Shippednot scoped

Evidence / 01

What one person can build now

The interesting part isn't the feature list. It's the ratio.

Constraint

One builder. No engineering hire, no outside capital, no agency. Eight service categories, and two sides of a marketplace that each refuse to arrive before the other.

Built

A homeowner books and pays into escrow. The job broadcasts to a pro network. First claim wins — race-safe under concurrency, so two pros clicking at once can't both take it. On completion, payment captures and transfers. Next.js, Supabase, Stripe Connect, Vercel.

Method

Context first — how quoting actually works, how pros actually behave, where the exceptions live. Then tight correct-and-continue cycles against live conditions, starting at roughly seventy percent of what there was to know.

Result

Live and transacting. The full money path — book, escrow, dispatch, claim, complete, capture, transfer — verified end to end.

So what

Whatever constraint your team is working under, it's a better one than that. The question worth asking is what this approach does with resources behind it.

Method / 02

Context × Loops

AI projects die in one of two places: a discovery phase that produces a document, or a demo that works on the happy path and collapses on real data.

Context. Before anything gets built, the system is loaded with the real material — actual records, actual constraints, the exceptions everyone knows about and nobody wrote down. Most implementations underperform because they were built against a description of the work rather than the work itself.

Loops. Then tight build–test–correct cycles against live conditions. Not one specification followed by one build. Small increments, each checked against reality, each correction feeding the next. The loop is where the specification actually gets written — by the work, not before it.

Why 70%. There's enough to start at about seventy percent of the information. The last thirty percent isn't sitting in a stakeholder's head waiting to be extracted — it doesn't exist yet, and it gets found faster by building than by asking. Waiting for certainty is how a four-week project becomes a four-month one that ships something nobody wanted.

Moving at 70% only works if someone carries the risk of being wrong. On a fixed fee against a fixed date, that's me. If the loop runs slow, it's my cost, not yours.

Terms / 03

How the work is structured

Ownership / 04

What can be owned outright

Advice is cheap and everywhere. The useful arrangement is handing over a whole function to someone accountable for it — a clean win on both sides, with nothing left ambiguous about who has it.

Most engagements start with a paid one-week diagnostic — a scoped plan and a working proof of the hardest part, $3,500, credited against the build if it goes ahead.

• AThe process nobody owns. The manual step everyone works around and no one has time to fix. Scoped, built, shipped, documented, returned.3–4 weeks · fixed fee from $10,000
• BThe gap between a good model and a hard customer. For AI companies selling into construction, trades, field service, property, logistics. Excellent technology, an audience that won't answer the phone. That gap is the whole job.4+ weeks embedded · fixed fee or monthly $12,000/month
• COverflow delivery. For firms with more demand than bench — someone who can sit in front of the client and build the thing.Per project or day rate $1,400/day
• DZero to one. An idea that needs to reach production rather than demo. Payments, edge cases, the parts that only appear under real load.6 weeks to production · fixed fee from $18,000