Plan. Build. Ship.

AI-driven product development in twelve phases. Twenty eight days of focused work for an experienced team, longer for first timers. This is the operating rhythm behind every AI Implementation we run, refined across a decade of mid market engineering.

The new map Plan · 01 to 03 Build · 04 to 08 Ship · 09 to 12 PRDs are everything Time expectations
When you need a tool that does not exist

Three options. The old map is broken.

For two decades, the choice was buy a SaaS or wait months for engineering. AI changes the middle of that map, and adds a new third option underneath.

Before · the old map
Buy
SaaS subscription
Minutes

Off the shelf, vendor priced, fits the shape of what someone else thought you needed.

Build
Traditional coding
Months

Custom, expensive, slow. Quarter long roadmaps that ship after the problem has already moved.

After · the new map
Buy
SaaS subscription
Minutes

Still the right call when an off the shelf tool fits the shape of the problem cleanly.

Build
AI-driven build
Days

Custom internal tools, ships in days not months, because the spec writes the system.

Automate
Agents with MCP
Hours

Workflow level automation that runs against your own systems and data, in hours.

The new third option changes everything. You can now build what does not exist.

1
Movement 01 · Phases 1 to 3 · Spec the product, build the toolkit

Plan. Sharp spec, shippable software.

The whole speed advantage of AI driven build collapses if the brief is fuzzy. The Plan movement is short by design, but it is where ninety percent of the engineering decisions actually happen. Skip it and you debug your spec in production.

01

Ideation & validation

1 to 2 days

Problem framing, user interviews if needed, kill criteria. We name the smallest version of the product that proves the bet before we touch a tool.

02

Planning & PRDs

4 to 5 days

Sharp written PRDs per surface, edge cases listed, acceptance criteria explicit, success metrics defined. This is the meatiest phase of Plan because every later phase reads from it. Fuzzy here means fuzzy everywhere, so we take the time to make the contract real before any code is written.

03

Design system

~half day

Tokens, type, components, layout primitives. Half a day, not two weeks, because we are codifying constraints for the AI build, not redesigning the wheel.

2
Movement 02 · Phases 4 to 8 · Make the engine, connect it, polish it

Build. Days, not months.

With a sharp spec and a tight design system, the actual code is the easy part. Five phases, roughly two weeks of focused work. The order is deliberate: frontend first so the spec proves itself in pixels before backend complexity lands.

04

Frontend build

6 to 8 days

The biggest single phase. We build the interface against fixtures first so the PRD proves itself before backend work begins.

05

Backend

1 to 2 days

Endpoints, schema, jobs. Short because the contract was already nailed down in the PRD and exercised in the frontend.

06

Wireup

~1 day

Replace fixtures with real endpoints, integrate auth, plug in third party APIs. Where most build phases bog down. Ours doesn't because the contract held.

07

Debugging

2 to 4 days

The honest day. Real users, real edge cases, real timing bugs. We budget for it because pretending it isn't there is how schedules slip.

08

Polishing

1 to 3 days

Microcopy, empty states, error states, loading states, motion. The difference between something that ships and something people actually use.

3
Movement 03 · Phases 9 to 12 · Harden it, get it live, keep it alive

Ship. Live, hardened, maintained.

Shipping is not deployment. It is hardening, refactor, deploy, and the maintenance posture that starts the day after launch. Most teams treat phase 11 as the finish line. We treat it as the start of phase 12.

09

Hardening

~1 day

Security pass, rate limits, input validation, secrets audit, observability. The unglamorous day that prevents the glamorous incident.

10

Refactor + final debug

1 to 3 days

The last pass at the parts you knew you'd come back to. Now the spec is final, the code can be made coherent rather than just correct.

11

Deployment

3 to 4 days

Infra, CI, staging, prod, DNS, observability dashboards, rollback plan. Boring on purpose. The boring part is what stays up at 3am.

12

Maintenance

forever

Patches, dependency drift, schema migrations, on call. The phase no one wants to talk about, which is exactly why it kills more products than any other.

PRDs are everything.

Fuzzy PRD, fuzzy software. Sharp PRD, shippable software.

AI driven build only collapses months into days if the brief is crisp. A team of agents will move very fast in whatever direction you point them. The point of phases 1 to 3 is to point them at the right thing before a single line of code is written.

This is where most teams trying to ride the AI wave stall. It is not the tooling, the model, or the framework. It is the spec. The teams that ship are the teams that learned to write PRDs that survive contact with real users and real data.

BMAD style multi agent setups make this concrete. Each role has an owner, each artifact has a producer. The brief is no longer a wiki page, it is a contract.

Time expectations

Roughly twenty eight days of focused work.

For an experienced builder. Longer for first timers. Each phase has an honest estimate so the team plans calendar time, not hopeful hours.

Bring the method to your roadmap.

Start with AI Diagnose for a written plan you keep. Or book a thirty minute call if you want to scope an Implementation directly.

Start AI Diagnose → Book a 30 min call