🧠 Community Wisdom: PRDs for agents, tracking competitor messaging, whether software moats still exist, exercising startup options when you leave, and more

We have spent the last decade perfecting the art of building software for humans. We mastered the user interface, the onboarding flow, and the dopamine loops that keep people clicking. But the landscape has shifted underneath us. Now, we are building for agents. This shift is not just a technical upgrade; it is a fundamental rewrite of how products are conceived and protected.

The Death of the Traditional PRD

For years, the Product Requirements Document (PRD) was the holy grail of the dev cycle. You defined the user, the pain point, and the functional requirements. You sketched out wireframes and mapped out buttons. But when your primary user is an LLM-powered agent, a button is just noise. In a world where an agent is interacting with your API or your frontend to complete a task, the PRD needs to evolve or it becomes a liability.

Builders need to stop obsessing over where the pixel goes and start obsessing over intent and outcome. An agentic PRD requires a focus on machine-readability. If an agent cannot parse what your software is supposed to do in milliseconds, it will move on to a competitor that has better documentation. We are moving from a visual-first design philosophy to a logic-first architecture.

Building for the Non-Human User

When you build for an agent, you have to account for unpredictability. Humans are relatively predictable in their errors; they miss clicks or forget passwords. Agents, however, can hallucinate, loop indefinitely, or try to exploit an edge case you didn't know existed. This means your requirements must include rigid guardrails and clearly defined success states that a machine can verify.

• Define the Objective: What is the absolute binary success state?
• Structured Inputs: How does the agent feed data into the system without breaking it?
• Error Handling: How does the system communicate a failure back to a non-human entity?

If you are still writing PRDs that focus on how a human feels when they see a dashboard, you are building for a shrinking market. The future is built on how efficiently an autonomous script can execute a workflow within your environment.

The Vanishing Software Moat

There is a growing panic in the founder community about moats. For a long time, your moat was your code, your data, or your network effect. AI has effectively commoditized code. If a junior dev can prompt a sophisticated feature into existence over a weekend, your proprietary codebase isn't the fortress you think it is.

Some argue that software moats no longer exist. I think that is an oversimplification, but the traditional walls are definitely crumbling. High-velocity shipping used to be a competitive advantage. Now, it is just the baseline for survival. If your only defense is that you built a better mouse trap, you are in trouble, because AI can build a thousand mouse traps by dinner time.

Where the Real Defensibility Lives

If code is no longer a moat, what is? It comes down to two things: distribution and proprietary context. Distribution is the hard part that AI hasn't solved yet. Having the trust of a customer base and a seat at the table is still the most valuable asset a founder has. You can have the best AI agent in the world, but if no one knows you exist, you have nothing.

Context is the other half of the equation. This isn't just raw data; it is the specific, messy, tribal knowledge of a particular industry. An AI can write code for a legal tech platform, but it doesn't understand the nuances of a specific courthouse's filing system unless you provide that context. Founders who win will be the ones who embed their deep domain expertise so deeply into the product that a generic LLM cannot replicate the result.

Competitive Intelligence in the Age of Noise

Tracking what your competitors are doing has also changed. It used to be enough to sign up for their newsletter and watch their pricing page. Now, you have to monitor their integration ecosystem. You need to know which models they are plugging into and how they are structuring their data for agentic access. If you aren't watching how your rivals are positioning themselves for the machine-readable web, you are flying blind.

The goal is not to track every feature release, but to understand the underlying logic of their ecosystem. Features are temporary; architecture is permanent.

The Equity Trap

Outside of the product itself, founders and early employees are facing a different kind of reality check: startup options. As the market tightens and valuations get a haircut, the old advice of "just exercise and hold" is becoming dangerous. Builders need to be more clinical about their equity. If you are leaving a startup, you have to treat that exercise decision like a cold-blooded investment, not a sentimental parting gift.

With the current volatility in the AI sector, many options are underwater or will be soon. Before you sink your life savings into exercising shares in a company that might be disrupted by a new GPT release next month, look at their moat. Do they have proprietary context? Do they have distribution? If the answer is no, those options are just lottery tickets with bad odds.

The Founder Perspective

The takeaway here for anyone building in the trenches is simple: the rules have changed, but the fundamentals of business haven't. We are seeing a massive shift in how we document work and how we protect our intellectual property. The PRD is becoming a technical manual for machines, and the moat is moving away from the screen and back into the real world of relationships and specific, hard-to-get data.

Don't get distracted by the hype of "AI doing everything." Focus on the things AI can't do—navigating complex human hierarchies, building brand trust, and understanding the subtext of a customer's real problem. Build your software to be agent-friendly, but build your business to be human-centric. That is the only way to stay relevant while the machines take over the busy work.

Read the original at Lenny's Newsletter →