What greenfield software development looks like on a twelve-week clock

How greenfield projects start

A greenfield project starts with an empty repository and a room full of opinions.

Nothing in the codebase argues back. No integration deadline forces anyone to commit. You can spend four months building the wrong thing, and the system will let you, right up until the first real user opens the product and doesn’t come back.

CB Insights went through a pile of startup post-mortems and put one cause at the top. 42% of those companies built something the market didn’t need. Most people reading this article might not be working at startups; however, the same thing happens to well-funded product teams inside established companies, and it happens most quietly in greenfield.

What greenfield software development actually means

It means building from scratch. An empty repo, and none of the architectural decisions someone made in 2019 that you’d otherwise be working around for the next two years.

Pure greenfield is rare. You’ll be wiring into an identity provider somebody else already chose by week six anyway. The part that matters still holds: you pick the architecture and the stack on the merits, and no legacy schema gets a vote.

When to consider greenfield

If the system already exists and the job is to bring it forward, that’s legacy application modernization, and it runs on a different clock.

Where AI changed the math

Google’s DORA team surveyed technology professionals in 2025. 90% now use AI at work, and more than 80% say it made them more productive. So far, so good.

The same research found something else. As AI adoption went up, delivery throughput rose, and delivery instability rose right alongside it. Same curve, same direction. 30% of developers said they have little or no trust in the code their own tools produce.

DORA’s read is that AI amplifies whatever it lands on. A team with a decent internal platform and real testing discipline compounds. A team with fragmented tooling and fragile infrastructure compounds too, but in the wrong direction.

Greenfield is the highest-stakes version of that. In week two, you choose what gets amplified for the next three years.

The Notch 3S Method™

Freedom without structure is how greenfield MVPs die, slowly. Scope drifts because nothing forces a decision. Version one grows a settings page, then a permissions model, then an admin view, and by the time anyone asks what it was supposed to prove, the people who could have answered have moved on to the next project.

NotchForge™ runs underneath all three phases. It’s our delivery framework: the documented standards that set how our senior engineers and coding agents work together, from the first working session through deployment. The same framework governs every Notch engagement, so the standards don’t shift with the team you get or the quarter you start in.

SCOPE. Weeks 1- 3

Working sessions with your decision-makers and our senior team. The pain points, the outcomes you need, and the argument about what version one actually has to prove.

Sovera AI™, our agent-orchestration platform, runs the domain research behind it all. It surfaces the edge cases nobody thought to mention and flags the shaky assumptions before the session starts, so the time in the room goes to decisions.

You leave with a working prototype and a plan you’ve signed off on. Nothing enters the Ship phase that you haven’t already seen.

SHIP. Weeks 4 to 12

The plan becomes working software, deployed in production, in the hands of real users.

This is where NotchForge™ does the work. In Ship, this means addressing which agents can touch what, and what must be true before their output moves anywhere. Agents take the implementation work that sits below senior attention. Engineers spend the week on the calls that decide whether the thing holds up under real load.

Every piece of work starts from a written specification with explicit acceptance criteria. Agents operate inside a defined harness with the policy checks and observability already wired in. Output arrives as small reviewable diffs, with tests written alongside them.

One rule sits under it all. Every commit has to be explainable to another engineer before it lands. That rule is what stops the codebase from reaching the eighteen-month wall, the point where a fast AI-assisted build gets tangled enough that maintaining it costs more than starting over.

You see working software every week. You give feedback; we act on it in the next cycle. By week twelve, you have a live MVP with real users and a repo that’s yours.

STRENGTHEN. Ongoing

Same senior team. Two tracks running in parallel.

1. Harden covers what ten users never surfaced and ten thousand will: performance under real load, security, observability.
2. Evolve covers the next set of user journeys, ranked by what your first cohort actually did with the product.

Monthly checkpoints, and the product is measurably better at the end of each one.

Why does twelve weeks hold up

Twelve weeks used to mean cutting something. Usually reliability, sometimes quality, occasionally both, while nobody said so out loud.

Agents narrow that trade-off. Sovera AI™ compresses the domain research that used to eat the first month of a greenfield project. NotchForge™ is what makes that speed safe to take. Every change passes the same checks whether a person or an agent produced it, so throughput goes up, and delivery stays stable.

The judgment sits exactly where it always sat. Senior engineers, 80% of our team, make the architectural calls that define whether the product survives its first year of real use. Agents clear the space around those calls. The product that comes out the other side is one your team can still change in year two.

What twelve weeks won’t get you

A 100% finished product. It gets you an MVP – a real one, in front of real users – on a codebase that can absorb whatever they tell you. If your version one has to be feature-complete on launch day, the Notch 3S Method™ isn’t the right fit, and you’ll hear that from us in the first session.

Start scoping your project

Notch has been building since 2014. 300+ projects, 50+ clients, a 4.9 on Clutch, and an average client relationship of five years.

Launch your MVP with real users in twelve weeks.  Book a scoping call