Making AI Useful, Not Just Everywhere

The situation

When I joined Wildkind, using LLMs day-to-day inside a business was still new territory. The founder wanted the whole team finding ways to use AI. My job was making sure that happened well — efficient, targeted — rather than bolted onto everything because it was trendy.

The approach

Getting there wasn't a straight line. Plenty of things got tried and dropped — I came at it with an open mind but a critical one, testing tools and workflows rather than assuming a new one would work just because it was new. What stuck: Codex for code review first, then agents that could commit code themselves, then workflows to turn a rough idea into a proper task spec before any of that happened.

Alongside that, I introduced PromptQL and built MCP servers so non-technical staff could ask questions of our own data directly, in plain language, instead of going through the dev team.

Outcome

• Every pull request now goes through the custom reviewer before a human looks at it — from a trial to the default way the team works.
• Shipped prototype and "fun" features that wouldn't have cleared the bar for dev time otherwise — not business-critical enough to justify — but landed well with customers and generated real engagement and feedback.
• Reduced (not eliminated — still refining this) how often staff had to come to tech for basic information; PromptQL and MCP tools now answer a lot of that directly.