A proof of concept proves the technology works, a prototype proves people understand the idea, and an MVP proves someone will use it in their real workflow. The right choice is the cheapest artifact that tests your riskiest assumption.
Founders pay MVP prices for prototype questions every month, and the gap between those two invoices is five figures. The three terms get blurred on agency calls, and every vendor has an incentive to sell the biggest one.
This guide separates the three by the assumption each one kills. You get a comparison table with 2026 numbers, two anonymized project stories from webvise's own delivery log, and a decision rule that takes five minutes to apply.
- A proof of concept answers one question: can this be built at all. It takes hours to a week, and the code is throwaway.
- A clickable prototype answers: do people understand the offer and flow. It takes 2 to 5 days and needs no backend.
- A focused MVP answers: will a real user complete the core workflow with real data. It takes 3 to 5 weeks and costs €5,000 to €25,000 at AI-assisted agency rates.
- Pick the cheapest artifact that tests your riskiest assumption. Buying an MVP to answer a prototype question burns five figures.
- A paying manual process counts as validation. When demand is already proven, skip the prototype and build the MVP in production grade.
What Each Artifact Proves
The three artifacts sit on one axis: how much reality they let in. A proof of concept touches no users. A prototype touches users in interviews but handles no real data. An MVP runs in production with real accounts, real data, and real consequences.
| Artifact | Question it kills | Timeline | Cost (2026) | Who sees it |
|---|---|---|---|---|
| Proof of concept | Can this be built with today's tech? | Hours to 1 week | Internal time, usually under €2,000 | The team, nobody else |
| Clickable prototype | Do people understand the offer and flow? | 2 to 5 days | €1,500 to €5,000 | Interview users, demo audiences, investors |
| Focused MVP | Will one user complete the core workflow with real data? | 3 to 5 weeks | €5,000 to €25,000 | Real users, in production |
The timeline rows match the scope table in how long an MVP takes to build, which breaks the 3 to 5 week window into a week-by-week plan. The cost bands come from webvise's own quoting range for AI-assisted builds.
If the question you need answered is the MVP one, webvise's MVP development service ships a focused first version in 3 to 5 weeks, production-grade from the first commit.
Pick by Your Riskiest Assumption
Write down the one assumption that kills the project if it turns out false. That sentence chooses the artifact for you. It is the same sentence that anchors the MVP requirements document template webvise uses as a learning contract.
- Feasibility risk sounds like: the model can extract line items from scanned invoices at 95% accuracy. Build a proof of concept. A script against 50 sample documents answers it in two days.
- Comprehension risk sounds like: a construction buyer understands outcome-based pricing on first contact. Build a clickable prototype and watch five people react to it.
- Behavior risk sounds like: agents will upload documents weekly without being reminded. Build an MVP, because only production behavior answers a behavior question.
Most projects carry all three risks, and the order matters: kill the cheapest risk first. Spending 5 weeks on an MVP while a feasibility question sits unanswered underneath it means betting the whole build on something a two-day script could have tested.
Mini-Story: The One-Week Prototype That Answered the Question
In June 2026, webvise built a delivery-engine concept for a Berlin team working on AI-native service operations. The idea: service companies sell finished outcomes, agents run the workflows, humans approve the risky transitions, and every run shows its margin. The riskiest assumption was whether operators would understand and believe that operating loop at all. Feasibility could wait.
The artifact was a landing page plus an operator dashboard concept: a workflow library with prices and target margins, a delivery board, run timelines with evidence attached, and a margin view. No production backend, no live agent runs. Delivery took one week.
An MVP for the same idea would have meant agent orchestration, connector integrations, and review gates: months of work and a mid five-figure budget. The one-week concept put the same story in front of operators for a fraction of that, and their reactions now decide what gets built.
Where Founders Overpay
The expensive mistake runs in both directions. Direction one: paying MVP prices for a prototype question. Traditional agencies quote €60,000 to €150,000 for MVPs that spend their first two months in workshops and mockups, which is prototype work at 20 times the price. The tier breakdown in MVP development cost in 2026 shows where those quotes come from.
Direction two: shipping a prototype and calling it an MVP. Vibe-coded builds demo well, then collapse when real accounts, real data, and edge cases arrive. That failure mode has its own post: the vibe-coded MVP tech debt trap.
The test is blunt. If no real user can sign up and complete the core workflow unaided, it is a prototype, whatever the invoice says. If it cannot survive a second user role or a malformed input, it is a prototype with a production URL.
Mini-Story: When Skipping the Prototype Was Right
In February 2026, webvise built a financing-certificate platform for a Berlin real estate service. Buyers use it to prove their creditworthiness to agents and sellers, and the service promises a binding certificate within 24 hours. Demand was already proven the expensive way: the team issued certificates manually, and customers paid for them.
With demand settled, the open risk was operational. Could a 10-step financing form, automated PDF certificate generation, and a comparison across more than 550 partner banks run without the manual bottleneck? Only production answers that, so the engagement skipped straight to a 6-week full-stack MVP.
The platform shipped with an admin dashboard covering the full request lifecycle, a Lighthouse performance score of 96, and page loads under 1.2 seconds. A prototype phase would have delayed the answer and tested nothing the manual process had not already tested.
That is the rule hiding in both stories. A paying manual process is a completed validation phase. An unproven operating story in a new category earns a prototype first.
The Upgrade Path: What Survives Each Step
The three artifacts form a ladder, but the material does not carry up automatically. Knowing what survives keeps you from paying for the same thing twice.
- From proof of concept: the learning survives, the code dies. A PoC script that proved extraction accuracy becomes a requirement line, never a foundation.
- From prototype: the flow and the screens survive into the MVP build, with production wiring added behind them. The prototype's shortcuts stay behind.
- From MVP: everything survives, which is the point. webvise builds MVPs with TypeScript, a real database schema, CI/CD, and monitoring from day one, so the codebase grows into the product instead of getting rewritten.
The reverse path is the trap. Promoting a prototype to production keeps its shortcuts alive under real load, and the cleanup usually costs more than a scoped MVP would have in the first place.
The five-minute version: write your riskiest assumption in one sentence, then buy the cheapest artifact that kills it. webvise scopes exactly that in a short call and builds MVPs and prototypes in 1 to 5 weeks. Send your one-sentence assumption through the contact form and you get back which artifact it needs.