floe studio. — The method

The method

Your AI product team, installed in yours.

Product-Aligned Agent Process, the delivery system we bring into engagements. 9 specialized agents, a reactive automation layer, and Agent Teams cover the full product lifecycle, from challenging an idea through shipping and measuring outcomes.

9Agents

13Skills

7Gated stages

3Environments

What changes.

Before

—“Done” is a claim
—AI output you re-check by hand
—Docs and designs drift from the build
—One person prompting hard
—You can't reconstruct what the AI did

After

—“Done” is a gate that passed, evidence attached
—Output arrives pre-verified, tested before you see it
—Drift is caught at the gate, not discovered in production
—Nine agents with defined roles and enforced hand-offs
—Every artifact traces from requirement to test to build

We won't promise you 40% faster. We promise you'll be able to prove what shipped.

Nothing advances on trust alone.

Gates are enforced by scripts and hooks, not by remembering to check. A gate can be waived only by the product manager, only in writing, on the ticket.

Design can't drift.

Built components are compared against the design source element by element before QA opens. Drift is a blocked gate, not a production surprise.

Docs can't drift.

Documentation is verified claim-by-claim against what actually shipped, twice per release. Our drift-detection tooling watches Jira, Confluence, and Figma between releases.

“Done” can't be declared.

Hard gates block, they don't warn: type safety, no personal data in logs, an audit entry on every change. Closure requires a 14-field completion certificate: evidence, not assertion.

Prod is the tested build.

The exact version that passed the dress rehearsal goes live, byte for byte. No rebuilds, no “small last changes.”

Lessons are laws.

Every failure becomes a written prevention rule (trigger, bad pattern, fix), loaded automatically before any similar task starts.

Humans decide.

The system surfaces conflicts and discovers dropped scope, but never settles either. AI challenges, surfaces, and checks; production promotion stays a person's call.

Four phases. One continuous loop.

Ideate

Challenge ideas, run discovery, map opportunities.

Product StrategyDesign

Validate

Test hypotheses, check feasibility, make go/no-go decisions.

Hypotheses

Execute

Build incrementally from validated hypotheses.

ArchitectureBackendFrontend

Measure

Track outcomes, feed learnings back to discovery.

Metrics

Learnings feed back from Measure to Ideate — continuous discovery & delivery

Reactive layer: always on

Hooks fire on events: a doc saved, a risky command, a production push. Scheduled routines run reports and monitoring on their own clock. The process keeps running between working sessions.

Seven stages. Three environments.

Every gate runs before the status changes, never after. Work moves from the workshop (a developer's private space) through the dress rehearsal (a full copy of the product, no audience) to opening night.

Planning· The briefLocal· The workshopStaging· Dress rehearsalProduction· Opening night

1
To Do
Is this actually ready to build?
Complete brief, or blocked with exactly what's missing named.
2
In Progress
Built in the workshop.
Code and tests on a branch; existing parts reused from the catalogue.
3
Code Review
First inspection, machine-checked.
Pass/block verdict: type safety, security, no personal data, audit trail.
4
Ready for QA
The rehearsal opens.
Design-fidelity diff and doc-sync report attached before the status flips.
5
In QA
Every user, every language, every screen.
Every acceptance criterion tested in a real browser, with screenshots.
6
QA Passed
Promotion to opening night.
14-field completion certificate; the identical tested build goes live.
7
Done
Shipped isn't finished.
Outcome metrics defined, tracked weekly, and fed back to discovery.

1 · PlanningThe brief
To Do
Is this actually ready to build?
Complete brief, or blocked with exactly what's missing named.
2 · LocalThe workshop
In Progress
Built in the workshop.
Code and tests on a branch; existing parts reused from the catalogue.
3 · LocalThe workshop
Code Review
First inspection, machine-checked.
Pass/block verdict: type safety, security, no personal data, audit trail.
4 · StagingDress rehearsal
Ready for QA
The rehearsal opens.
Design-fidelity diff and doc-sync report attached before the status flips.
5 · StagingDress rehearsal
In QA
Every user, every language, every screen.
Every acceptance criterion tested in a real browser, with screenshots.
6 · ProductionOpening night
QA Passed
Promotion to opening night.
14-field completion certificate; the identical tested build goes live.
7 · ProductionOpening night
Done
Shipped isn't finished.
Outcome metrics defined, tracked weekly, and fed back to discovery.

Prod = the tested build.

This exact pipeline shipped Nomu and Lighthouse.

See the case studies

How it arrives in your team.

Through an engagement

The method arrives with us, not an installer. The audit maps where it lands in your workflow first.

Configured to your domain

Agents, skills, and gates tuned to your product and process, wired into your Jira, Confluence, Figma, and repositories.

Yours to keep

The system and the skills stay when we leave. Your team runs it, and we stay available as it evolves.

Works in the tools your team already uses

Claude CodeJiraConfluenceFigmaPostHog

Start with an audit.

Start a project

A fixed-scope map of where AI pays off in your delivery, and the process to install it.

PAAP