"Weeks, not quarters" reads like a slogan until you make someone show their work. So here's ours.
Nothing below is about working harder or cutting quality. Every item is a specific piece of overhead that traditional software delivery treats as unavoidable — and the thing we replaced it with.
1. Delete the hand-off
The single biggest tax in software isn't writing code. It's the transfer of understanding between people: strategy → design → engineering → QA. Every hand-off loses context, and every loss gets paid back later as rework.
Replaced with: one small team that carries a piece of work from decision to production. The person who understands why also decides how. There is no document whose only job is to move an idea to the next department.
2. Delete the specification phase
Long specs are an attempt to be certain before building. It doesn't work — you learn the real requirements by looking at working software, not by describing it more carefully.
Replaced with: a one-page decision doc (the loop, the data model, the non-goals) and then a working URL by day three. Feedback on a real thing beats sign-off on an imagined one, every time.
3. Delete the waiting
Most calendar time on a project is not work. It's queueing — waiting for review, for environments, for the weekly sync where the decision will be made.
Replaced with: preview deploys on every change, decisions made in the thread where the work is, and a standing rule that nothing waits for a meeting that a message can settle.
4. Delete the ceremony
Estimation theatre, sprint rituals that report on rituals, status decks assembled to describe work that could simply be shown.
Replaced with: a live environment. Status is the deployed build. If someone wants to know where things are, they open the link.
5. Delete the busywork
Boilerplate, scaffolding, test fixtures, migrations, glue code, the third variation of the same form. Genuinely necessary, genuinely not where craft lives.
Replaced with: AI leverage, applied at exactly this layer. Not "AI writes the product" — AI removes the volume so the humans spend their hours on architecture, product judgment, and the details users feel. That's the difference between AI-native and AI-decorated.
6. Delete the false scope
Half the timeline in a typical project is features nobody asked for, added because they seemed obviously needed.
Replaced with: ship one loop, watch real usage, then decide. Most of the cut list turns out to be right. Some of it comes back — now with evidence, which is the only good reason to build anything.
Speed is not the opposite of care. It's the absence of waste.
What stays slow on purpose
This is the part people skip. Some things must not be compressed:
- The data model. Decided carefully, changed rarely, worth an extra day.
- Security and permissions. No shortcuts, ever.
- The details users touch. Empty states, error copy, loading behaviour, motion. This is where a product feels expensive or cheap, and it's the last place to save time.
Everything else is negotiable. Those three are not.
The honest limit
Not every project fits in weeks. Regulated systems, large migrations, and deep integrations with slow counterparties have irreducible calendar time — usually created by other people's processes, not by engineering.
The method still applies: you can't remove the external wait, but you can remove all six of the internal overheads while you wait. Most teams do the opposite, and then blame the counterparty.
Read the companion piece: the two-week MVP that isn't throwaway code. Ready to run this on your project? Start here.