Know what to build before your agent builds it.

A planner is the full feature list for one kind of software, counted from real products. Your coding agent asks what you want, then builds from your answers.

See the planner

$49 once per planner. Works with Claude Code, Cursor and Codex.

A restaurant system built by an AI agent from the restaurant planner. Opens the live app.
The planner asks

15 of 29 restaurant POS products have a kitchen display. Do you want one?

IncludeLaterSkip

An agent builds whatever you describe. Most people leave things out.

“I want to start my own SaaS.”

You know the market you want to sell into. You may not know what the products already there treat as standard.

“A client wants it built.”

Go through the feature list with them, agree what is in and what waits, and quote from that.

“I already tried with an agent.”

It looked finished, then broke when a real person used it. No roles, no audit trail, features only half built.

“I pay for software that nearly fits.”

You have outgrown the tool, or you pay per seat for the small part you use. Building the version that fits is realistic now, if you know what the current product does.

Sample questions, real counts

All 29 restaurant POS products have menu categories and items. Include them?

Include

28 of 29 take payments. Record them by hand, or connect a card terminal?

By hand

15 of 29 have a kitchen display. Do you want one?

Later
Written to CLAUDE.md

How it works.

  1. Buy it and unzip it into your project.

    Pick your market, pay once and download the zip from your library. Put the folder in your repo, next to your code. About 2 minutes.

  2. Point your agent at it.

    Open Claude Code, Cursor or Codex in that project and send one line:

    Read START-HERE.md and act as the planner.
  3. Answer the planner.

    It goes through the feature list with you. For each feature you say include, later or skip. You do not need to know the market, because the count already does. 30 to 60 minutes.

  4. Tell your agent to build.

    The planner writes the spec into CLAUDE.md or AGENTS.md: scope, build order and what counts as done. Your agent builds from that.

Not sure what a planner is, or whether you need one? Read the FAQ.

We give the same planner to every model.

Then we audit what each one built and publish the score, including the bad ones.

Read the full board

Claude Fable 5.1Best Overall
97
GPT-6 AstraBest Hardening
92
Grok 4.7Leanest Stack
80
Gemini 3.8 FlashFast-Tier Build
68

Online community: the model build-off build-off, 4 models, audited 23 Sep 2026. Scored as found at audit; later fixes are noted, not re-scored. 14 build-offs so far, 65 builds audited.

Click around the apps before you buy.

Each one was built by an AI agent from a planner. They are live, with sample data.

See all 68 live apps

16 markets. $49 each.

Pay once. Lifetime access, and free updates as the count grows. The number on each card is how many real products or sources were read for that market.

Browse all planners

What are you building?

See the planner

$49 once per planner. Works with Claude Code, Cursor and Codex.