From Prompt to Production: A Founder's Playbook for Shipping AI-Built Apps

How do I take an AI-generated prototype to production?

You take it the same way any code goes to production: by adding the layers AI tools skip. Authentication, observability, deployment pipelines, error tracking, database migrations, and the edge-case handling that AI generators don't think about because they were never shown a real production incident. The code you have is usually fine — it's structurally sound, it works on the happy path, and it looks polished enough to demo. The problem isn't the code. It's everything around it. The path from prototype to production is well-trodden: audit what's missing, scope ruthlessly, layer the missing infrastructure in, refactor only what blocks the launch, set up observability, then run a launch dress-rehearsal. Most teams complete the productionization in 4–12 weeks. Founders typically estimate 1–3.

Why prototypes stall at 60%

AI coding tools are optimized for one thing: generating code that looks like what was asked for. They aren't optimized for the boring, invisible 40% that makes software actually work in front of real users — handling expired sessions, malformed inputs, network timeouts, deployment rollbacks, alarms when things break at 3am, and the hundred small decisions a senior engineer makes without thinking. A prototype is software that works once. Production is software that works always. That gap is where teams burn months. They keep tweaking the AI prompt, regenerating components, switching tools — none of which builds the missing infrastructure. The fix isn't a better prompt. It's bolting on the layers AI doesn't generate.

“”

The 6-step playbook we use

When a founder shows up with a vibe-coded prototype, we run the same playbook every time. Step 1: Audit. We read the code, identify the missing layers, and produce a punch list. Most prototypes have between 12 and 25 production gaps. Step 2: Scope. Cut everything that isn't required for the launch. The features that survive get production-grade treatment; the rest goes in the backlog. We've covered the scope-cutting framework in detail — see our MVP scope-cutting framework. Step 3: Add the missing infrastructure — auth (typically Clerk or NextAuth), error tracking (Sentry), monitoring (basic metrics + alerts), deployment (Vercel or Railway), and database migrations. Step 4: Refactor for the actual bottlenecks, not for aesthetics. If the code works and is testable, it stays. Step 5: Build the observability layer — structured logs, traces, alerts on the metrics that matter. Step 6: Run a launch dress-rehearsal with load and chaos tests before real users see it. This is also where toolchain choices matter; our vibe coding acceleration playbook covers the senior-pair model in detail.

How long does productionizing take?

The honest answer ranges from 4 weeks to 12 weeks depending on three things: how much code already exists, how much technical debt the AI tools left, and whether you have compliance requirements. A 5,000-line prototype with no compliance, single product, no integrations: 4–6 weeks. A 15,000-line prototype with a payment integration and basic auth: 6–10 weeks. An app that handles regulated data (health, finance, EU users): 10–16 weeks because compliance work doesn't shortcut. Founders almost always estimate 1–3 weeks because the existing code looks done. It isn't — and the gap between "looks done" and "actually done" is what we close. For most engagements we scope inside the MVP development framework so timelines stay predictable and the founder knows the price before we start.