Manifesto

An essay on what we're building and why.

I. The terminal becomes the place where ideas are born

Most developers already spend half their working day with an AI agent in the loop — Claude Code, Cursor, Cline, Codex, ChatGPT, the next one. By 2027 it will be most of the day. Not as a chat sidebar to consult occasionally. As the place where the work happens.

When the work happens there, the ideas happen there. A refactor that wants to be its own tool. A database schema that wants to be a market. A CLI flag that exposes a primitive worth charging for. A throwaway script that turns out to be the seed of a product. The activation energy of inspiration drops to near zero — you can articulate the idea to the agent in fifteen seconds.

But the activation energy of acting on it is still measured in days. You have to leave the agent window. You have to register a domain. You have to set up a landing page. You have to find someone to fund it. You have to convince yourself it's worth the weeks of work between idea and validation. The friction is so high that most ideas die in the same window they were born in, as a comment that begins // TODO: and ends six months later, untouched.

This is the wave. The next decade of work will produce orders of magnitude more ideas than the previous one. Almost none of them will become anything, because the path from agent-surfaced thought to validated public artifact is still routed through the old machinery of pitch decks, demo days, and warm intros.

P&L exists for the path that machinery never built.

II. Why "less friction" is the wrong answer

The default crypto answer to friction has been to remove it entirely. Memecoin launchpads — pump.fun and its imitators — bet that the bottleneck was permission. If anyone could mint a token in sixty seconds for less than a dollar, more good things would happen. Talented people would launch, communities would form, and the noise would self-correct via price.

The market called the bluff. 98.6% of pump.fun tokens rug. Most of the rest go to zero through inattention rather than malice. Real builders stopped using these surfaces because the signal-to-noise made them unusable as either a discovery layer or a fundraising layer. The launchpad genre got captured by extraction primitives — bonding curves designed to harvest attention into liquidity into volatility into PnL for someone other than the founder.

The lesson isn't that low-friction launchpads are bad. They're not — they removed real gatekeeping, and they let a generation of people experiment with capital formation in ways VC never permitted. The lesson is that a launchpad is only as good as its filter. Without one, every legitimate idea is buried under ten thousand copies of the same scam. With one, the same launchpad becomes an instrument.

Standard prediction markets get the filter half right. Polymarket, Kalshi, Augur — they aggregate beliefs about external events: will Trump win, will the Fed cut, will GTA VI ship. They've proven that crowds with skin in the game predict better than experts or polls. The $3.3B in 2024 election wagers wasn't noise; it was an information machine that called the result before the media did.

But prediction markets aren't productive. They resolve and disappear. They don't fund anything, build anything, or graduate any idea into the world. The conviction generated stays as a number in a database. The market becomes right and then becomes irrelevant.

P&L takes the prediction-market insight — crowds with skin in the game have better signal than gatekeepers — and points it at the question that actually matters for the AI-builder wave: does this idea deserve to launch?

III. Conviction as capital

The mechanic is simple. The implications take a while to land.

An idea is posted. Two camps form. Believers stake SOL on YES — they think it should exist. Critics stake SOL on NO — they think it won't work, or that the world is better off without it. Both sides have to put real capital at risk. The pool grows. The implied probability moves. Eventually the market resolves.

When YES wins, the pool funds the launch. A token is created on pump.fun with seeded liquidity. The founder earns ongoing royalties from trading. The believers receive token airdrops proportional to their conviction. The idea passes through a real gate and arrives in the world as a credible artifact — vetted by people who paid to vouch for it.

When NO wins, the critics split the pool. They get paid for filtering. This is the part that almost no other system has: an explicit, rewarded role for skepticism. Memecoin launchpads pay no one to call out scams. Prediction markets don't pay anyone to maintain the commons. Open-source pays nothing for code review. P&L makes skepticism a job with a paycheck, denominated in the same currency as belief.

What this produces, over enough markets, is a curated stream. Not a feed of ideas — an index of which ideas had enough conviction behind them to graduate. Every idea that bloomed went through real economic resistance. Every idea that withered was opposed by people who staked money on it being wrong. The signal is dense.

That's the layer the AI-builder wave needs. Not lower friction. Different friction — friction that consists of other humans betting real money that your idea is worth their time.

IV. The terminal as the planting tool

Here is the workflow we're building toward.

You're in Claude Code or Cursor, mid-task. Something surfaces — a small thing, the kind of thing you would have left as a TODO in 2024. The agent reads the context and asks: "This isn't blocking what we're working on, but it could be its own thing. Want to plant it on P&L?"

You say yes. The agent extracts the description from your conversation, generates a thesis paragraph, picks a sensible target pool, names the would-be token. You confirm in your wallet. The market is open. The link to it gets dropped back into the chat. You go back to what you were doing.

Sometime later, the market resolves. Maybe believers won and your idea is now a token, with a treasury and a community and a path forward. Maybe critics won and a small stake was the entire cost of finding out it wasn't worth pursuing. Either way, the question got answered with real money and real attention, while you were doing other things.

We are aiming P&L to be the protocol that runs in the background of the AI-builder workflow. Read endpoints that any agent can call. An MCP server in ~/.claude/skills. A pnl-pitch slash command that turns a paragraph into a market. Reasonable defaults so the founder doesn't have to fill a form. Skepticism that does the filtering, automatically, on the same surface where the idea was born.

The site you're reading is the developer-facing documentation for that protocol. It will grow with the product. The thesis here is a stake in the ground.

V. Why this matters beyond crypto

We have to be honest about what we're claiming.

We are not claiming P&L will solve attention markets, or AI alignment, or capital allocation, or any of the other problems people sell crypto into. We are claiming something narrower and harder to refute.

Right now, in 2026, there is no good path from agent-surfaced idea to publicly-validated artifact. Pitch decks are too slow. Twitter is too noisy. Memecoin launchpads are too captured. Standard prediction markets don't fund anything. AngelList and Y Combinator are excellent for some founders and accessible to almost none of the people whose best ideas show up at 2am while debugging.

The closing of that gap matters because the volume of ideas is changing. Twenty years ago, having a startup idea was rare enough that the screening problem was finding any at all. The job of an investor was sourcing. In the next decade, the screening problem inverts. Every developer with an AI assistant produces good ideas weekly. The screening problem is which of these are worth pursuing?

Crowds with skin in the game answer that question better than any other mechanism we have invented. P&L turns that mechanism on, points it at idea-stage projects, and lets it run.

We are generating dreams into realities. That phrase sounds soft until you sit with the literal claim: the things people imagine and articulate to their tools are about to become tokenized, public, capital-formation events, in volumes nobody has prepared for. Most of those events will be small. Some will be enormous. P&L is the rail for the whole distribution.

If that is what's coming, the protocol that meets it has to be open, on-chain, agent-readable, transparent in its economics, and run by the conviction of the people using it rather than the curation of a committee.

That is what we are building. The rest of this documentation is the implementation.

VI. What you should read next

The thesis is committed. From here the docs explain the actual machinery.

• How PNL works — the conviction-market mechanic in detail, with the math
• Economics — fees, distribution, AMM pricing, where every lamport goes
• Lifecycle — state machine, permissions, when an idea graduates
• Public read API — JSON endpoints for reading live market state
• On-chain program — Anchor instruction reference, PDA seeds
• Agent integration — llms.txt, MCP, the path to terminal-native markets
• Transparency — fees, treasury wallets, known limitations, fund usage

The docs are written to be read in order. Each chapter is short enough to read in one sitting; together they explain the whole protocol with no implicit knowledge.

If you only have time for one more chapter, read How PNL works.