The foundation
Why enterprise AI failures are failures of structure, not intelligence — the seams where AI meets your systems, and the handoffs between steps, are where value is won or lost.
Essays
-
The Pilot Graveyard
These pilots didn't die of a technical failure. They passed. They died at the boundary between the lab and the business that was supposed to run on them: a boundary that is organizational, not technical, and that almost no one is staffed to cross.
-
AI Theater
A flawless demo tells you almost nothing. A perfect performance is evidence of perfect staging, not of anything that will hold the first morning a real customer walks on and refuses to read the lines.
-
Fix It Because It's Broken (Not Because It's AI)
Strike the letters "AI" out of the proposal and read what's left. If the case falls apart the moment those two letters are gone, you haven't found a reason to fix your foundation — you've found a reason to distrust the project.
-
Spend It All in One Place
The other eleven requests wait — not because they are bad ideas, but because the people who would do them are the same few now doing the one. Concentration isn't a sequence you'll work through later; it's an honest admission that the scarce thing can only be in one place at once, so you had better choose that place on purpose.
-
Fix One Corner First
The first move was never a strategy offsite or a platform decision. It is a choice, and a narrower one than the size of the problem makes it feel: which single corner — legible problem, measurable win — do you fix first, and let the result fund and direct everything that comes after.
-
GPS with Last Year's Map
A smarter engine reading the same old map only takes the wrong turn with more authority. You don't have an intelligence problem — you have a currency problem, and no amount of model quality closes the gap between what your records say and what is true right now. Only the freshness of the source does.
-
Why All Those AI Upgrades Won't Help You
Every upgrade you buy makes one tool smarter and stops at its edge. But your work doesn't stall inside a tool — it stalls in the handoffs between them, and the handoffs are on no vendor's roadmap, because they belong to no single product.
-
The Conveyor Break
Throughput is a property of the whole line, not the fastest station. You can automate one stretch to brilliance and the work will still stall at the very next break — because a process is only ever as strong as its weakest handoff.
-
Better Boots, Not Smarter Machines
Governance is not about how the AI thinks. It is about what it can commit to — the seams that decide where intelligence is allowed to touch reality, and the laces that decide how much it can promise once it gets there.