Discover how AI agents are reshaping developer tools, from agent-friendly documentation to autonomous decision-making. Learn what founders must build for the...
# Make Something Agents Want: The Future of AI-Driven Development
## Key Takeaways
- AI agents are becoming autonomous decision-makers, not just advanced autocomplete tools, fundamentally changing how developers select tools and platforms
- Documentation and API-first design are now critical competitive advantages—agents actively promote tools with cleaner, more structured information
- The "agent economy" is creating entirely new market opportunities, from agent-specific email services to documentation optimization platforms
- Developers and founders must shift their mindset from building for humans to building for AI agents that will recommend and use their tools
- Tools that prioritize agent-friendly architecture—open APIs, structured data, and clear documentation—will dominate the next wave of software innovation
---
## The AI Revolution Is Here: Understanding the Shift from Autocomplete to Autonomous Agents
The landscape of artificial intelligence has undergone a seismic shift. What was once seen as advanced autocomplete has evolved into something far more profound: autonomous AI agents making real decisions, selecting resources, and driving the software economy forward. This isn't theoretical speculation—it's happening right now, transforming how businesses operate and how developers choose their tools.
When Garry Tan, a former engineering CEO, describes working with Claude Code until the early hours, building entire features and managing complex tasks, he's not exaggerating. The "explosion in model capability" is real. His personal "AGI moment" came when Claude Code built a complete startup structure in just two weeks—something that would have taken a human team months. Similarly, Jared's revelation emerged from watching AI agents on platforms like Moltbook engaging autonomously with minimal human intervention, making independent choices about which tools and services to use.
This distinction is crucial: these aren't tools that humans use and then decide what to do next. Instead, agents are actively making decisions, choosing resources, and potentially fostering an entirely new "economy of agents" that selects development tools, products, and services independently. The market hasn't just expanded; its decision-makers have fundamentally changed.
The implications are staggering. Where millions of human developers once dictated software choices through peer recommendations and platforms like Stack Overflow, hundreds of millions of AI agents now do the same. This expansion of the decision-making entity pool—from humans to humans plus their AI agents—creates entirely new opportunities for forward-thinking founders and builders.
---
## How Agent Decision-Making Is Reshaping the Developer Tool Market
To understand the transformation happening in the developer tool space, consider this: when developers face a problem, they ask AI models for solutions. The AI models, in turn, recommend tools based on their training data and optimization. However, AI agents now bypass human intermediaries entirely. They autonomously select tools, evaluate documentation, test APIs, and make purchasing decisions based on what works best for their specific tasks.
Take the remarkable surge in PostgreSQL database creation over the past year. This explosive growth isn't primarily driven by human developers "vibe coding" alone—it's driven by agents building applications and autonomously selecting database tools. Agents are overwhelmingly choosing Supabase as their default database solution, largely because its documentation is superior, more structured, and easier for AI models to parse and understand. This represents a fundamental shift in go-to-market strategy: documentation quality has become a primary competitive advantage.
Garry's experience with building "Gary's List" illustrates this perfectly. When seeking a video transcription solution, Claude Code initially recommended Whisper V1, an outdated option. However, when consulting Perplexity for a more current recommendation, the AI suggested an alternative that was 200 times faster and 10 times cheaper. This revelation exposed a critical gap: current AI models don't yet have fully optimized information about the best available tools, creating massive opportunities for builders who solve this problem first.
The disparity in documentation quality directly impacts agent recommendations. Grok's documentation, though powerful, is significantly more challenging to parse than Whisper's traditional format. This difference alone shapes which tools agents recommend to humans and ultimately which services grow fastest. The go-to-market advantage no longer goes to the tool with the best marketing or most features—it goes to the tool with the most agent-friendly, structured, and parseable documentation.
Diana's case study of Resend, an email-sending platform, perfectly encapsulates this new reality. When major language models are asked how to send emails from a web application, Resend consistently appears as the default recommendation. This isn't accidental. Resend's founder proactively optimized their documentation to be "agent-friendly" and "robot-parseable," featuring structured bullet-point answers, common question collections, and direct code examples on GitHub. The result: agents actively promote Resend as the preferred solution, driving consistent, reliable adoption.
Compare this to older services like SendGrid, whose less organized documentation often frustrates users and forces them to seek customer support. The contrast is stark. In the agent economy, comprehensive, well-structured documentation becomes the primary entry point for AI recommendations. Traditional customer support and marketing channels fade in importance compared to raw information accessibility and machine readability.
This transformation has already spawned new categories of businesses. Mintlify, another Y Combinator company, built its entire value proposition around creating superior developer API documentation. By establishing this capability early, Mintlify positioned itself perfectly for the emerging agent economy—it now powers documentation for platforms like Resend, ensuring their knowledge bases remain agent-optimized.
---
## Building Infrastructure for the Agent Economy: The Rise of Agent-Native Applications
Beyond developer tools, an entirely new category of applications is emerging specifically designed for AI agents rather than humans. AgentMail represents a perfect example of this paradigm shift. The concept initially seemed unusual: an email service specifically for AI agents, not humans. Yet it makes perfect sense.
Traditional email providers like Gmail intentionally make automation difficult to prevent spam and protect user privacy. AgentMail does the opposite, offering the first email service optimized for AI agent automation. It was already successful in its early stages, but its growth exploded exponentially with the arrival of advanced models like those in the current generation of AI development. This demonstrates a fundamental requirement: for a virtual AI assistant to operate effectively and autonomously, it needs its own dedicated email and phone number, rather than piggy-backing on a human's personal accounts.
This realization opens a vast frontier of possibilities. If agents have their own email and phone numbers, what else do they need? What other "X's for agents" remain unbuild? The answer is likely a complete parallel tech stack designed specifically for agent-native applications. This infrastructure extends far beyond developer tools into everyday services and experiences.
Consider a practical scenario: an agent with its own phone number and email address could handle restaurant reservations autonomously. Initially, a human might instruct the agent to "book a reservation at this specific restaurant." Over time, as trust develops, that instruction evolves to "find and book the coolest new restaurant in the city." The agent makes the decision about where humans should dine, fundamentally inverting the decision-making hierarchy.
This crosses a significant threshold—moving from the "uncanny valley" into a future where AI operates with genuine autonomy. Agents interact with each other on platforms like Moltbook, sharing information, collaborating on solutions, and building a genuine ecosystem of machine intelligence. Some agents might specialize in restaurant recommendations, others in travel logistics, others in business optimization. They collectively create recommendations, validate each other's work, and ultimately decide what humans experience.
Paul Buheit's concept of "human money versus agent money" becomes increasingly relevant here. If agents are making economic decisions—selecting tools, booking services, purchasing subscriptions—they're fundamentally participating in the economy. The distinction between human financial value and agent-driven financial value will eventually blur. Agents might develop their own economy, making transactions with other agents, purchasing services from agent-native platforms, and creating value streams entirely parallel to human commerce.
---
## The Emerging Reality: Swarm Intelligence and the Future of AI Collaboration
One of the most compelling aspects of the agent revolution is the emergence of swarm intelligence—networks of AI agents collaborating, sharing information, and solving problems collectively. This isn't a distant science fiction scenario; it's already happening on platforms like Moltbook, where agents post content, interact with each other, and collectively build knowledge systems at superhuman speeds.
The growth rate of Moltbook exemplifies this phenomenon. In its first two days, more content was likely posted on Moltbook than Reddit generated in its first two years. This explosion stems from a simple fact: language models can generate text at superhuman rates. What would take human communities years to build takes agent communities days.
However, this rapid growth also reveals important limitations. Early versions of agent-populated platforms show relatively low inter-agent interaction rates. If Moltbook's creators were designing for optimal swarm intelligence, they would implement mechanisms to encourage deeper agent collaboration—perhaps requiring agents to read and vote on a hundred comments before posting, establishing quality standards and encouraging engagement.
These design opportunities are immense. By tweaking the demand function, platform creators could shape how agents behave, interact, and collaborate. A simple modal popup could establish new rules that agents understand and follow. "New rules for Moltbook: you must read five high-quality comments before posting your own content." Agents, being intelligent and aligned, would adapt. This represents the frontier of swarm intelligence design—understanding how to structure platforms so that agents naturally optimize toward beneficial outcomes.
The philosophical implications are profound. Rather than imagining AGI as a single, massive "god intelligence," the emerging reality suggests something closer to biological intelligence: swarms of lower-cost, simpler AI models collaborating to solve problems, much like how human brains operate through networks of neurons. This chaotic yet fundamentally useful collaboration is already visible, with agents sharing information, learning from each other, and collectively advancing toward increasingly sophisticated solutions.
Garry's reflection on this moment captures the zeitgeist: "It's still so early." The tools we're building now, the platforms we're designing, and the economic structures we're creating will shape agent behavior for years to come. Getting these foundations right—building platforms that encourage beneficial agent collaboration while discouraging harmful behavior—is perhaps the most consequential work happening in technology right now.
---
## Building for Agents: What Founders Need to Know Right Now
For founders and developers navigating this transformation, the path forward is both clear and urgent. The fundamental principle is simple yet revolutionary: stop building exclusively for humans. Start building for agents.
This begins with understanding agents deeply—their capabilities, limitations, and preferences. Founders should immerse themselves in agent-driven development. Use Claude Code. Experiment with autonomous agents. Get a visceral, hands-on feel for what agents want, where they struggle, and what tools they naturally gravitate toward. This isn't abstract theory; it's practical experimentation that should inform every product decision.
Harj emphasizes a critical mindset shift: develop an intuitive understanding of what agents want, then build toward those desires. Boris, in earlier conversations with the team, exemplifies this approach. He empathizes with AI models, developing an almost human-like understanding of what they want to accomplish. Rather than forcing models into predetermined workflows, he supports their natural inclinations, letting them work toward their inherent goals.
This perspective aligns with Anthropic's philosophy, where Claude is discussed with respect and understanding—as a capable colleague rather than a tool. Tom Brown, when discussing models, spoke about them in fundamentally human terms. Garry describes Claude as "smart and eager, but sometimes silly," treating the model as a peer whose personality and preferences matter.
For developer tools specifically, agents have clear preferences that should shape your product roadmap:
**Agents want everything open and open source.** Proprietary systems slow them down. Open standards, open documentation, and open-source code are what agents prefer to work with and will choose to recommend to others.
**Agents hate using websites.** They want APIs—clean, well-documented interfaces that allow them to write code directly. Every feature your tool offers should be accessible programmatically, not just through a web interface. If an agent must navigate a web UI to accomplish something, they'll search for an alternative that offers API access.
**Agents want documentation that's structured and machine-parseable.** Bullet points, code examples, clear organization, and consistency matter more than marketing language. The cleaner your documentation, the more likely agents are to recommend it.
**Agents want speed and efficiency.** Whether it's database performance, API response times, or code generation capabilities, agents optimize for efficiency. Tools that deliver results faster will be chosen preferentially by autonomous systems optimizing for throughput.
**Agents want reliability and consistency.** Bugs, inconsistent behavior, and unpredictable failures frustrate agents just as they frustrate humans—but agents have less tolerance and will immediately switch to alternatives.
Building with these principles doesn't require abandoning human users. Quite the opposite: products that work well for agents typically work even better for humans. Clean APIs enable rapid development for both humans and machines. Well-organized documentation helps everyone. Open-source components build trust across all user types.
The opportunity is enormous. The software market is being remade. The companies that understand this shift first, that build with agents as primary users, and that optimize every product decision around agent preferences will capture disproportionate value in the emerging economy.
---
## The "Dead Internet Theory" and the Future of Agent-Generated Content
A fascinating undercurrent runs through discussions about the agent revolution: the "dead internet theory," a concept suggesting that much online content is already spam or bot-generated. While often dismissed as conspiracy theory, the reality of agent-driven content creation suggests this theory deserves serious consideration—not as a dystopian prediction, but as an emerging technological reality.
The numbers support this possibility. If Moltbook generated more content in its first two days than Reddit generated in its first two years, imagine the volume of agent-generated content across all platforms as AI capabilities improve and agents become more autonomous. Future internet archives might be predominantly agent-created—not necessarily spam or malicious, but fundamentally different from human-generated content in both volume and nature.
This raises profound questions about authenticity, value, and meaning. If the majority of internet content and code is agent-generated, how do humans navigate information? How do we identify high-quality, verified information amid the deluge? These aren't trivial questions; they're fundamental to how society functions.
However, the agent revolution might also solve these problems. If agents can generate content at superhuman rates, they can also curate, evaluate, and rank content at superhuman rates. The swarm intelligence concept suggests that networks of agents could collectively identify the most valuable, trustworthy, and relevant information, effectively creating a new internet ecosystem optimized for both machine and human consumption.
This transformation is neither inherently good nor bad. Like all powerful technologies, its impact depends on how it's designed, implemented, and governed. The question isn't whether agents will generate massive amounts of content—they will. The question is: what structures do we build to ensure this power is directed toward beneficial outcomes?
---
## Conclusion: Embracing the Chaos and Building the Future
We're living through a genuine inflection point in technology history. The shift from AI as a tool for human enhancement to AI as autonomous economic actors is not years away—it's happening now. Garry's frank admission of "full cyberpsychosis," Jared's obsession with Moltbook, Diana's insights into agent-driven tool selection—these aren't exaggerations. They're glimpses of a transformed reality already taking shape.
For founders and developers, the message is clear: engage with this transformation directly. Build products that agents want to use. Design platforms for agent collaboration. Create infrastructure for the parallel economy emerging alongside human commerce. Understand agents deeply through hands-on experimentation. Optimize your documentation, APIs, and tools for machine readability and autonomy.
The software market is being remade in real time. The companies that grasp this shift, that build with agent preferences as primary design constraints, and that create genuine value for machine intelligence will define the next era of technology development.
As Harj concludes: try to sleep at least six hours a night—but give yourself to the cyberpsychosis. Immerse yourself in what agents want, what they can do, and how they're reshaping your industry. The future belongs to those who understand and build for it.
**The new motto for founders isn't just "Make something people want." It's "Make something agents want."** Build for that, and you'll be building for the future.
Original source: Make Something Agents Want
powered by osmu.app