Can you really build an AI MVP in 28 days?

The skeptical question

Every founder we talk to asks some version of "is 28 days realistic?" The honest answer: yes, but only if you're willing to accept four constraints. Most founders aren't until they've already burned a quarter on a slower process.

This post lays out the constraints, the tradeoffs, and what actually happens during those 28 days. If you can't accept the constraints, the 28-day timeline isn't for you — and that's useful to know early.

Constraint 1: One feature, fully shipped

The most common reason 28-day timelines slip: scope creep. "Can we also add X?" is a normal request — and the answer in this engagement model is "no, in this sprint."

We ship one production-ready AI feature. Not three half-finished ones. Not a feature plus an admin panel plus a billing page. One feature, end-to-end.

What counts as one feature: a customer support agent. An onboarding agent. A pricing-recommendation engine. A KYC review co-pilot.

What doesn't fit: "AI everywhere across our app." That's a 6-month roadmap, not a 28-day sprint. We can ship the first feature in 28 days, then continue building the next on a retainer.

Constraint 2: Founder availability in week 1

Days 1–3 (the Inference phase) require ~10 hours of your time. Discovery questions, decision graph, architecture review. This is the highest-leverage time you'll spend on the project — every hour saved here saves five days of engineering rework.

If you can't carve out that week, the timeline doesn't work. We've had founders try to do this in 30-minute slots between meetings. It always extends the engagement by 1–2 weeks because decisions are deferred.

Constraint 3: One stack decision

During Days 4–7 (Compile phase) we lock the stack: Next.js or Remix. Supabase or PlanetScale. OpenAI or Anthropic. Pinecone or pgvector. Vercel or AWS.

The constraint: we don't change these mid-build. If you decide on Day 14 that you'd like to switch from Supabase to Firebase, the engagement extends — and we don't cover the cost of the rebuild.

This sounds restrictive but it's how every successful 28-day shipment works. Stack indecision is one of the top reasons MVP timelines slip.

Constraint 4: Pre-existing infrastructure or willingness to use ours

If you have existing infrastructure (a deployed backend, a CI pipeline, a domain), great — we integrate. If you don't, we deploy on a Skygnosis-template stack: Vercel + Supabase + your domain. Either works.

What doesn't work: arriving at Day 8 and saying "actually we want to use AWS Lambda + DynamoDB + custom CI." That's a different engagement.

What actually happens during the 28 days

Days 1–3: Inference

The deliverable is a decision graph: 30 questions answered, 3 founder decisions locked, 1 architecture spec drafted. By the end of Day 3, you know exactly what we're building and we have a binding scope to build against.

Most founders are surprised by how rigorous this phase is. The questions are pointed: What's the worst-case behavior we'll tolerate? What's the cost ceiling per interaction? Who reviews edge cases when the agent is uncertain?

These are decisions you'd normally make at week 6 of a slower project. Making them upfront is what enables the 28-day timeline.

Days 4–7: Compile

The deliverable is the architecture artifact — a 4–6 page document that's the binding contract for what gets shipped. Includes:

System diagram (data sources, agent, tools, integrations, outputs)
Stack choices (with rationale)
API surface
SLA targets (response time, uptime)
Cost projections (per-interaction and monthly)
Rollout plan (canary, feature flag, full launch)

We review this with you on Day 7. After your sign-off, this document is locked.

Days 8–24: Execute

Seventeen days of build. The structure:

Daily Loom updates (3–5 minutes, what shipped today)
Weekly stakeholder demo (30 min, working software)
Continuous deployment to staging from Day 10

Week 1: scaffolding, auth, data layer. Week 2: AI feature core, integration with existing systems. Week 3: UI, polish, testing, infrastructure.

By Day 24, the system is in staging end-to-end and we're fixing bugs.

Days 25–28: Deploy & Evolve

Day 25: staging review with founder
Day 26: production deploy
Day 27: monitoring + handover doc finalized
Day 28: handover meeting, IP transfer, support window opens

The 30-day support window

Days 28–58 are post-launch. We fix bugs found in production, monitor performance, and answer questions from your team. This is included.

Most engagements transition from support to a $4.5k/mo retainer after Day 58 — but that's your call.

When 28 days doesn't fit

The real test: can the project be reduced to one feature with one stack decision and one founder making decisions in week 1? If yes, 28 days works. If you need three features, or you need stakeholder buy-in across departments, or you're in a regulated industry with full audit trails — that's a 60–90 day Enterprise AI Pilot, not a 28-day MVP.

We'll tell you which one fits during discovery.

The guarantee

If we don't ship in 28 days, the next sprint is on us. We've never had to extend at our cost in 30+ engagements. The constraints above are how that math works.

Read the full 28-Day AI MVP Playbook for the discovery question list, the architecture artifact template, and the daily Loom format we use.