Most failed agentic work fails before any agent is invoked, because the problem was framed wrongly or not at all. The agent receives an underspecified instruction, makes confident assumptions about what was meant, and builds something coherent that misses what was actually wanted. Problem framing is the first and most consequential human contribution to any agentic workflow.

Engineering and design literature positions problem framing as a respected discipline with its own rigour. Polya’s How to Solve It establishes the primacy of understanding the problem before attempting a solution. Donald Schön shows that expert practitioners reframe problems continuously as they engage with them. Frederick Brooks, in The Design of Design, supplies the empirical case: his own house-design log shows that requirements emerge from attempted solutions rather than from prior reflection. Solving the wrong solution is how the right problem is discovered.

Above the surface question of “what features should we build” sits the deeper question of primary usefulness. What is this thing fundamentally for? Who needs it to exist? An agent can produce features without limit; only the human can answer whether the project should exist at all.

Drawing on Brooks’s chapter on user models, “better wrong than vague” is a practical discipline. Every designer carries an implicit user model. Making the model explicit, even by guessing, allows it to be debated and corrected. The vague ASE practitioner ends up letting the model design for an imaginary user.

Constraints are shaping material. Constraints shrink the search space and make design easier, not harder. Telling the agent what properties are needed produces a better result than telling it how to achieve them. Brooks’s four-way taxonomy of constraints (real, obsolete, misperceived, intentional) is the diagnostic to apply to any framing.

The deliverables of good framing are first drafts in pencil rather than finished artefacts. A problem statement describes the situation rather than the solution. The stakeholder list names who needs the project to exist. The definition of done is specific enough to be testable. The out-of-scope list passes the test “why might a reasonable person have expected this to be in scope?” A practical procedure delegates the framing interview to the agent itself: the practitioner writes a short project sketch, then directs the agent to ask one focused question at a time across six clusters (purpose, user, success, boundaries, constraints, testability).