When AI misreads what someone meant, it rarely fails right then. It fails later, and bigger. The misread becomes a plan, the plan becomes work, the work passes its own tests, and the next person builds on it.
Better models still misunderstand. This is about how the work runs, and that’s where I work.
Four shifts that keep AI on what people mean
Understanding before action
The AI says what it understood before it acts. A misread shows up in one sentence, not a week later.
Care that fits the risk
A typo moves fast. A change to what customers pay gets every check.
Done means proven
Work is finished when the checks that matter have run and passed, not when it looks finished.
Corrections carry forward
A correction that keeps coming up becomes standing guidance. Nobody has to say it again.
How I work
I don’t promise a number up front. We start by measuring where drift actually gets in for your team, then change what matters most, with one team first.
When I audited my own system, I found checks that looked right and did nothing. I rebuilt them. Most teams have a few of these, and finding them is usually where the work starts.
Engagements range from a two-week assessment to working alongside your team as the practice grows.
Selected work
A live AI coach, built by breaking down what a great coach actually does
In production with paying members. Over 14,000 real coaching conversations so far.
ixcoach.com →Checks at each moment where AI agent work drifts
In one real session, before changing prices, the agent stopped to ask whether existing customers should get the new price too.
See the harness →A new session starts knowing what was already decided
One search across past decisions, research and intent. Nobody re-explains the same context twice.
How it works →925 coaching concepts in 7 languages, free to read
Google search clicks grew from 306 to 3,347 in a month (September 2026).
Browse it →Where drift gets in
A one-page field guide to the seven moments in an AI working session where misunderstanding slips in, with the question to ask at each. It’s free to use with your team.
Open the field guide →Selected writing
Six agent alignment patterns the industry hasn’t published
What running AI agents in production taught me that the published frameworks don’t cover yet.
Don’t replace the gut. Calibrate it.
Why I score how accurate intuition is, instead of A/B testing everything.
From human development to production AI
Why the point of AI isn’t answering questions. It’s developing the person asking them.
Background
I came to AI from human development. For twenty years I’ve studied and practiced how people actually grow, including graduate study in applied integral theory at JFK University and training in Integral Circling.
In 2023 I founded Next AI Labs and built IX Coach on one conviction: AI should develop the person asking the question, not just answer it. I built most of it with AI agents, which taught me exactly where they go wrong.