The pattern
Teams get access to AI and the first question is always the same:
"We have GPT-4/Claude, what can we build?"
That question is the problem.
A poorly defined problem fed into an AI produces a well-crafted wrong answer — at scale, in minutes. That's not progress. That's acceleration in the wrong direction.
The methodology
I built Problem-Driven AI, an open methodology that structures
the work before the build. The core idea: the bottleneck was never execution. It was always understanding the problem deeply enough to deserve a solution.
It's organized in 5 phases with quality gates. You can't move forward until the current phase is complete:
1. Problem Phase
Define and validate the real problem with the people who live it
every day. If the team can't agree on the problem, nothing gets built.
2. Solution Phase
Align product, engineering, and business on a solution that deserves to be built. Not the first idea — the right one.
3. Context Phase
Build the structured knowledge the AI needs to work well. Not
improvised prompts — organized context documents that tell the AI
exactly what it needs to know.
This is where most teams fail. They jump from a vague problem
to a prompt and expect good results.
4. AI Build Phase
Build with prepared context, not against ambiguity. The AI becomes
a collaborator, not a slot machine.
5. Market Phase
Iterate with real usage data. Validate with the market, not with
assumptions.
Why gates matter
Each phase has defined roles, artifacts, and exit criteria. The
friction is intentional. Skipping the Problem Phase to start coding
feels fast. Rebuilding because you solved the wrong problem feels
slow.
What this is not
- Not a prompt engineering guide
- Not an AI tutorial
- Not a product for sale
It's an open, documented methodology for teams building AI products
who want to stop confusing speed with progress.
Try it
The full methodology is open and bilingual (EN/ES):
It covers principles, a step-by-step framework with artifact
templates, anti-pattern detection, a glossary, and a roadmap.
Built with Docusaurus. Everything is documented. No paywalls.
I'd appreciate honest feedback — especially from anyone who's
shipped AI products and dealt with the gap between what the AI
can do and what it should do.
Top comments (0)