Opinionated, because opinions save weeks.
Four phases, in order. Each phase has a fixed duration, a fixed scope, and a written artefact. Nothing is bespoke except the work itself.
Phase 01
First call
20 minutes
A scoping conversation, not a pitch. We listen to the problem, share an opinion, and together decide whether an audit makes sense.
You
- Describe the feature or workflow you want to build
- Share any rough timeline or AI-Act deadline pressure
- Tell us honestly what you have tried already
Us
- Ask enough questions to understand the shape of the problem
- Tell you within the call whether the audit fits — or whether you would be better off with a different partner
- Send a short written summary by end of day
Artefacts
- Written call summary
- Clear yes / no / not-yet recommendation
Phase 02
AI Opportunity Audit
1 week elapsed · about 20 combined engineering hours
A one-week, fixed-scope engagement. You get a ranked map of AI features worth building into your product, plus a working prototype of the top candidate.
You
- Grant temporary, least-privilege access to the product and one representative data slice (or an equivalent synthetic slice we will build)
- Dedicate ~2 hours of CTO / Head of Product time across kickoff and review
- Nominate one internal champion for mid-week questions
Us
- 90-minute kickoff workshop to pin down constraints and success criteria
- Discovery across your product, data model, roadmap, and regulatory context
- Ship a working prototype of the highest-ROI feature, runnable locally
- 60-minute review session to walk through findings
Artefacts
- Opportunity map document — 3–5 candidates ranked by ROI, risk, and AI-Act tier
- Runnable prototype with README
- Short recorded walkthrough you can share internally
Phase 03
AI Feature Implementation
3–4 weeks elapsed · about 60–80 combined engineering hours
We pick up the top candidate from the audit and build it for production. Fixed fee, fixed scope, fixed cadence.
You
- Grant repository access with branch-protection rules we can honour
- Keep an internal reviewer available for one weekly 30-minute sync
- Commit to a single source-of-truth for decisions (we default to a shared notion or linear doc)
Us
- Daily commits to a shared branch — no long-lived forks
- Tests, observability, and deployment handover included in scope
- Human-in-the-loop checkpoints documented where the AI Act expects them
- Security and data-flow documentation written for your compliance reviewer, not us
Artefacts
- Production-ready feature merged to main
- Model-choice memo, data-flow diagram, and human-oversight plan
- AI-Act risk-tier classification with cited obligations
Phase 04
Handover + 30-day review
30 days of quiet support
You own the code the moment it merges. We stay reachable for one month to answer questions, fix any post-launch issues, and hand over cleanly.
You
- Flag anything unexpected in production
- Forward any questions from your compliance reviewer
- Tell us honestly where we over- or under-delivered
Us
- Respond within one business day for the full 30 days
- Fix any bug traceable to our implementation at no additional cost
- Write up the engagement as an internal lessons-learned, shared with you
Artefacts
- Post-launch review document
- Open-ended invitation back — at no obligation on either side
Constraints
Six things we will not negotiate.
Not because we are inflexible. Because every one of these came from a lesson we already paid for — or saw a peer pay for — and we would rather be upfront than surprise you later.
01
Audit before implementation
Without an audit, the scope is a guess. We have yet to meet a team whose first-guess feature was the one actually worth building.
02
Fixed fee, never hourly
Hourly billing rewards slow work. Fixed fee aligns our incentives with yours: ship well, ship on time, and move on.
03
We turn down custom one-offs
Custom one-offs are how small services companies drown. Our scope this year is deliberately narrow. We would rather refer you to a better fit than pretend.
04
You own the code
Everything we write ships to your repository. We keep no proprietary layer you must license forever.
05
Async by default
We do one scheduled 30-minute sync per week. Everything else is written. This is how two people serve multiple clients well.
06
No NDAs for first conversations
We will sign a mutual NDA before the audit begins. Signing one to talk for 20 minutes is a signal that the engagement will be painful.
Ready for a 20-minute scoping call?
No NDA. No proposal. No sales call. If an audit does not fit, we will say so on the call.