The craft didn’t disappear. It moved.
An open manifesto, taxonomy and rubric for building software with AI, and for evaluating the people who do it.
2 people have signed
AI changed how software gets written. Technical interviews did not keep up. Some companies ban AI in interviews, others allow it with no criteria, and candidates rarely know what is being evaluated. Memorized puzzles measure less than ever, and nobody agrees on what should replace them.
We are engineers and people who hire engineers. We believe the craft did not disappear. It moved. It now lives in how we frame a problem, how we choose to direct the model, how we verify what comes back, and how we decide what ships. That can be named, taught and evaluated fairly.
We value
- Judgmentovervolume of generated code
- Verificationoverblind trust in the output
- Intentional choice of styleovera single prompting habit
- Contextoverclever phrasing
- Open use of AIoverpretending it is not there
- Real problemsovermemorized puzzles
The items on the right still matter. The items on the left matter more.
Principles
Evaluate the process, not only the result.
In an AI-assisted interview, the final code is the least informative artifact. Watch how the candidate frames the problem, which style they pick, what they reject and why.
Name the styles.
Declarative, imperative, spec-driven, test-driven, example-based, iterative, context engineering, vibe coding. A shared vocabulary lets candidates prepare and lets interviewers compare notes.
No style is wrong. Choosing blindly is.
Good engineers move along the spectrum according to risk, complexity and how well they know the codebase. Vibe coding a prototype is fine. Vibe coding a payment flow is not.
Verification is part of the job.
Reading, testing and questioning generated code is engineering, not overhead. Code the candidate cannot explain is not their code yet.
Be transparent on both sides.
Companies say up front which AI tools are allowed, what is being evaluated and with which rubric. Candidates use AI in the open and explain their decisions.
Keep it open.
The taxonomy and the rubric are public, versioned and free to adopt, adapt and criticize. Anyone can propose a change with a pull request.
Signatures are personal. Company names appear for identification only and do not mean the employer endorses this manifesto.
8 prompt styles, one shared vocabulary
A shared vocabulary for the ways engineers direct AI when they write software. Each style has a definition, an example prompt, what it reveals about the person using it, when it fits, what to watch out for, and where the term came from.
Explore the stylesA rubric you can use tomorrow
4 levels, 5 dimensions and a 65-minute interview script. Built to be shared with candidates before the interview, not hidden from them.
See the rubric- 1
Operator
Uses a single style and accepts the output without real review.
- 2
Practitioner
Switches between describing and dictating, reviews the code and fixes errors.
- 3
Engineer
Chooses the style by risk and complexity, uses tests and specs as contracts, manages context.
- 4
Architect
Sets standards for the team, knows when not to use AI, weighs cost, security and maintenance.
Sign the manifesto
Signing takes a minute. You open a short form on GitHub while logged in, a bot checks it and adds your name. Every signature is tied to a real GitHub account and kept in the git history.
2 people have signed