How to build an AI app in a week (the realistic playbook)

When we tell people we ship working AI apps in a week, the most common reaction is a raised eyebrow. The second most common is 'yeah but how.' This is the how. The literal sequence of days we follow, what we ruthlessly cut, and where we slow down on purpose.

Day 0: The one focused call

Before any code, there's a 45-minute call. We don't ask about your org chart or your 'strategic vision.' We ask: what's the specific task, who does it, how often, and what does the input and output look like. If we can't draw the app on a napkin at the end of the call, the scope isn't right yet and we say so.

Day 1: Scope, not discovery

By the end of day one, you have a written scope: one page, a sketch of the UI, the exact prompt strategy, and the fixed price. No Gantt chart. No 40-page proposal. We'd rather get it wrong on paper cheaply than right on paper expensively.

“A scope doc longer than the app is about to be is a red flag. It means the agency is selling paperwork, not software.”

Day 2: The ugly working version

By end of day two, there's an ugly, live, working version running on a preview URL. It's the wrong color, the copy is placeholder, the layout is a disaster. But it accepts your real input and returns a real AI output. You can poke it.

This is the single most important day of the week. It forces every unknown unknown to the surface. 90% of the 'surprises' that derail longer projects show up on day two. We'd rather meet them in hour ten than in week ten.

Day 3: Prompt tuning and structure

Day three is spent almost entirely on the prompt and the response structure. A good prompt is worth more than a good model, and both are worth more than a fancy UI. We iterate the prompt against a handful of your real examples until the output is consistent, voice-appropriate, and hard to break.

Day 4: Design and polish

Now it gets pretty. Real typography, real colors, real microcopy, loading states, empty states, error states. The thing starts to feel handmade instead of scaffolded. This is also where we add the small details that make an AI app feel alive: copy-to-clipboard, smart placeholders, example inputs.

Day 5: Ship it

Domain connected. Analytics in. A short handoff video so your team knows what's running and where. By Friday afternoon you have a live app on your domain that your team can use on Monday morning. And if Monday morning reveals something we missed, we fix it. Post-launch tweaks are priced in, not an upcharge.

What makes shipping an AI app in one week possible?

• Small scope. We kill feature creep in the napkin sketch, not in week three.
• One team. Same two humans designing, building, and shipping. No handoffs, no ceremony.
• Modern foundations. We don't reinvent auth, hosting, or LLM orchestration every time.
• Candour. If we can't do it in a week, we say so on day zero and quote Pour Over instead.

Is building a real AI app in five days realistic?

Shipping fast doesn't mean shipping sloppy. It means refusing to do the parts of the job that don't move the product forward. No discovery theatre. No consensus-building middle management. No status meetings about status meetings. Just design, build, ship, learn.

If you have a small, stubborn problem you'd like quietly solved by next Friday, we'd love to hear about it.