Every fixed price written before discovery is a guess. Both sides pay for the guess: the client in change requests, the builder in margin. Here is what we do instead, and what you own at the end of it even if you never sign the build.
Someone asks what their platform will cost. We say we do not know yet, and that we will not pretend to. That answer loses us deals, and we keep giving it. The Phase 0 Rule is defined as a refusal to price a software build until a short, separately priced discovery and architecture engagement has produced the architecture decision records, the data model, a working software factory, a tested answer to each risky unknown, and a priced milestone plan. Phase 0 runs two to four weeks at a fixed price in our AI Product Studio, and the build that follows is quoted per milestone once we know what we are actually building. After 60+ products shipped since 2013 and 90+ tech due diligences spent reading other people's unfinished projects, we have never seen a build go wrong for a reason that Phase 0 could not have surfaced in week two.
Here is the uncomfortable mechanic. A buyer asks for a number before anyone has looked. The builder has three options and all of them are bad.
Quote low and win the deal. The unknowns arrive in month four, the change requests start, and the relationship becomes a negotiation about what the word “included” meant in a document written by people who had not yet read the legacy database.
Quote high and pad it. The buyer pays for the builder's uncertainty. Half the padding was never needed, and the buyer has no way to tell which half.
Quote honestly, with a range wide enough to be true. The range is so wide it reads as incompetence, and the deal goes to whoever quoted low.
A fixed price for undiscovered work is not a commitment. It is a bet, written in the language of a commitment, and the client is on the other side of it.
An agency will typically quote six figures and the better part of a year for a serious platform, built on exactly this kind of guess, with the discovery phase bundled inside the number it is supposed to validate. That bundling is the problem. Discovery cannot test the estimate if it has already been priced into it.
Phase 0 is sold on its own, at a fixed price, with its own deliverables. The build is quoted afterwards and paid per milestone. Two contracts, two decisions, two chances for you to stop.
This is not a softer commitment. It is a harder one, moved to the point where it can be kept. We will hold a milestone price we wrote after seeing the data model. We will not hold a milestone price we wrote after a discovery call.
Five deliverables. All of them are artefacts, not opinions, and all of them land in your repository as they are produced.
Every consequential technical choice written down with the options we rejected and why. The record is what keeps the build consistent once agents are writing most of the code.
Entities, relationships, lifecycle, ownership. The data model is where scope hides. Most change requests we see in other people's projects are data model discoveries arriving nine months late.
The repo, CI/CD, environments, the product brain, and the agents configured against your stack. Not a plan to set it up. Set up, in your repo, with a pull request already merged through it.
The three or four things that could blow the estimate: a third-party API that does not behave as documented, a migration from a system nobody has read, a latency budget, a compliance constraint. We prototype them in Phase 0, while they are still cheap.
Milestones with acceptance criteria, sequence, and a price per milestone. This is the deliverable the other four exist to make possible.
Note what is not on the list. No slide deck. No wireframe pack that nobody opens after week three. No estimate spreadsheet with a confidence column. Phase 0 ends with a running pipeline, a merged pull request and a price.
The obvious objection: if agents write the code, why spend weeks on architecture? Just build it and iterate.
We have built 5+ AI-native products and teams this way, and the answer is the reverse of what people expect. Slow, expensive execution used to be an accidental safety mechanism. When a feature took three developers two weeks, somebody noticed in week one that the requirement made no sense. The cost of typing the code was the feedback loop.
An agent will build the wrong thing very fast, very consistently, and with excellent test coverage. Cheap execution does not forgive a bad specification. It industrialises it.
And the failure is harder to spot than human failure. A confused developer produces visibly confused code: inconsistent naming, dead branches, a comment that says “not sure about this.” An agent working from a wrong but internally coherent spec produces clean, conventional, well-tested code that implements the wrong model of your business in forty files. Nothing looks broken. Everything is.
This is why the architecture decision records and the data model come before the build rather than during it. They are the specification the agents work from, and they are what the product brain holds so that the hundredth ticket is built on the same assumptions as the first. The quality controls we run on top of that, from test-driven development to senior review of every merge, are covered in how we ensure quality with AI agents. They catch bad code. They cannot catch a good implementation of the wrong thing. Only discovery does that.
The same logic shows up on the other side of our work. In due diligence, the fastest-moving engineering teams we score are not the ones with the most agents. They are the ones whose agents are pointed at a written, agreed model of the domain. That gap is the practical difference between the maturity levels we describe in the 4 levels of AI maturity.
Phase 0 is designed so that leaving is a real option. Everything it produces is in your repository, your cloud accounts, your AI subscription. If you read the milestone plan and decide not to continue, you keep:
That last one is portable in a way clients do not always expect. Take it to another builder and you will get a better quote than you would have got from a blank brief, because the unknowns that inflate quotes have been removed. Take it to your own team and they can start. Take it to an investor and it is a credible technical plan rather than a roadmap slide.
If Phase 0 tells you not to build this, Phase 0 has done its job and saved you the expensive version of the same conclusion.
We would rather say this out loud than have you work it out. Phase 0 is also how we stay solvent.
A builder who quotes blind and is wrong has two ways out. Absorb the overrun, which destroys the margin on a project the team is now resentful about. Or defend the scope line by line, which destroys the relationship. We have watched both happen from the outside in due diligence, repeatedly, and the damage is never confined to the budget. It shows up as a codebase that was built to a contract instead of to a design.
Charging for discovery removes the incentive to be optimistic. We are not trying to win a build by underpricing the part we cannot see. And because Phase 0 is paid, we staff it with the senior architect who will own the build, not with a salesperson producing a free proposal.
Our model is cash, by the way, and the first phase is paid. We do not build products for equity. Aligned incentives are better than entangled ones. Scale-ups carrying the results of a few blind quotes already, usually around Series B, will recognise the pattern we described in the Series B tech foundation.
Phase 0 is for building a product. It is the wrong instrument for three other situations we see often.
You already have a team and you want it to build this way. That is an AI Factory Assessment at EUR 4-6K, and then an AI Transformation Sprint from EUR 26K if the assessment says the gap is worth closing.
You are about to invest in or acquire a company and you want to know what the engineering is really worth. That is tech due diligence, from EUR 8,000, with a verdict in five to seven days.
You do not yet know whether the product should exist. Then the honest answer is that no amount of architecture helps, and the money belongs in customer conversations first. The methodology we use to get from an idea to something worth architecting is in the AI-native startup playbook.
Phase 0 is a short, separately priced discovery and architecture engagement that runs before any build is quoted. Two to four weeks. It produces architecture decision records, a data model, a working software factory, tested answers to the risky unknowns, and a priced milestone plan. It is a fixed price, and the build that follows is paid per milestone.
Because a fixed price written before discovery is a guess, and both sides pay for the guess. You pay in change requests and in an argument about what “included” meant. We pay in margin, or we protect the margin by having that argument. We would rather sell a small piece of honest work that makes the big number real.
The opposite. Agents removed the cost of typing code, and that cost used to be the feedback loop that caught a bad spec in week one. An agent will build the wrong thing very fast and very consistently, in clean well-tested code that looks entirely correct. Cheap execution raises the value of a correct specification.
All of it, in your own repository and accounts: the architecture decision records, the data model, the configured agents and product brain, the pipeline and environments, the prototypes, and the priced plan. The plan is portable. Another builder, your own team or an investor can all act on it.
No. A discovery phase produces documents. Phase 0 produces a running system plus the documents: the repo exists, the pipeline is green, the agents are configured against your stack and a change has already been reviewed and merged. The build does not restart from zero afterwards, which is why the milestone prices hold.
Above The Clouds runs the AI Product Studio and AI Transformation Sprint. If someone has handed you a fixed price for a platform nobody has architected yet, the number is not the risk. The silence behind it is. Get in touch to discuss your company or portfolio.