Deterministic vs fuzzy engineering
Deterministic vs fuzzy
Traditional software is deterministic: same input, same output, every time. AI software is fuzzy: the same question can produce different answers. This is both its strength (flexibility) and its weakness (unpredictability). The best systems combine both.
Think like a manager
Traditional software is structured like microservices — data engineers, UI, business logic. Agentic AI is structured like delegating to a team: a graphic designer, a marketing manager, a data analyst. You design roles and handoffs, not code modules.
The McKinsey credit memo case
McKinsey studied a financial institution where credit risk memos took 1–4 weeks. By redesigning the workflow with AI agents that gathered data, subdivided tasks, and drafted the memo, the process was cut by 20–60%. The human stakeholders stayed the same — only the process changed.
Know when to be deterministic
Not everything should be fuzzy. Multiple-choice scoring should be deterministic. Voice assessments might be fuzzy but need a human-in-the-loop appeal system. The skill is deciding where AI flexibility adds value and where predictability is non-negotiable.
In your work, which tasks need a perfectly predictable answer every time, and which could benefit from a flexible AI-generated response?