Case study

Enlyte

Service: AI Development & Integration
Year: 2026
Live: enlyte.ai

01The work

Overview

A decision engine for the calls that are genuinely hard — where the stakes are high, the trade-offs are not obvious, and reasonable people would disagree. You put a question to it, of the shape "should we pivot to enterprise", and it returns a recommended action, the risk that action carries, and the blind spot you had not accounted for, evaluated through thinking patterns modelled on founders and investors. The interface reports how many perspectives it applied, and calls the result a council analysis rather than an answer.

Challenges

An AI that answers strategy questions is easy to build and nearly impossible to trust. Ask a general model whether to pivot and it will produce something articulate, balanced, and useless — advice with no position in it, or a confident position with the counter-argument quietly omitted. The hard part is not generating a recommendation. It is making one worth acting on, which means the system has to be capable of disagreeing with itself in front of the user, and has to show enough of its reasoning that a sceptical operator can decide whether to believe it.

Solution

Fix the shape of the output and make the disagreement structural. Several distinct lenses are run over the same question, and the interface says how many were applied, so a recommendation arrives with its provenance attached rather than as a single voice. The response is always the same three parts — recommended action, risk, blind spot — which means the counter-argument is never something the user has to think to ask for; it is placed next to the recommendation by construction. Naming the method on the site as decision lenses, and giving methodology its own place in the navigation, is part of the same idea: a strategy tool that hides how it reasons has no way to earn trust.

Conclusion

Forcing the risk and the blind spot into every response is what separates this from a chat window with a system prompt. A tool that only ever tells you what to do is one you stop believing the first time it is wrong; a tool that tells you what it might be missing is one you can keep using.

02More work3 projects