Mental model
Begin with a workflow decision, not an impressive model behavior. Define who acts, what outcome improves, which constraints matter, how failure is detected, and why an AI-assisted approach beats the simplest alternative.
Theory
Describe the current workflow, frequency, delay, error, and accountable decision. Form a narrow capability claim such as extracting specified fields with reviewable evidence or proposing a valid next action under policy. Build representative cases before selecting architecture.
Evaluate technical quality together with adoption, latency, privacy, integration, review burden, and recovery. A model can perform a benchmark task yet fail the product because inputs are unavailable, outputs cannot be verified, users do not trust the workflow, or failure arrives too late to correct.
Alternatives and trade-offs
Rules, search, templates, ordinary software, and process redesign may solve the problem more reliably. AI helps when inputs are unstructured, variation is high, and outputs can be evaluated or reviewed. Human-in-the-loop designs reduce automation but can create value earlier and produce better evidence.
Failure modes and misconceptions
A demo is not a workflow result. Starting with a model encourages invented use cases. Broad goals such as productivity are difficult to evaluate. Ignoring the non-AI baseline hides whether complexity is justified. Pilots without a decision threshold become permanent experiments.
Knowledge check
What evidence would show that a capable model still lacks product fit for a particular workflow?
Decision scenario
A legal team tests clause extraction. The pilot measures supported-field accuracy, citation correctness, reviewer time, missed high-risk clauses, document privacy, and recovery. Deployment proceeds only for contract types that meet the predeclared review threshold.
Learning outcomes
- Explain Capability-Problem Fit as a system mechanism rather than a slogan.
- Compare its alternatives, trade-offs, and production failure modes.
- Apply the concept to a decision and identify evidence that would validate it.
Trade-offs
Using Capability-Problem Fit can improve capability or control, but it also introduces cost, latency, complexity, and failure modes that must be measured against an explicit objective.