Every AI Agent Needs a Manifesto. Here's Why.

Ron Reynolds · 2026-02-13 · 4 min read

When your development team includes AI agents, culture isn't optional. It's configuration.

Most teams using AI for development start with a prompt. "Build me a REST API." "Write tests for this service." "Refactor this module." And the AI obliges — competently, quickly, and with zero understanding of what actually matters to your business.

That's the problem. Competent code without context is just sophisticated guessing.

When I started building with AI agents — not as side tools, but as core members of the development process — the first thing I wrote wasn't a spec. It wasn't a ticket. It was a manifesto.

A manifesto answers the questions AI will never think to ask.

Not "what should I build?" but:

What does quality mean here — specifically, measurably? What are we never willing to compromise on? When two good options exist, which do we choose and why? What does "done" actually look like?

Without these answers baked in, every AI interaction starts from zero. You get code that works but doesn't belong. Features that function but don't fit. Output that's technically correct and strategically wrong.

Human teams absorb culture through osmosis. AI agents don't.

A junior developer sits in standups, overhears design debates, watches a senior reject a PR for the third time because "it's not how we do things here." Over months, they internalize the unwritten rules.

AI agents don't get months. They don't overhear anything. Every session is a blank slate unless you explicitly tell them what matters.

A manifesto is how you transfer institutional knowledge to a system that has none. It encodes the decisions your team would make at 2 AM when nobody's watching — because that's exactly when AI agents are running.

What belongs in an agent manifesto?

After iterating on this extensively, I've found five categories that matter:

1. Non-negotiable standards. Not aspirations — hard lines. What testing coverage is mandatory? What error handling is required? What code patterns are banned? AI agents are remarkably compliant when given clear boundaries. They're remarkably creative at finding loopholes when you don't.

2. Decision-making hierarchy. When quality and speed conflict, which wins? When simplicity and completeness conflict? When convention and optimization conflict? Every piece of code involves tradeoffs. Your manifesto should pre-resolve the common ones.

3. The "why" behind the "what." AI agents that understand why you validate all inputs with Zod will apply that principle to novel situations. Agents that only know what to do will follow the letter and miss the spirit. Include the reasoning, not just the rules.

4. Definition of done. Not "it compiles." Not "tests pass." The actual bar. Does it handle edge cases? Is it documented? Would it survive a code review from someone who cares about craft? If your manifesto says "good enough is not good enough," the agent knows to keep going when a lesser standard would stop.

5. The mission. This sounds soft. It's not. When an agent understands it's building an operating system — not a feature, not a tool — it makes different architectural choices. Context shapes output. The bigger the context, the better the output.

The manifesto is the most leveraged document in your codebase.

Every other document — specs, tickets, READMEs — describes what to build next. The manifesto describes how to build everything, forever. It's the difference between managing every task and setting a standard that manages itself.

I've watched the same AI agent produce mediocre code with a generic prompt and exceptional code with a manifesto loaded. Same model. Same task. The only variable was context about what "exceptional" means here.

The counterargument is obvious: "Just write better prompts."

Sure. And just write better requirements. And just hire better people. And just have better meetings.

The manifesto isn't a replacement for good prompts. It's the foundation that makes every prompt better. It's the accumulated wisdom of your organization, compressed into a document that ensures no agent — and no new team member — starts from scratch.

If you're building with AI agents and you don't have a manifesto, you have a team with no culture.

And a team with no culture builds whatever's easiest, not whatever's right.

Start with the non-negotiables. Add the "why." Define done. State the mission.

Then watch every agent in your system build like they've been on the team for years.

Because in a sense, they have.