When should you use AI?

Use AI when the task needs interpretation

AI is compelling where language, interpretation, synthesis, pattern recognition or uncertain reasoning are intrinsic to the task: reading free text, summarising evidence, drafting, classifying messy inputs or choosing among allowed actions.

Use deterministic software when the rule can be written down

Eligibility checks, permissions, calculations, policy thresholds and most routing decisions belong in deterministic logic. Code gives the same answer every time, costs less to run and is simple to test.

Make errors detectable before you add a model

Where a wrong output is consequential, the product needs a way for the system or the user to catch it: constraints, verification steps, confidence signals or human review. Design those first, then decide how much the model should do.

Plan the evaluation up front

Define representative tasks, failure modes and acceptable thresholds before building. With an evaluation plan in place, the team can see whether each change improves the product.

Check for a simpler fix

Many “AI problems” turn out to be information architecture, workflow or integration problems. Better defaults, search, forms, rules or data access often produce a more reliable result at lower cost.

The strongest products combine both

Place AI capabilities inside deterministic boundaries. Code owns the workflow, the permissions and the safety gates; the model handles the parts that need judgement.