Two-Way Doors
Decide fast. Reverse cheaply.
Low-cost, reversible choices: UI layouts, synthetic test data. We execute at maximum velocity in our open-source Prototype Factory sandbox, testing ideas at thought-speed.
02 / HOW IT WORKS A LOGIC-CENTRIC DELIVERY MODEL
How the Factory works.
We restructure how software is built around what an agentic pipeline can manufacture. We don't bolt AI onto a legacy process.
The Premise
Traditional delivery says you can pick two of speed, cost, and scope. AI-native delivery breaks the picking. Here is why. Below is how. On a current UK retail data platform engagement: at least 5x lower cost than a traditional build, and more than 10x faster in pace, measured against documented agency benchmarks. Tracked in live velocity logs, not claimed in slides.
Three zones of ownership.
01 / MODEL01
Intent
The client owns the outcome. You define what success looks like, provide the inputs, and set the constraints. The outcome leads; everything downstream serves it.
02
Orchestration
We translate intent into deliverables. The agentic Code Factory pipeline manufactures microservices, schemas, and test matrices concurrently under senior human oversight.
03
Validation
You verify outputs meet intent. No implementation-approval theatre. We present working outputs; you make the judgement calls that only a human should make.
The shift, in plain English.
02 / THE SHIFTTraditional: long discovery, slow handoffs
- Six to ten weeks of scoping.
- Specs that drift before the build starts.
- Teams documenting work nobody reads.
- Cost stacks up before value lands.
AI-native: one focused week, then code
- Clearer inputs upstream.
- Automation handles the translation work.
- Humans focus on the judgement calls that actually matter.
- Working software starts to land in week two.
Your first week.
03 / WEEK ONESix steps. One calendar week. At the end of it, a brief your delivery pipeline can act on without a single follow-up meeting.
01
Find what each side knows
We sit with the executive sponsor, the functional owner, and the technical lead. We surface what each one knows that the others do not.
02
Split reversible from irreversible
Every decision goes into one of two buckets. Reversible choices like UI layout, naming, packaging (Two-Way Doors) move fast. Irreversible choices like data schema, authentication, core integrations (One-Way Doors) get proper scrutiny. Speed where it is safe. Care where it is not.
03
Pull in what already exists
Old specs, prior tickets, existing schemas, naming conventions, recent post-mortems. Nothing gets rewritten from scratch when a credible source already has the answer.
04
Prototype the outcome before scoping it
Before any production code, we build a throwaway working slice in an open-source sandbox so everyone can react to the shape of the thing. Disagreement surfaces in front of a screen, not inside a 40-page specification.
05
Run automated checks against your repos
A command sweep validates the emerging brief against your real codebase, conventions, and constraints. Gaps show up here, on day four, not in week eight when the build is half done.
06
Lock the brief
The week ends with a machine-readable brief that captures the seven things a delivery pipeline needs before it can run without surprises. See below.
What gets locked before the first commit.
04 / THE BRIEFSeven things. If any of them is fuzzy on the day delivery starts, the build will pay for it later.
1
The outcome
Who has the pain, and what does success look like for them.
2
Success measures
Specific, observable, traceable. Not aspirations.
3
Scope edges
What is in. What is out. What we are deliberately ignoring.
4
Affected systems
Which repositories, services, and data sources the work touches.
5
Blast radius and rollback
If it goes wrong, what breaks, and how we get back.
6
Quality bars
The standards the work has to meet before it ships.
7
Data reality
Real data, synthetic data, or a static fixture. The boundary is explicit.
One-Way vs Two-Way Doors.
05 / GOVERNANCEOne-Way Doors
Cross-examine before you commit.
Critical, irreversible architecture: database patterns, auth models, core API integrations. We subject these to rigorous, evidence-based scrutiny before manufacturing a single line of production code.
Frequently asked questions.
06 / FAQOne focused week of discovery, then code. The discovery week splits every decision into reversible and irreversible, pulls in what your team already knows, prototypes the outcome in a sandbox, runs automated checks against your real codebase, and ends with a locked brief the delivery pipeline can act on without further meetings.
Working software starts to land. On a current UK retail data platform engagement, the build is running at more than 10x the pace of a traditional team and at less than a fifth of the cost, measured against documented agency benchmarks.
Traditional delivery is bottlenecked by sequential handoffs and specs that go stale. Bolting AI onto that process speeds up the typing, not the structure. We restructure the work upstream so that automation can run end to end. Faster typing is incremental. Restructured delivery is a different physics.
Reversible decisions (sometimes called Two-Way Doors) are low-cost choices like UI layout or packaging variants. We move on them fast. Irreversible decisions (One-Way Doors) are choices like database schema, authentication model, or core integration patterns that are expensive or impossible to unwind. We give those proper evidence-based scrutiny before any production code is written.
Two safeguards. First, the prototype in step four surfaces disagreement before scoping is finalised, in front of a working screen rather than a document. Second, every decision in the brief is tagged for reversibility, so when something needs to change in flight, we know whether it is a cheap pivot or an expensive one.
See if it fits See whether it fits
See whether it fits
your build.
We will walk you through a real engagement, the brief that came out of it, and what week two delivered. No pitch deck. No commitment.