, with greater accuracy and confidence.

Taking Personalization and Insights to the Next Level

Agentic AI doesn’t just stop at segmentation. It opens the door to unprecedented levels of precision and creativity, allowing marketers to achieve what once seemed unattainable. Here are a few possibilities that can unlock things that were previously impossible or unfeasible:

• Hyper-Personalization: Conversion rates for direct mailing campaigns are typically very low. Agentic AI can massively improve your conversation rates by utilizing hyper-personalization. Imagine crafting individual postcards tailored to each individual recipient’s profile—not just using their first name but designing the text, image, and messaging to resonate deeply with their lifestyle and your brand’s messaging. For instance:

  • Family Households with Children: The postcard might highlight benefits or products that appeal to parents, using imagery of happy families and messaging that emphasizes safety and comfort.

  • Young Single Professionals: These recipients could receive sleek, modern designs featuring messaging that highlights efficiency, independence, or career growth.

• Simulation of Consumer Surveys: Additionally, Agentic AI can simulate hundreds or thousands of consumer surveys for advanced insights. Normally, crafting 50 micro-segments and surveying representative consumers would be expensive and time-consuming. With AI, you can simulate these interactions, gaining a statistical approximation of customer preferences and messaging effectiveness. While no model is perfect, these simulations provide valuable insights without the time and cost overhead of traditional surveys.

Agentic AI helps marketers achieve what once seemed impossible: hyper-personalized messaging and scalable, data-driven insights, all within the constraints of modern campaigns.

Why Agentic AI Matters for Marketers

By combining the intelligence of LLMs with real-world grounding and iterative refinement, it tackles segmentation challenges that previously felt impossible.

For marketers, this means:

• Fewer Hours Spent on Repetitive Tasks: The AI agent handles the heavy lifting, freeing you to focus on strategy.

• More Confidence in Your Results: Grounded insights and systematic refinement ensure accuracy.

• Scalability Across Campaigns: With AI agent, you can apply this process to multiple regions or audiences simultaneously.

If segmentation has ever felt like an impossible puzzle, agentic AI is the tool that solves it—quickly, intelligently, and at scale.

Marketing is just one of many professions poised to be transformed by agentic AI. Stay tuned to learn how this powerful technology can revolutionize your work. Subscribe to this newsletter for more insights and practical guides on implementing agentic AI across industries.

Subscribe now

In one of my previous articles, AI is Eating the Jobs, I explored how agentic AI workflows are rapidly replacing entire teams, ushering in a new era of automation that is as transformative as it is disruptive. This is not a new story—automation has been reshaping industries since the Industrial Revolution, from the assembly line to the rise of computers. More recently, trends like Software-as-a-Service (SaaS), Robotic Process Automation (RPA), and Low-Code/No-Code platforms promised to streamline workflows and make businesses more efficient.

The goal of automation has always been to do more with less—but today’s tools aren’t solving problems; they’re creating new ones.

While SaaS platforms became ubiquitous as the backbone of modern business operations, they still depend heavily on human operators to configure, manage, and extract value. RPA and Low-Code/No-Code (LCNC) systems followed, aiming to push automation even further, but they have struggled to live up to their promise. Instead of enabling businesses to scale seamlessly, these tools have left many companies grappling with inefficiency, high maintenance costs, and limited flexibility.

The Shortcomings of SaaS

SaaS platforms revolutionized how businesses use software, offering flexibility and scalability without the headaches of traditional installation or maintenance. Tools like Salesforce, Slack, and HubSpot became the backbone of modern business operations, promising streamlined workflows and improved collaboration.

These tools often fall short when applied to internal automation. They rely on users to manually configure workflows, integrate systems, and monitor operations. For instance, CRMs require sales teams to spend hours inputting and maintaining data, while managers must constantly optimize workflows for reporting and lead tracking. This effort shifts the workload rather than removing it.

Instead of automating work, SaaS often turns employees into operators.

Another issue is fragmentation. Most businesses depend on multiple tools for different functions—one for project management, another for customer support, and yet another for HR. These systems rarely integrate seamlessly, forcing employees to toggle between them and manually move data.

Tool overload adds to the complexity. HR teams might use separate platforms for payroll, recruiting, and employee engagement, each with its own setup and maintenance needs. This patchwork of disconnected workflows creates inefficiencies and demands constant oversight, leaving businesses far from the streamlined operations they envisioned.

The promise of SaaS was to simplify, but its reality often ties automation to human effort.

The Broken Promises of RPA

Robotic Process Automation (RPA) emerged in the 2000s as a solution for handling repetitive workflows. Companies like UiPath, Automation Anywhere, and Blue Prism built multi-billion-dollar businesses on the idea that bots could mimic human actions—clicks, keystrokes, and form fills—to complete tasks like data entry, invoice processing, and customer service triage.

RPA promised to simplify workflows, but in reality, it often creates more work than it saves.

The problem lies in how RPA operates. Bots follow rigid scripts, navigating software interfaces based on predefined rules. These systems work well when processes remain static, but the moment something changes—a form layout updates, a button is renamed, or a vendor introduces a new invoice format—the bot fails. Companies then spend valuable time and resources fixing broken automations.

Take customer service as an example. While bots can triage basic tickets, they often misclassify complex requests, leading to longer resolution times and frustrated users. Similarly, RPA systems used for data migrations frequently stumble on edge cases, requiring constant human intervention.

These issues expose RPA’s fundamental flaw: it doesn’t understand the work—it merely mimics it. This brittleness has left many businesses questioning whether RPA is worth the investment, especially when ongoing maintenance costs often exceed initial implementation savings.

The Frustrations of Low-Code/No-Code

Low-Code/No-Code (LCNC) tools—offered by companies like Retool, Airtable Apps, and Microsoft Power Apps—were designed to democratize software development. Their promise was simple: empower non-technical users to build custom workflows without relying on developers.

For simple use cases, LCNC tools work well. Users can quickly create dashboards, build forms, or design approval workflows. But as workflows grow in complexity, these platforms struggle. Many “no-code” tools require users to write scripts for advanced functionality, undermining the very premise of no-code. Meanwhile, more developer-centric platforms, like Retool, are inaccessible to non-technical teams, leaving businesses stuck between two extremes.

LCNC platforms claim to lower the barrier to entry, but they often introduce new bottlenecks instead—ironically requiring the very expertise they were meant to replace.

Another major issue is the lack of opinionated defaults. Most LCNC platforms provide blank canvases, requiring users to define everything themselves—from data models to integrations. This leads to inconsistent implementations across teams, inefficiencies, and overreliance on technical support.

For example, an HR team might use an LCNC platform to build a performance review tracker, but without predefined templates or best practices, the result is often clunky, hard to maintain, and unable to scale. Similar problems arise when customer support teams attempt to create integrated dashboards without clear guidance.

What the Next Generation of Automation Should Look Like

To truly transform how work gets done, the next generation of automation tools must combine the adaptability of horizontal platforms with the specificity of vertical solutions. Generative AI makes this possible by enabling tools that are both flexible and deeply integrated.

Next-generation automation must be AI-native, opinionated, and seamlessly integrated into enterprise systems.

An ideal platform would:

1. Leverage AI for workflow creation: Users describe their needs in plain language, and the platform generates a working prototype.

2. Provide opinionated frameworks: Prebuilt templates and best practices guide users, reducing complexity and ensuring consistency.

3. Ensure seamless integration: Robust connections to systems of record like CRMs, ERPs, and HR tools ensure enterprise readiness.

4. Offer enterprise-grade governance: Built-in role-based access controls (RBAC), audit logs, and monitoring tools provide visibility and control.

In upcoming issues, we’ll dive deeper into how to design agentic AI workflows that deliver on this vision.

Subscribe now

Businesses today face mounting complexity in their operations, from supply chain disruptions to rapidly evolving customer expectations. Traditional automation systems, such as Software-as-a-Service (SaaS), Robotic Process Automation (RPA), and Low-Code/No-Code (LCNC) platforms, have promised transformation but often fall short of delivering true efficiency. These tools frequently require heavy manual intervention, struggle with adaptability, and create inefficiencies rather than solving them. Multi-agent systems offer a way forward: a network of specialized agents that interact, refine their outputs, and create a continuously improving system.

OpenAI’s recent announcement of its o3 reasoning model highlighted a significant breakthrough, achieving an unprecedented score on the ARC-AGI benchmark. Many view this as a significant step toward the broader vision of Artificial General Intelligence (AGI), where systems exhibit the ability to generalize and adapt across diverse challenges.

Insights from the latest research1 into the o1 model’s inner workings—such as its modular reasoning, dynamic context adaptation, and efficient knowledge integration—offer valuable lessons for creating networks of agents that interact dynamically and solve complex problems. By applying these principles, we can design more robust and adaptive multi-agent systems capable of tackling real-world challenges.

Designing Agentic Systems for the Next Levels of AGI

To understand the trajectory of advanced AI systems, we can use the framework that OpenAI uses internally to guide their path toward AGI development. Each level builds upon the last, moving from basic interaction to systems capable of managing complex operations autonomously:

• Level 1 (Chatbots): Conversational AI is like the entry-level customer service representative who can handle FAQs, route inquiries, and manage basic tasks. These systems excel in providing quick, straightforward answers but lack the capacity for deeper problem-solving or strategic insights.

• Level 2 (Reasoners) ← we’re here: Reasoners function like a skilled analyst in your organization. With OpenAI's recent introduction of o3, AI reasoning is reaching a level of intelligence where systems can tackle novel challenges effectively, paving the way for more autonomous decision-making and complex problem-solving.

• Level 3 (Agents): Agents are comparable to a highly competent project manager. For example, an AI-driven supply chain agent could reorder inventory, adjust shipping schedules, and negotiate with vendors without human intervention.

• Level 4 (Innovators): Innovators take on the role of a creative product designer or strategist. These systems generate novel ideas and develop groundbreaking solutions, such as conceptualizing entirely new business models or product lines based on emerging market trends.

• Level 5 (Organizations): Organizational Equivalents represent an autonomous C-suite, capable of running the entire organization. These systems manage workflows, make strategic decisions, and adapt to market shifts—essentially functioning as an independent enterprise.

Achieving Levels 3 through 5 will require robust multi-agentic designs that inherit adaptability from reasoning models while introducing capabilities for creativity and strategic management.

The Four Key Principles for Multi-Agentic Systems

To bridge the gap between reasoning capabilities and fully autonomous multi-agentic systems, four foundational principles lay the groundwork for collaboration, learning, and adaptability among agents:

1. Policy Initialization: Policy initialization ensures agents start with a clear understanding of their role, equipped with the tools and training to succeed. Just as an HR department provides new hires with company guidelines and access to necessary software, this principle sets up agents with domain-specific expertise, ethical guidelines, and seamless integrations to begin their tasks effectively.

2. Reward Design: Reward design works similarly by aligning agents’ actions with the organization’s objectives. Agents are encouraged to prioritize collaborative outcomes and iteratively improve through feedback, much like how a sales team adjusts its strategies based on performance metrics.

3. Search Mechanisms: Search mechanisms empower agents to explore and refine strategies dynamically, ensuring that innovative and practical solutions emerge even under uncertainty. This is akin to iterating on a product concept until it fits market needs.

4. Learning Frameworks: Learning frameworks enable agents to continuously improve individually and collectively. These frameworks ensure that agents learn from their experiences and adapt to evolving challenges, fostering a culture of continuous improvement.

Let’s now explore these four key principles in greater detail.

1. Policy Initialization: Building Strong Foundations

Policy initialization is the process of equipping agents with the foundational tools and frameworks necessary to operate effectively in complex environments. Drawing from advanced reasoning models like o1, this principle focuses on three core components:

1. Dynamic Policy Creation: Agents are initialized with policies that adapt through iterative evaluation and reasoning. Techniques like Monte Carlo Tree Search enable agents to refine strategies by balancing exploration and exploitation, which is critical for decision-making in novel scenarios.2 This process is further enhanced by token-level action granularity, where agents operate at the finest level of decision-making, selecting individual tokens from a vast vocabulary to construct coherent responses.3 Much like a startup founder refining their business plan based on market feedback, agents dynamically adjust their policies to optimize performance. Additionally, step-by-step reasoning and alternative proposal mechanisms allow agents to explore multiple solutions and self-correct when errors are detected, significantly improving accuracy and adaptability.4

2. Integrated Knowledge Models: Agents are initialized with modular knowledge bases that combine pre-trained datasets and domain-specific updates. This approach integrates cross-domain knowledge with real-time data for situational accuracy, enabling agents to remain highly relevant and effective in their operational domains.5 Consider a consultant who tailors their advice to each client’s unique needs. Pre-training establishes language understanding, world knowledge acquisition, and basic reasoning abilities, while instruction fine-tuning transforms these capabilities into task-oriented behaviors.6 For example, multilingual training data enables cross-lingual transfer, and exposure to diverse corpora fosters domain expertise in areas like mathematics, science, and programming.7

3. Human-like Reasoning Behaviors: Agents are equipped with sophisticated reasoning behaviors that mimic human problem-solving. These include:

   • Problem Analysis: Reformulating and analyzing problems to reduce ambiguity and construct actionable specifications.8 It’s similar to a CEO breaking down a complex challenge into clear, actionable goals.

   • Task Decomposition: Breaking complex problems into manageable subtasks, dynamically adjusting granularity based on context.9 This mirrors a project manager dividing a large project into smaller milestones.

   • Self-Reflection: Evaluating and correcting outputs, enabling agents to recognize flaws and refine their reasoning processes.10 It’s like a company conducting a post-mortem after a product launch to identify lessons learned.

Policy Initialization Lessons from the o1 Reasoning Model

The o1 model highlights several key insights that can inform the policy initialization of agentic systems:

• Long-Text Generation: Agents must generate lengthy, coherent outputs to handle complex reasoning tasks. Techniques like AgentWrite and Self-Lengthen enhance long-text generation capabilities by fine-tuning on specialized datasets.11

• Logical Orchestration of Reasoning Behaviors: Agents need to strategically sequence reasoning behaviors, such as deciding when to self-correct or explore alternative solutions. Exposure to programming code and structured logical data strengthens these capabilities.12

• Self-Reflection: Behaviors like self-evaluation and self-correction enable agents to recognize and address flaws in their reasoning. This self-reflection capability is critical for improving accuracy and reliability.13

Policy initialization provides agents with the foundational policies, knowledge bases, and reasoning frameworks they need to perform effectively. This ensures that every agent operates seamlessly within the larger system, maintaining both efficiency and adaptability.

2. Reward Design: Aligning Actions with Goals

Reward design is the cornerstone of aligning agent behavior with system-wide objectives, ensuring that every action contributes to the overall goals of the organization. Drawing insights from reinforcement learning and advanced reasoning models, this principle involves three core concepts:

• Dynamic Reward Structures: Incentive mechanisms must evolve alongside the system's changing goals and environments. Techniques like policy-gradient methods and adaptive reward shaping allow systems to fine-tune incentives dynamically, ensuring agents remain aligned with overarching objectives14. For instance, a customer service agent might initially prioritize response speed but later focus on customer satisfaction as the system’s goals evolve.

• Collaborative Incentives: To foster teamwork, agents must be rewarded not only for individual success but also for contributions to collective outcomes. Methods for multi-agent credit assignment, as described in coordination algorithms, ensure agents work toward shared objectives like improving overall customer retention or maximizing team productivity. This is akin to how high-performing organizations balance individual bonuses with team-based rewards to encourage collaboration.

• Feedback Loops for Iterative Learning: By providing agents with continuous feedback on their performance, systems can refine agent behavior iteratively. Concepts like reward modeling, as highlighted in scalable alignment research, allow agents to adapt through both reinforcement signals and post-task evaluations15. For example, a sales agent might adjust its approach to prospecting based on reward adjustments for achieving long-term revenue goals. This iterative process is similar to how businesses use performance reviews and real-time analytics to optimize employee output.

Reward Design Lessons from the o1 Reasoning Model

The o1 reasoning model might provide some valuable insights into designing reward systems for agentic systems, particularly for complex, multi-task environments. Here are key lessons and challenges:

1. Process Rewards for Complex Reasoning: For tasks like mathematics and code generation, where responses involve long chains of reasoning, o1 likely employs process rewards to supervise intermediate steps rather than relying solely on outcome rewards16. This ensures that agents are rewarded for correct reasoning processes, even if the final outcome is incorrect. Techniques like reward shaping can transform sparse outcome rewards into denser process rewards, enabling more effective learning17.

2. Learning from Preference and Expert Data: When reward signals from the environment are unavailable, o1 might leverage preference data (e.g., ranked responses) or expert data to infer rewards18. This approach, inspired by methods like Inverse Reinforcement Learning (IRL), allows agents to learn from human-like decision-making patterns, even in the absence of explicit feedback19.

3. Robust and Adaptable Reward Models: o1’s reward model is likely trained on a large and diverse dataset, enabling it to generalize across domains. It can be fine-tuned using few-shot examples, making it adaptable to new tasks with minimal data. This flexibility is crucial for agentic systems operating in dynamic environments.

While traditional businesses use a mix of monetary incentives, career advancement, and recognition to drive performance, reward design creates a sophisticated framework of reinforcement signals that shape agent behavior. This multi-layered approach mirrors how the most effective organizations balance individual achievement with team success, creating a self-reinforcing cycle of continuous improvement and collective achievement.

3. Search Mechanisms: Exploring and Refining Strategies

Search mechanisms empower agents to navigate uncertainty, test strategies, and refine their approaches dynamically. Based on insights from the research paper, these mechanisms can be broken into five core concepts:

1. Guiding Signals: Internal vs. External Guidance
   Search relies on guiding signals to direct the exploration process. Internal guidance leverages the model’s own capabilities, such as model uncertainty or self-evaluation, to assess the quality of solutions. For example, self-consistency techniques use majority voting to select the most reliable answer from multiple candidates. External guidance, on the other hand, uses task-specific feedback, such as rewards or heuristic rules, to steer the search. This is akin to a business using customer feedback (external) versus internal performance metrics (internal) to refine its strategies.20

2. Search Strategies: Tree Search vs. Sequential Revisions
   Search strategies fall into two broad categories: tree search and sequential revisions. Tree search, like Monte Carlo Tree Search (MCTS) or Best-of-N (BoN), generates multiple candidate solutions simultaneously, exploring a wide range of possibilities. Sequential revisions, in contrast, iteratively refine a single solution based on feedback or self-reflection. Think of tree search as a brainstorming session where multiple ideas are generated at once, while sequential revisions are like polishing a single idea through multiple drafts.21

3. Exploration vs. Exploitation
   Effective search strategies balance exploration (trying new possibilities) with exploitation (leveraging known successful strategies). Techniques like epsilon-greedy algorithms ensure agents remain innovative while capitalizing on proven outcomes. For instance, during training, agents might explore diverse solutions (exploration) while also leveraging high-reward solutions (exploitation) to iteratively improve their policy. This is similar to a company investing in R&D (exploration) while maximizing profits from existing products (exploitation).22

4. Training-Time vs. Inference-Time Search
   Search plays distinct roles in training and inference. During training, search helps generate high-quality data by exploring diverse solutions, often guided by external feedback like task-specific rewards. In inference, search refines outputs through internal guidance, such as self-evaluation, to improve accuracy. For example, during training, a model might use MCTS to explore a wide range of solutions, while during inference, it might use sequential revisions to refine its final answer.23

5. Efficiency and Scaling Challenges
   Scaling search mechanisms introduces challenges, such as inverse scaling (where increased search degrades performance) and over-thinking on simple tasks. For example, using a reward model trained on limited data can lead to poor generalization during large-scale search. To address this, researchers propose techniques like speculative rejection (discarding low-quality solutions early) or combining tree search with sequential revisions to balance efficiency and performance. This is akin to a business optimizing its processes to avoid wasted effort while maintaining high-quality outcomes.24

Search Mechanism Lessons from the o1 Reasoning Model

The o1 reasoning model may offer some insights into optimizing search mechanisms, providing a blueprint for designing efficient and effective systems. Here are three critical takeaways:

1. Training-Time Search: Parallel Exploration with External Guidance
   During training, o1 likely employs tree search techniques like Best-of-N (BoN) and Monte Carlo Tree Search (MCTS) to explore multiple solutions simultaneously. This parallel exploration is guided by external feedback, such as task-specific rewards or environmental signals like code execution results, ensuring alignment with real-world performance. Systems can adopt similar strategies to efficiently explore and validate solutions during training, leveraging external feedback to refine their approaches.25

2. Inference-Time Search: Iterative Refinement with Internal Guidance
   During inference, o1 might shift to sequential revisions, iteratively refining outputs using internal guidance like self-evaluation or model uncertainty. This minimizes computational overhead while improving solution quality. By adopting iterative refinement, systems can optimize decisions in real-time without relying on costly external feedback, ensuring efficiency and accuracy.26

Just as military planners use simulations to evaluate tactical approaches or chess grandmasters analyze multiple move sequences, search mechanisms enable agents to explore decision trees and optimize their choices through systematic evaluation. Together, these mechanisms ensure that agents navigate complex decision spaces with confidence and precision, much like a well-oiled corporate strategy team.

4. Learning Frameworks: Continuous Improvement and Collaboration

Learning frameworks in multi-agent systems are not just about continuous improvement; they encompass a broader set of principles that enable agents to adapt, innovate, and scale effectively. Here are the core concepts, enriched with insights from the research:

1. Reflective Analysis: Agents evaluate their own performance to identify strengths and weaknesses, refining their strategies iteratively. This is akin to a business conducting a post-mortem analysis after a project to understand what worked and what didn’t. For example, an agent might analyze its past decisions to improve future actions, much like a sales team refining its pitch based on customer feedback.27

2. Collaborative Knowledge Sharing: Agents share their experiences and solutions, creating a collective intelligence that benefits the entire system. This mirrors how cross-functional teams in a company share insights to solve complex problems. When one agent discovers an efficient solution, it’s shared across the network, enhancing overall performance—similar to how a company’s R&D department shares innovations with other teams to drive organizational growth.

3. Adaptive Learning Algorithms: Agents continuously update their strategies based on real-time data, ensuring they remain effective in dynamic environments. This is akin to a business adapting its operations in response to market trends. For instance, an agent might adjust its decision-making process based on new information, much like a retail company dynamically changing its inventory based on consumer demand.28

4. Search-Driven Exploration: Agents use search algorithms (e.g., beam search, Monte Carlo Tree Search) to explore high-value actions rather than relying on random sampling. This ensures that the training data is of higher quality, similar to how a company uses data analytics to identify the most promising opportunities. The iterative process of search and learning—where the system refines its strategies based on search results—mirrors how businesses refine their operations based on performance metrics.29

5. Hybrid Learning Approaches: Systems often combine multiple learning methods, such as behavior cloning and policy optimization, to achieve the best results. Behavior cloning focuses on replicating successful actions, while policy optimization learns from both successes and failures. This hybrid approach is like a business starting with proven strategies before experimenting with innovative approaches to achieve breakthroughs.30

6. Scaling and Efficiency: As systems grow in complexity, ensuring efficient learning becomes a critical challenge. The process of generating and analyzing data can be resource-intensive, much like a company’s R&D efforts consuming significant time and budget. To address this, systems can reuse data from previous iterations or focus on the most promising areas of exploration, similar to how businesses optimize their operations to reduce costs and improve efficiency.31

7. Dynamic Problem Generation: As agents improve, previously challenging problems become trivial, requiring the generation of more complex challenges. This is akin to a company continuously innovating to stay ahead of competitors. Systems must dynamically adjust their focus to ensure they’re always tackling meaningful problems, much like a business pivoting its strategy to address emerging market trends.32

Learning Framework Lessons from the o1 Reasoning Model

The o1 reasoning model may offer some insights that can inform the design of learning frameworks for agentic systems, particularly in policy initialization and training efficiency:

1. Warm-Starting with Behavior Cloning:
   The o1 model suggests that behavior cloning is an efficient way to initialize policies, as it leverages high-reward solutions to quickly establish a strong baseline. This is akin to onboarding new employees with proven best practices to ensure they start on the right track. However, behavior cloning has limitations—it ignores lower-reward solutions, which can provide valuable learning signals. Thus, while it’s effective for warm-starting, it should be complemented with other methods for further optimization.33

2. Transitioning to Policy Optimization:
   Once the initial policy is established, the o1 model suggests transitioning to policy optimization methods like PPO or DPO. These methods utilize all state-action pairs, including lower-reward solutions, to refine the policy. This is similar to a company moving from standardized training programs to more dynamic, data-driven strategies that incorporate lessons from both successes and failures. PPO, while memory-intensive, offers robust performance, whereas DPO is simpler and more memory-efficient but relies on preference data.34

3. Handling Distribution Shifts:
   A potential challenge in o1 is the distribution shift that occurs when search-generated data (from a better policy) is used to train the current policy. This issue can also arise during policy initialization if the initial data doesn’t align with the system’s operational environment. To mitigate this, initialization can incorporate behavior cloning to narrow the gap between the initial and target policies, similar to aligning new hires’ skills with organizational needs through targeted training.35

4. Improving Training Efficiency:
   The o1 model highlights the possible computational cost of train-time search, which can slow down policy initialization. To address this, systems can reuse data from previous iterations or focus on high-value actions, much like a company optimizing its workflows to reduce costs and improve efficiency. This ensures that initialization is both effective and resource-efficient.36

In essence, just as a successful corporation thrives on continuous improvement, collaboration, and adaptability, multi-agent systems achieve their goals through reflective analysis, knowledge sharing, adaptive learning, and strategic exploration.

Challenges Ahead

As we move toward implementing sophisticated multi-agent systems, several significant challenges emerge that organizations must navigate carefully. These hurdles reflect the complexity of creating systems that can think, learn, and collaborate effectively at scale.

1. Lack of Published Research on o1

   A major challenge in reproducing or understanding the o1 model is the absence of published research from OpenAI. OpenAI has not released any detailed papers or technical reports about the o1 model, forcing researchers and developers to reverse engineer its capabilities based on limited information and public demonstrations. This lack of transparency makes it difficult to fully understand the underlying mechanisms, architectures, and training methodologies used in o1. As a result, the research community must rely on open-source projects and independent studies to piece together the roadmap to achieving similar capabilities. This underscores the importance of open-source models and the need for more transparency in AI research.

2. Technical Scalability and Performance: Just as businesses face challenges when scaling their operations, multi-agent systems encounter unique obstacles as they grow. Counter-intuitively, adding more computational power doesn't always improve performance—in fact, it can sometimes degrade it. This is similar to how adding more team members to a project doesn't necessarily increase productivity. The way these systems process information, particularly when making complex decisions, creates bottlenecks that become more pronounced at scale. Consider how a video conference becomes unstable with too many participants—multi-agent systems face similar coordination challenges but at a much more sophisticated level.

3. Balancing Learning and Action: Much like developing effective performance metrics for employees, creating the right incentive structures for AI agents presents unique challenges. These systems need clear signals about what constitutes success, but defining these metrics becomes increasingly complex as tasks become more sophisticated. For instance, how do you balance a customer service agent's need to resolve issues quickly with the quality of their interactions? The system must learn from experience while maintaining consistent performance, similar to how businesses must innovate while maintaining day-to-day operations.

4. Orchestrating Collaboration: Perhaps the most significant challenge lies in managing how multiple agents work together effectively. This mirrors the complexity of coordinating different departments within a large organization—each unit needs to operate independently while contributing to overall goals. The system must manage resources efficiently, ensure smooth communication between agents, and maintain performance across the entire network. Just as a company needs robust systems to coordinate between sales, operations, and customer service, multi-agent systems require sophisticated frameworks to manage their interactions and collective learning.

Ironically, as we work to develop AI systems to solve these organizational challenges, we could find ourselves grappling with the same fundamental problems that have plagued organizations for decennia—just in a different form. Whether managed by humans or artificial intelligence, it seems the core challenges of running complex, adaptive organizations will remain a difficult problem to solve.

Autonomous agents, like humans, rely on core elements such as identity, memory, planning, and action to function effectively. These components draw inspiration from human behavior, enabling agents to think, learn, and act in ways that feel natural and intuitive. At the same time, AI agents possess unique strengths, such as processing vast amounts of data, optimizing decisions in real time, and scaling operations far beyond human limitations. These insights are grounded in the latest research1 into autonomous agents, providing a clear framework for building systems that are both practical and groundbreaking.

The Four Pillars of Autonomy

At the core of every autonomous AI agent are four critical components: Profile, Memory, Planning, and Action. These systems function as an interconnected system, enabling agents to not only perform tasks but also adapt and improve continuously. Each component addresses a distinct aspect of intelligence, working together to create a cohesive and dynamic entity.

The Profile serves as the agent’s identity and personality, defining its roles, objectives, and the constraints within which it operates. This foundational element shapes how the agent interacts with the world, ensuring that its behavior aligns with its intended purpose and the expectations of users.

The Memory system acts as the agent’s experience bank, storing and retrieving information to inform future actions. By integrating short-term and long-term memory systems, agents can maintain context over ongoing tasks while building a repository of knowledge that enhances their capabilities over time.

The Planning system provides the agent with the ability to strategize and decompose complex objectives into actionable steps. Whether executing static plans or adapting dynamically to real-time feedback, this component ensures that the agent operates methodically and with purpose.

Finally, the Action component translates decisions into tangible outcomes. By leveraging internal reasoning capabilities and external tools, this component enables the agent to execute plans effectively while refining its approach based on feedback.

Together, these four pillars form the foundation of modern autonomous AI agents, empowering them to function as adaptive, intelligent entities in complex environments.

Profile: Defining the Agent’s Identity

The profile component establishes the foundation for an agent’s behavior and decision-making, defining its purpose, personality, and role. By tailoring these attributes, the profile ensures agents interact effectively with their environment and align with user expectations, making them more relatable, functional, and human-like.

Basic Attributes

Basic attributes outline the core characteristics of an agent, such as its expertise, role, or professional domain. For instance, a financial advisor agent may be assigned attributes emphasizing analytical rigor and familiarity with economic trends. These attributes act as the agent’s “resume,” ensuring its actions are informed by relevant context and knowledge.

Psychological Traits

Psychological traits shape how agents behave, mimicking personalities like empathetic, assertive, or collaborative. For example, a customer service agent designed to handle sensitive issues might prioritize empathy and patience, creating a supportive and positive user experience. These traits allow agents to engage users in a manner that feels intuitive and human-centered.

Social Context

Social context defines the agent’s relationships and collaborative dynamics. This can include specifying its role within a team or its interaction model with other agents and humans. For instance, in a collaborative software project, one agent might function as a “project manager,” delegating tasks to others, while another acts as a “developer,” implementing the assigned components.

Creating Profiles

• Manually Defined Profiles: Manually specifying profiles involves explicitly writing descriptions and constraints for the agent. For instance, “You are an analytical product manager focused on delivering data-driven insights to drive innovation.” While this approach ensures precision, it can be labor-intensive, particularly for large-scale implementations.

• Automated Profile Generation: Automated systems can generate diverse profiles by leveraging AI models. For example, by seeding profiles with a few initial traits, such as age or industry, the model can produce variations suited for different applications, saving time while ensuring scalability.

• Dataset-Informed Profiles: Real-world data can enhance profiles, making agents more relevant to specific scenarios. For instance, a sales agent’s profile could be informed by regional customer preferences, enabling it to tailor interactions effectively.

The profile component acts as the backbone of an agent’s identity, shaping its interactions and decision-making processes. By carefully defining attributes, psychological traits, and social roles, profiles ensure agents align with their intended purpose and resonate with user expectations. Whether crafted manually, generated through AI, or informed by real-world data, robust profiles empower agents to perform seamlessly and adapt dynamically across various applications. This foundational layer ensures that every interaction, memory, and action reflects the agent’s unique purpose and value.

Memory: The Agent’s Experience Bank

The memory component empowers agents to accumulate, recall, and reflect upon experiences, creating a feedback loop that enhances learning and adaptability. Inspired by human cognitive processes, the memory system integrates short-term and long-term components to navigate dynamic environments effectively.

Memory Structures

Much like human memory, an agent's memory is divided into short-term and long-term systems. Short-term memory acts as a transient workspace for managing immediate context. This is akin to remembering what someone said just moments ago during a conversation - it helps maintain coherence and enables meaningful responses. For example, an AI assistant might use short-term memory to hold recent customer queries, ensuring its answers remain relevant and conversational.

Long-term memory provides a more durable repository, consolidating key insights over time. Similar to how people remember important milestones or professional lessons learned, agents use long-term memory to store and retrieve knowledge that informs future decisions. A sales forecasting agent, for instance, could analyze years of sales data stored in long-term memory to detect seasonal trends and recommend optimal inventory levels.

Memory Formats

The way memories are stored also plays a vital role in agent performance.

• Natural Language Memory: Memories can be recorded as plain text, capturing the nuances of user interactions or observations. This format is highly interpretable and enables rich, context-aware decision-making. An agent managing customer support might store previous interactions as natural language logs, allowing it to ensure continuity in follow-ups.

• Embedding-Based Memory: By encoding memories as vector embeddings, agents can retrieve relevant information efficiently based on similarity searches. To illustrate, when recommending a product, an agent could match current user preferences with past customer behavior stored in embeddings.

• Structured Data: Agents can store information in structured formats like databases or hierarchical lists. This allows for systematic organization and precise querying, such as retrieving specific financial records during audits or customer details for targeted marketing.

Memory Operations

Memory operations govern how agents interact with stored information.

• Writing Memory: Storing new information is akin to taking detailed meeting notes. An agent must decide which details to preserve - filtering for relevance and avoiding duplication. For instance, an agent tracking project updates might consolidate repetitive task notes into a concise summary.

• Reading Memory: Retrieving stored knowledge requires prioritization based on relevance, recency, and importance. An agent planning a delivery route could pull recent traffic data (recency), prioritize major road closures (relevance), and consider weather conditions (importance) to optimize decision-making.

• Reflection: Reflection allows agents to summarize past experiences and derive insights. Similar to how a team evaluates project outcomes to refine future strategies, an agent could analyze failed and successful customer service interactions to improve its performance.

Hybrid Memory Systems

By integrating short-term and long-term memory, agents gain the ability to reason effectively in both immediate and strategic contexts. Imagine an inventory management agent that tracks real-time stock levels (short-term) while leveraging historical sales trends (long-term) to anticipate seasonal demand spikes. This dynamic capability enhances both day-to-day operations and long-term planning.

The memory component is like a knowledge management system in a business; short-term memory resembles an active project board for immediate tasks, while long-term memory acts as an archive of insights and strategies.

When designed effectively, it ensures continuity, allowing agents to draw from past experiences to inform present decisions. By integrating and balancing both short-term and long-term memory, agents can adapt to immediate demands while leveraging historical knowledge to make strategic, consistent, and impactful choices across dynamic and complex scenarios.

Planning: Charting the Path Forward

The planning component provides agents with the ability to deconstruct complex goals into actionable steps, enabling coherent and methodical execution. Much like humans, agents rely on structured planning to approach challenges in a manageable way.

Planning Without Feedback

In some scenarios, agents generate and execute plans without receiving intermediate feedback to adjust their strategies. These plans are created using predefined frameworks or processes, suitable for straightforward tasks.

• Single-Path Reasoning: This involves breaking down a task into sequential steps, with each step leading directly to the next. It mirrors a to-do list approach, where completing one step logically progresses to the next. For instance, an agent organizing an event might follow steps such as booking a venue, arranging catering, and finalizing invitations in a fixed sequence. While effective for predictable scenarios, this approach lacks adaptability when conditions change.

• Multi-Path Reasoning: This strategy involves exploring multiple possible solutions concurrently, akin to brainstorming different routes to solve a problem. By evaluating various pathways, the agent can select the most promising one. For example, a product design agent might explore different prototypes simultaneously and refine the most viable option based on user testing results.

• External Planning Tools: In complex or domain-specific scenarios, agents can integrate external tools to enhance planning. A healthcare scheduling agent could use specialized software to optimize appointments based on patient preferences, doctor availability, and equipment constraints. By leveraging such tools, agents can generate precise and efficient plans.

Planning without feedback is like a project manager creating a detailed plan for a product launch and sticking to it rigidly, regardless of market changes or team setbacks. While this approach ensures speed and consistency for predictable tasks, it falls short in dynamic scenarios where flexibility is crucial.

Planning With Feedback

For dynamic and unpredictable environments, agents employ iterative planning processes that incorporate feedback to refine their strategies.

• Environmental Feedback: This involves adjusting plans based on changes in the environment or observed outcomes. A logistics agent may revise a delivery route in real time to account for traffic congestion, ensuring timely deliveries despite evolving road conditions.

• Human Feedback: By soliciting guidance from users, agents can align their strategies with human preferences and expectations. A design assistant might request feedback on a prototype before finalizing its design, allowing iterative refinements based on client input.

• Model Feedback: Agents can self-assess their reasoning and actions using internal evaluative models. A financial analysis agent might critique its investment recommendations against historical data patterns, iterating on its conclusions to enhance accuracy.

Planning with feedback, on the other hand, is akin to a sales team adjusting its pitch in real time based on customer reactions during a meeting. While more resource-intensive and complex, it enables agents to respond effectively to unforeseen challenges, making it indispensable for tasks involving long-term reasoning or high variability.

Each method has its place: the former is ideal for stable, routine operations, while the latter is better suited to complex, evolving environments where adaptability is key.

Action: From Decision to Execution

The action component serves as the bridge between planning and real-world impact, transforming decisions into tangible outcomes. It is the final, yet pivotal, step where all preceding components converge to interact with the environment effectively.

Action Goals

Actions taken by an agent are goal-oriented and shaped by the overarching objectives of its tasks. These goals often fall into three primary categories:

• Task Completion: In many scenarios, the agent’s purpose is to complete specific, well-defined tasks. For example, an AI agent in e-commerce may automate order processing and inventory updates, ensuring operational efficiency. Similarly, in the context of software development, agents may execute unit tests or compile code based on predefined requirements.

• Communication: Effective communication is a crucial action goal, especially for agents working in collaborative or customer-facing roles. For instance, a customer support agent may communicate empathetically with users to resolve issues, while collaborative agents within a development team exchange structured information to coordinate efforts.

• Environment Exploration: Certain agents are designed to explore unfamiliar territories or environments to gather data or expand their capabilities. For example, a supply chain agent might analyze market trends by scanning competitor websites and extracting useful insights, refining its decision-making process iteratively.

Action Production

Translating decisions into executable actions can occur through various strategies:

• Memory-Based Actions: Actions often leverage stored knowledge. For instance, an agent retrieving prior successful strategies for handling specific customer complaints can adaptively address recurring issues. Memory-driven actions ensure continuity and relevance in execution.

• Plan-Based Actions: Some actions strictly adhere to pre-generated plans. A marketing automation agent, for example, may execute a pre-approved campaign schedule, systematically deploying ads, and analyzing engagement metrics without deviation unless prompted by dynamic inputs.

Action Space

The breadth of possible actions an agent can perform is dictated by its capabilities and integration with external tools:

• External Tools: Agents often utilize APIs, databases, or specialized models to extend their action space. For instance, a financial agent might access external economic data APIs to inform portfolio adjustments.

• Internal Knowledge: LLMs' inherent reasoning, conversational, and common-sense capabilities enable agents to make informed decisions. For example, an AI consultant might provide nuanced business recommendations purely based on its internal reasoning processes.

Action Impact

Actions inevitably have consequences that ripple across various dimensions:

• Environmental Changes: Agents can alter their surroundings, such as updating inventory levels in a warehouse management system or generating new content for social media campaigns.

• Internal Adaptations: Actions often result in updated internal states. An agent completing a sub-task may refine its long-term memory with insights, improving its future performance.

• Triggered Follow-Up Actions: In complex workflows, one action may cascade into others. For instance, sending an invoice could trigger payment tracking and follow-up reminders if deadlines aren’t met.

The action component is akin to a customer service representative implementing solutions. Without a robust system, actions might be reactive or misaligned with company goals. With an optimized action system, like an agent using CRM data, every interaction aligns with customer needs and business objectives, whether solving a support ticket, upselling a product, or adjusting based on feedback.

By effectively leveraging internal knowledge, external tools, and real-time inputs, the action component ensures agents not only execute tasks efficiently but continuously improve their performance, driving meaningful business outcomes.

Capability Acquisition: Building Agent Skills

While the architecture of an agent serves as its “hardware,” the true effectiveness of an autonomous AI agent lies in its ability to acquire the “software” it needs: task-specific skills, knowledge, and experiences. Capability acquisition is a critical process that enables agents to grow and adapt, transforming them from general-purpose tools into highly specialized entities capable of handling complex and diverse tasks.

Acquiring Capabilities Through Fine-Tuning

One of the most effective ways to enhance agent performance is by fine-tuning large language models (LLMs) using task-specific datasets. These datasets can be created through several methods, including human annotation, LLM-generated content, or real-world data collection. For example, fine-tuning with datasets such as those collected from e-commerce platforms or web interactions, provide valuable context to optimize agents for domain-specific challenges.

Fine-tuning allows agents to integrate substantial task-specific knowledge into their model parameters, making them highly effective at addressing particular use cases. However, this method is most suitable for open-source LLMs, as it requires direct access to the model’s architecture.

Enhancing Capabilities Without Fine-Tuning

When fine-tuning is not feasible, agents can acquire new capabilities through techniques such as prompt engineering and mechanism engineering.

• Prompt engineering leverages the natural language understanding of LLMs to describe desired behaviors or provide few-shot examples. For instance, including intermediate reasoning steps in a prompt can significantly improve an agent’s problem-solving capabilities. Similarly, prompts that incorporate social or reflective contexts can enhance conversational adaptability and self-awareness.

• Mechanism engineering, on the other hand, involves creating new operational strategies for agents. This can include iterative trial-and-error processes, where agents learn from feedback to refine their actions, or the development of self-driven evolution systems, where agents set goals and explore environments autonomously. These strategies enable agents to continuously improve their performance without altering their underlying model parameters.

Building a Learning Framework

Agents can also acquire new capabilities through frameworks that emphasize experience accumulation and collaborative problem-solving. By storing and refining successful actions in a memory system or a skill library, agents can draw on past experiences to solve similar tasks more efficiently in the future. Collaborative approaches, where agents exchange knowledge and adjust their roles dynamically, further enhance their ability to tackle complex problems that require collective intelligence.

Evaluation: Measuring Agent Performance

Evaluating the performance and effectiveness of LLM-based autonomous agents is a challenging yet crucial task. While these agents hold the potential to revolutionize workflows across industries, assessing their capabilities requires a nuanced approach that balances qualitative insights with quantitative rigor.

Subjective Evaluation

Subjective evaluation focuses on human judgment to assess the agent’s capabilities in tasks where no standardized datasets or metrics exist. This approach is particularly useful for evaluating aspects such as user-friendliness, creativity, or the agent’s ability to mimic human-like behavior.

• Human Annotation: In this method, human evaluators score or rank the agent’s outputs. For instance, annotators might rate agents on qualities like engagement, helpfulness, and honesty, as seen in studies where agents’ outputs are compared against human benchmarks. This approach captures nuanced feedback that reflects the real-world impact of an agent’s behavior.

• Turing Test: The Turing Test involves asking human evaluators to differentiate between outputs produced by agents and those created by humans. If the evaluators are unable to distinguish between the two, the agent demonstrates human-like performance. This strategy has been widely used to assess agents’ ability to generate human-like responses, emotional intelligence, and decision-making skills.

Objective Evaluation

Objective evaluation employs quantifiable metrics to measure agent performance, offering a more systematic and scalable approach. This method focuses on three key aspects: metrics, protocols, and benchmarks.

Metrics
Evaluation metrics are designed to capture specific dimensions of agent performance.

• Task Success: Metrics such as success rate, accuracy, and goal completion are used to assess how effectively the agent achieves its objectives.

• Human Similarity: Metrics like dialogue similarity, trajectory accuracy, and mimicry of human responses gauge how closely the agent’s behavior aligns with human norms.

• Efficiency: This includes measures like planning speed, cost of execution, and inference time, which evaluate the agent’s operational efficiency.

Protocols
Protocols define how metrics are applied in different contexts:

• Real-World Simulation: Agents are tested in immersive environments such as games or simulators, where task success and human similarity metrics can be observed in action.

• Social Evaluation: Agents are assessed based on their interactions in collaborative or competitive settings, analyzing qualities like teamwork, empathy, and communication.

• Multi-Task Evaluation: A diverse set of tasks from different domains is used to measure the agent’s generalization capability.

• Software Testing: Metrics like bug detection rate and test coverage are used to evaluate agents in coding and debugging scenarios.

Benchmarks
Benchmarks provide standardized environments and datasets for consistent evaluation. Examples include AgentBench2 for general-purpose assessments, WebShop3 for e-commerce capabilities, and EmotionBench4 for evaluating emotional intelligence. These benchmarks ensure that agents are tested against a wide range of real-world challenges, providing valuable insights into their adaptability and robustness.

Challenges: Key Barriers Ahead

While LLM-based autonomous agents have achieved significant milestones, the field is still in its early stages. There are numerous challenges that researchers and developers must overcome to unlock the full potential of these systems. Below, we outline some of the most critical challenges shaping the trajectory of development in this domain.

Role-Playing Capability

Autonomous agents are often required to assume specific roles, such as a researcher, programmer, or teacher, to complete tasks effectively. While LLMs can simulate some roles convincingly, they struggle with less common or emerging roles, as well as aspects of human cognition like self-awareness in conversations. This limitation stems from the datasets on which these models are trained, which may lack diverse role-specific data. Fine-tuning with curated datasets or designing optimized prompts and architectures may improve role-playing, but balancing these enhancements with general-purpose performance remains a significant challenge.

Generalized Human Alignment

For agents to serve humans effectively, they must align with human values. However, in applications like real-world simulations, agents may need to replicate both positive and negative human behaviors to provide accurate models of societal dynamics. This dual alignment is particularly challenging as most LLMs are optimized for positive, unified human values. The development of prompting strategies or controlled re-alignment techniques is needed to tailor agents to diverse scenarios without compromising ethical standards.

Prompt Robustness

The integration of complex modules like memory and planning requires structured, reliable prompts to ensure consistent agent behavior. However, even slight modifications to prompts can cause significant deviations in agent outputs. This challenge is exacerbated by the interconnected nature of agent modules, where changes in one prompt may affect others. Developing unified and resilient prompt frameworks that are robust across diverse tasks and LLMs remains an open problem.

Hallucination

Hallucination, where agents confidently produce incorrect information, poses a major challenge in high-stakes applications. For example, in coding tasks, hallucinations can generate erroneous outputs with potentially severe consequences. Addressing hallucinations requires iterative feedback mechanisms, improved training datasets, and fail-safe systems to validate outputs before execution.

Knowledge Boundary

LLMs possess extensive knowledge from training on vast datasets, which can sometimes hinder their ability to simulate realistic human behaviors. For instance, when tasked with replicating user behaviors with limited prior knowledge, agents may leverage their extensive training corpus inappropriately. Constraining agents to appropriate knowledge levels requires innovative methods for controlling context and access to information.

Efficiency

Agents often need to query LLMs multiple times for tasks such as memory retrieval, planning, and decision-making. The inherent slowness of LLMs due to their autoregressive architecture creates significant efficiency bottlenecks. Addressing this requires optimizing agent workflows, leveraging caching or batching strategies, and developing faster inference techniques.

In upcoming issues, we’ll dive even deeper into how to design autonomous agents. Subscribe to stay ahead of the curve and learn how to shape the future of work.

Subscribe now

More than a decade ago, Marc Andreessen famously wrote that software is eating the world. Back then, the idea that digital platforms would disrupt entire industries seemed bold. Yet time proved him right: software solutions did spread through every sector, transforming the way we consume media, shop online, and conduct business. Now, AI is not just eating the world; it’s eating the jobs along with it.

As we speak, countless AI startups are building vertical solutions that go beyond assisting human workers. They’re aiming to replace them entirely. No job is safe in this wave of automation, from the simplest data entry position to the most senior data analyst. We’re talking about entire teams being replaced by systems that learn continuously and never clock out.

Big Tech CEOs often downplay these concerns. They’ll tell you that people will find meaning in creative and artistic work, that new types of jobs will emerge requiring everyone to be retrained, or that society will adopt a wealth dividend so nobody has to work anymore. But none of these ideas hold up under scrutiny. AI is already replacing creative and artistic work, producing everything from paintings and music to screenplays and advertising campaigns. There’s no clear evidence that the pace of new job creation will match the speed at which AI displaces old ones. And the wealth dividend? It’s more a hypothetical discussion than a concrete plan, with no clear path for implementation. The reality is this: the only way to thrive in this new era is to embrace it and start a business yourself, leveraging AI as the powerful tool it is becoming.

Just as the SaaS boom of the 2000s created a new generation of tech giants by lowering barriers for software distribution, AI will be able to handle the same work (or better) without the need for large teams of people. For every iconic SaaS success story, there could soon be an $300B+ AI-driven agent company. The difference? These businesses don’t just save on software costs; they thrive with small, efficient teams while achieving massive scale.

Much like the early days of SaaS—when the idea of subscribing to software over the internet felt fresh—AI-based services are still in their infancy. But every three months, these models become smarter, faster, and more reliable, unlocking new verticals ripe for takeover.

We’re at a point where, if you look around for any boring, repetitive administrative jobs, you’ll likely find a billion-dollar AI opportunity. These tedious chores are exactly what AI excels at, and a determined founder can leverage these advanced tools to build a lean, high-margin business that runs on autopilot. We’ve already seen early wins. There are AI services handling full-scale testing automation, voice-based outreach for auto loan collections, and specialized chatbots that replace entire developer relations teams. Instead of humans spending their days on repetitive work, AI takes over, freeing individuals to do something else—or leaving them out of the loop entirely.

Of course, the go-to-market strategy for these AI startups looks very different from what we saw during the software boom. If you try to sell these AI solutions to the very team they’re going to replace, you’ll face fierce resistance. Those employees, knowing the automation threatens their roles, have no incentive to adopt the technology that makes them redundant. Instead, successful AI startups focus on high-level decision makers—executives who see the big picture and care about bottom-line efficiency. From their vantage point, fewer employees and more output is a no-brainer. And as AI agents deliver quantifiable results—no salaries, no HR overhead, no burnout—these decision makers will choose the technology that streamlines their operations.

This new dynamic means that old constraints on company size and management complexity will break down. With vertical AI agents at the helm, companies no longer need armies of staff to achieve massive scale. A small founding team, equipped with the right AI systems, can run an enterprise that serves millions of customers. A business that once demanded offices, HR teams, and multiple management layers can now exist almost entirely in the cloud, orchestrated by a handful of AI models that coordinate production, distribution, marketing, and sales. It’s the birth of the billion-dollar, one-person company—a vision that sounds wild today but could become routine tomorrow.

In the long run, AI will likely remain vertical and specialized rather than consolidating into a few dominant platforms. Just as SaaS products proliferated into highly specific niches, AI agents will become ever more tailored, filling tightly defined roles. From legal document review to dental insurance claim processing, each narrow market offers a chance for AI to replace human effort with near-perfect, always-on service. Vertical agentic systems will pave the way for entirely new categories of companies.

Personal Note:

👋 I’m Melvin Tercan. I’m the Head of Engineering at Octagon AI, an vertical agentic AI startup with a mission to build agentic systems that empower financial analysts to do more impactful work. Over the past three years, I’ve gained deep insights into the design of agentic systems and how they’re reshaping industries.

If you’re interested in learning how to design and build agentic systems, subscribe for free to this newsletter. Let’s build the future together!