Menu

Products

GSF-E Governed Software Factory for Enterprise software

GSF-E is a governed software factory from AI for Products. It guides your AI coding assistant (today, Claude Code) through 11 phases of building enterprise software, with a gate after each that only a person can sign, and locks against risky actions. You only chat. Your code, documents and the signed record stay in your repository.

Get early access See the GSF-E method

Release candidate 9 · November 2026

  • 11phases, from idea to learning
  • 6stages of Anthropic's AI-native SDLC
  • 11gates, one after each phase, signed by a person
  • 3tracks, from a full product to a prototype
  • 6ways in, existing code included
  • 179step cards your AI follows

Why GSF-E

The code is no longer the bottleneck. The decision is.

Teams are building faster.

  • 97%of 677 product people Product Focus surveyed said AI made them more productive
  • 64%said their product outcomes improved
  • 89%of executives in Atlassian's survey say AI increases speed
  • 6%are sure they have clear examples of organisation-wide return

Marty Cagan calls this the AI productivity paradox. Andrew Ng calls it the product management bottleneck: “the holdup is deciding what to build.” When building is cheap, what remains is deciding what is worth building, writing it down so an agent can build it, checking what comes back, and knowing what it cost. The tools product teams use were built for none of that. AI for Products exists to build them, and to teach the people who will use them.

Sources: the Product Focus 2026 survey of 677 product people; Atlassian, State of Teams 2026; Marty Cagan, SVPG, 2026; Andrew Ng, The Batch, issue 310, 2025.

The method

A delivery loop that people steer

Your AI does most of the work at every stage, and each stage leaves a record the next one reads. A person signs a gate after every phase, and the record shows who approved it and when.

  • Go/No-Go. After the idea, the product manager decides whether to go on.
  • Specs and API contract approved. For each feature, the product manager and the tech lead approve what will be built. Product code can't be written before a feature reaches Build.
  • Production deploy. The way to ship is the deploy workflow on GitHub, after the tech lead signs the release. Before this gate, a release goes out only with its reason kept in the record, such as an urgent fix, and the gate is still owed.
The delivery loop Six stages in a loop: plan, design, build, test, deploy and maintain. Each stage leaves a record the next stage reads: IDEAS.md, SPECIFICATION.md, BUILD-LOG.md, TEST-RESULTS.md, DEPLOY-REPORT.md and RETROSPECTIVE.md. Eleven gates sit on the loop, one after each phase, and a person signs each one. A breach in production starts the loop again. Each stage leaves a recordthe next stage reads. a breach in productionstarts the loop again PlanIDEAS.md DesignSPECIFICATION.md BuildBUILD-LOG.md TestTEST-RESULTS.md DeployDEPLOY-REPORT.md MaintainRETROSPECTIVE.md
  1. Plan IDEAS.md · 3 gates: Go/No-Go, Problem validated, Solution scoped
  2. Design SPECIFICATION.md · 3 gates: Flows and mockups approved, Data schema review, Specs and API contract approved
  3. Build BUILD-LOG.md · 1 gate: AI code review
  4. Test TEST-RESULTS.md · 1 gate: Security review
  5. Deploy DEPLOY-REPORT.md · 1 gate: Production deploy
  6. Maintain RETROSPECTIVE.md · 2 gates: Launch readiness, Retro complete
The six stages follow Anthropic's AI-native SDLC playbook (2026). The records and the gates are GSF-E's.

The map

Eleven phases on six stages, a gate after each

Each phase ends at a gate: a checklist, every item marked as checked automatically or by a person, then a person's sign-off.

  1. Plan

    1. IdeaGo/No-Go PM
    2. Problem validationProblem validated PM
    3. Solution designSolution scoped PM + TL

  2. Design

    1. DesignFlows and mockups approved PM
    2. ArchitectureData schema review TL
    3. SpecsSpecs and API contract approved, per feature PM + TL

  3. Build

    1. BuildAI code review, per feature Dev + TL

  4. Test

    1. TestSecurity review TL

  5. Deploy

    1. DeployProduction deploy TL

  6. Maintain

    1. LaunchLaunch readiness PM
    2. LearnRetro complete PM

This is the full track, with its usual approvers: PM is the product manager, TL the tech lead. The mvp track keeps gates 2, 3, 6, 7, 8 and 9; the prototype track keeps 2, 3 and 7 and never goes to production. Gates 8 and 9 can't be skipped: a release can go out before them only with its reason kept in the record, such as an urgent fix, and they are still owed. Existing code starts with a Brownfield assessment and its own gate. On a solo project, you sign for every role.

Who decides

People decide. AI approval never counts.

Your AI does the work. A person signs it off, and your project's record keeps the proof.

Named and dated

Every gate records who approved it, by name, and when.

Never the AI

An AI approval never counts. Your AI cannot sign its own work.

Someone else approves

A person other than the author approves the final commit, and two people for sensitive paths.

A human merges

A person merges each change, and the tech lead signs each release before it goes to production.

How it works

Three steps, then you just talk

  1. STEP 1

    Connect your project

    Add the GSF-E plugin to Claude Code, open your project and say "Set up this project with GSF-E". It adds a small safety kit to your repository: house rules for your AI, locks against risky actions, and GitHub checks. You confirm once.

    Signed by you

  2. STEP 2

    Just talk

    Say "Let's start a project", then "What's next?". Your AI asks where you are starting, how big the project is and who is on it, then does each step with you. Your documents and code are written in your own repository.

    Recorded when you say yes

  3. STEP 3

    Sign the checkpoints

    Ask "Are we ready to move on?". The checkpoint is checked, and a person signs it. The signed record is saved in your repository, so a change made by hand is caught, and code stays locked until it is put right.

    Checkpoint signed by a person

What's inside

A method, a safety kit and a record you keep

GSF-E is one source for building software with AI agents while people stay accountable: a method that says what to do and who decides, and a kit that enforces it in your tools.

The method

The 11 phases, written as 179 step cards. Your AI gets the next card only when your project reaches it, and does that step with you. Every gate is a checklist, each item marked as checked automatically or by a person.

The safety kit

Added to your repository when you set up:

  • Locks on your AI's own tools: no hand edits to the record, no product code before a feature reaches Build, no production deploy before its gate unless a reason is kept in the record, no destructive commands, no reading secrets.
  • Checks on GitHub: your project's install, lint, test and build, free security scans, and a production deploy workflow.
  • The watchdog opens an issue when a change reaches main without a person's approval.

The signed record

Which steps are done, which gates passed and who signed them: your project's record lives in your own repository, signed by the GSF-E service. A record changed by hand is refused, and code stays locked until it is put back.

Three tracks

  • Fullall 11 phases, 11 gates
  • MVP16 fast-track cards, 6 gates
  • Prototype3 gates, and never goes to production

Switching tracks is the tech lead's call, with the reason kept in the record.

Six ways in

  • A new product
  • A validated idea
  • Specs ready
  • Existing code, with a Brownfield assessment first
  • Partway through
  • A production metric out of bounds

Who it's for

For people who build products with AI

For product managers, founders and small teams who build with an AI coding assistant and want the speed without losing control.

  • Product managers who want each step done in order, and a say before the work moves on.
  • Founders and solo builders who want a careful method without the paperwork.
  • Teams who want a clear record of who approved what, and when, kept in their own repository.
  • What you need: Claude Code, git, the GitHub CLI, a repository on GitHub, and Node.js 18 or later. Then you only talk.

Status: release candidate 9 (November 2026). Every check, test and end-to-end walk passes on GitHub, on Linux and Windows. It has not run on a real product yet; the first pilot is next.

Pricing

Free for 1 month, then simple per-seat pricing

Every person gets their first month free, counted from their own sign-up. After that, each seat is paid by the team's owner or the company.

In India

₹199per seat, per month

or ₹1,499 per seat, per year (about a third less)

No GST is added at present. If GST registration happens, this price stays the same, with GST included.

Outside India

$5per seat, per month

or $39 per seat, per year (about a third less)

Prices follow your billing country: rupees in India, US dollars elsewhere. Not yet offered in the EU, the EEA or the UK.

  • 1 month free from each person's sign-up.
  • Seats are paid by the team's owner or the company, one seat per person on the team. A solo user pays for one seat.
  • If a plan is unpaid, new steps stop. Your code, documents and records stay yours, the locks work as before, and disconnecting from GSF-E always works.
  • Cancel any time. See the Refund and Cancellation Policy.

The online GSF-E service opens soon. Each person's first month is free.

Your work stays yours

Questions people ask

Does GSF-E see my code?

No. GSF-E is built so that your AI writes your code and documents only in your own repository. During each call the service sees only your project's record (which steps are done, which checkpoints passed and who signed them) and an outline of each document it checks (its labels and headings, not its text). It keeps no copy. The one thing it keeps about a project is who may act on it (which accounts are on its team), held with the accounts by our sign-in provider.

How much does GSF-E cost?

Each person's first month is free, counted from their own sign-up. After that, a seat costs ₹199 a month in India or $5 a month outside India, or ₹1,499 or $39 a year, paid by the team's owner or the company. The online GSF-E service opens soon.

What happens if I stop paying?

New steps stop. Your code, documents and records stay in your repository, the locks work as before, and saying "disconnect from GSF-E" always takes them off.

What if the GSF-E service ever ends?

You get at least 60 days' notice. Your project's record is always in your repository, and your AI can save a copy of it at any time without the service. The locks keep working from your last signed record, and "disconnect from GSF-E" takes them off without the service.

Which AI coding assistants does it work with?

Today, Claude Code. You choose and pay for it yourself, under its own terms.

Does it work if I'm on my own?

Yes. On a solo project you sign for every role and approve your own work, and the record still shows each step and who signed it. On a team of two, your teammate approves. The "someone else approves" rule works fully with three or more people. On GitHub Free, a change that reaches main without a person's approval is not blocked, but the watchdog opens an issue about it.

Get early access

Tell us a little about your team and what you are building. We will be in touch as places open.

Get early access Launch it with your team