A five-day MVP, hour by hour: what "fast and affordable" actually means
Last Thursday a founder forwarded me a quote: twelve weeks, ₩30,000,000, five people listed on the cover page. He asked, reasonably, how anyone can promise the same app in five days without lying. I build web apps and s
Last Thursday a founder forwarded me a quote: twelve weeks, ₩30,000,000, five people listed on the cover page. He asked, reasonably, how anyone can promise the same app in five days without lying.
I build web apps and small SaaS products alone, with AI tools doing a large part of the typing. Five days is real, but not for the reason people assume. It is not that the code comes out faster. It is that almost none of the twelve weeks was code.
Where the twelve weeks actually go
Take an ordinary booking app. The agency quote breaks down into discovery, wireframes, design, backend, frontend, QA, deployment, and a buffer. Five people, three months, fifteen person-months. At ₩3–4M per person-month, the raw cost is already above the quoted price, which tells you those five people are spread across several projects and your app gets a slice of each.
The largest block of time is not building. It is deciding, re-deciding, and waiting for the decision to travel between five people who each own one part of the answer.
When one person holds the whole thing, that travel time disappears. That is most of the five days, and most of the price difference. What is left is genuinely engineering.
What AI compresses, and what it does not
AI tools are very good at the first 80%: screens, forms, CRUD, routing, boilerplate, a working prototype by the end of an afternoon. If you have built something with Cursor, Lovable, Bolt or v0, you have seen that 80%.
The 20% is where launches die, and it barely moved:
- The data model. The list screen looks fine and fires 200 queries because the AI wrote a loop with a fetch inside it. Fine at ten users, fatal at a thousand, and expensive to fix later because the screens were built on top of it.
- Authorization. Admin and customer hitting the same endpoint, so changing a URL shows you someone else's booking. Generated code almost never draws this line properly.
- Payment consistency. Paid but no order, or charged twice. Every one of those becomes a human being doing a refund by hand.
- Deployment and handover. Works on my laptop, fails on the server; keys and accounts in the vendor's name instead of yours.
That list is what a five-day build spends its time on. It is also why a five-day quote is not a five-day promise of feature parity with the twelve-week one. Something has to be left out, and the honest version of the conversation is which something.
The five days, hour by hour
Day one, morning: one call, one document. What has to be true for this to be worth launching. We cut the feature list until it fits on one page, and we write down what we are deliberately not building. That document is the contract's scope, so it gets written before any code.
Day one, afternoon: data model and the three screens that carry the core flow. Schema first, screens second, because the other order is how you end up rewriting both.
Day two: the core flow, clickable end to end on a staging URL. Sign up, do the thing, see it saved. From here the client can look at it on their own phone, which ends most scope debates faster than any meeting.
Day three: money and access. Payments, login, admin screen, permission boundaries, and the failure paths — what the user sees when the card is declined, what the admin sees when the webhook never arrives.
Day four: deployment, domain, error tracking, and the boring list. Environment variables in one table, backups scheduled, rollback tested once. Nothing here is impressive and all of it decides whether month four is calm.
Day five: handover. Repository, database, payment account, domain, and every third-party key moved into the client's own accounts, plus a one-page runbook: how to deploy, how to roll back, who to call.
Four of those five days are things the twelve-week quote also contains. The compression is that they happen in sequence by one person instead of in parallel across a team with handoffs.
What five days cannot fix
I would rather say this up front than explain it on day six.
- App Store and Play review. Building takes days. Review queues take days more, and nobody can compress them. Five days gets you a submitted build, not a live app.
- Money that requires a business entity. Payment gateway approval, contract review, bank onboarding. Those run on the provider's calendar.
- Migration from a live system with real users. Moving data that people depend on needs a cutover plan, and that plan is worth more time than the build.
- Decisions nobody will make. If the answer to "should this be one screen or two" takes a week, the project takes a week longer. This is the most common reason a five-day build becomes eight.
What the client actually has to do
About four hours across the five days: two hours on day one to decide scope, one hour on day three to check payments with a real card, one hour on day five to take over the accounts. The rest is waiting, which is the point of hiring someone.
The projects that go wrong are rarely the technically hard ones. They are the ones where the client is unreachable during day one, and three unresolved questions turn into three days of guessing.
Five questions that tell a real five-day quote from a fantasy
- "What are you leaving out?" A quote with no exclusions is not a plan.
- "Where does the code live?" Yours, in your repository, from day one. Not the vendor's, transferred later.
- "Do I get a URL I can open on day three?" If there is no staging link until the end, there is no five-day build, only a five-day claim.
- "Who owns the accounts?" Deploy target, database, payment, domain, keys. All yours.
- "How many revisions, and what counts as one?" Written in the quote, not agreed verbally later.
None of these are about price. They are about whether the timeline is a plan or a number chosen to win the conversation.
About the cost
A scoped first version here runs 3–5 days, and lands around one-twentieth of a typical agency quote for the same scope — roughly ₩1,500,000 for the app example above, about $1,100. The gap is structural: no sales layer, no project manager, no office between the person who quotes and the person who builds. The work is not cheaper; the layers are gone.
Terms are deliberately boring. Quotes and consultations are free. Nothing is paid before a contract, a deposit starts the work, and the balance is due after you have used the build yourself. Contracts are direct with CPC AI Inc., a registered company in Seoul (business number 771-88-03539), with a proper invoice.
If you already got 80% of the way with an AI tool and are stuck, that is the other half of my week: I take the code as it is, review it for free, and finish it through launch.
- Studio (English): https://cpcai-studio.netlify.app/en/
- 한국어: https://cpcai-studio.netlify.app/
- 日本語: https://cpcai-studio.netlify.app/ja/
- Email: [email protected]
Originally published by Dev.to WebDev. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.