An ADHD coach who used to build software

Kat, an ADHD coach, spent years as a software engineer at a large tech company before being diagnosed with ADHD later in her career. She eventually left to build her own coaching practice, working with technically minded people and business owners on executive-function-friendly systems. The framework she teaches for working with AI comes out of that same practical, systems-first instinct.

Genius, but secretly — and only starting from zero

Her framing: AI is like a new intern who happens to be secretly a genius. It knows an enormous amount — effectively the entire internet — but won't volunteer any of it unless asked directly, and shows up with zero context about you specifically, every single time a new conversation starts.

Every new chat is still "day one"

Unless context is deliberately saved or connected, an AI tool has no memory of preferences from a previous conversation — a new chat in Claude or ChatGPT genuinely starts from nothing. That shifts the burden onto giving it the equivalent of onboarding documentation up front: who you are, what you're trying to achieve, and what a good outcome actually looks like.

Why connecting tools changes what the intern can even do

The metaphor extends naturally to connected tools: without them, the intern is like a personal chef who needs every ingredient handed over by hand, every time. MCP and connectors hand that chef the keys to the pantry instead — with Claude connected to xTiles, there's no need to re-explain a whole week from scratch, because the calendar, notes, and tasks are already reachable directly.

A practical technique: the "sandwich" approach

The method she recommends has three layers: brain dump everything raw and unfiltered, let the AI take a first pass at organizing it, and explicitly ask it to request clarification instead of guessing — with final review and decision-making staying on you. That last step matters most: without it, it's easy to fall into a loop of asking AI to just decide for you, second-guessing the answer, and asking again.

What this looks like for a weekly plan

In practice, that's one plain-language request — something like "help me plan my week" — pointed at Claude connected to xTiles, following the same three-part shape: raw context in, a first structured pass back, questions before anything gets locked in. No special setup is required beyond the connection itself; the framework is really about how you talk to it once it's there.