Why most AI pilots fail
Pilots don't fail on the model. They fail because nobody mapped where it was supposed to go. Five recurring causes, and what to do about each of them.
Learn
Blog
How an AI idea becomes a system that actually runs in production: process discovery, evals, exception handling, ROI and deployment in real enterprise environments.
- For leaders
- Discovery
- Evals
- Deployment
- Fundamentals
- Career
The documented process and the real one
Every company has two processes: the one written down, and the one that runs. AI deployments build on the first — here's how to capture the second.
How an AI discovery actually works, step by step
What an FDE does during the two to four weeks of discovery, who they talk to, what they pull from your systems, and exactly what you get at the end.
The ROI matrix: what's worth automating, and what isn't
Not every process deserves attention. How we rank workflows on volume, recoverable time, risk and feasibility — and why the first project isn't the biggest one.
Evals: how to measure whether it actually works
“Looks good” isn't a result. How a non-deterministic system becomes a measurable number, and how you find out where the errors really come from.
Building a golden dataset in practice
How to assemble a measurement set from your own historical cases that genuinely tells you whether the system works — and the mistakes to avoid.
How to evaluate a task with no single right answer
A summary or a deck can be done well a thousand ways. It's still measurable: break it into criteria and build in human feedback from the start.
Human in the loop: where approval should stay
Human-in-the-loop isn't a compromise, it's a design decision. Where to keep a person in the process, and how to stop approval degrading into rubber-stamping.
Unhappy paths: why everything rides on exception handling
A process goes right one way and wrong several hundred ways. Build only for the happy path and you've built a demo, not a system.
If you can't see what it did, you won't trust it
An audit trail isn't a nice-to-have, it's a precondition for going live. What to log in an agent, in what form, and what people actually do with it.
Don't replace your ERP. Build on it.
A company that spent two years migrating doesn't want another migration. Why building on existing systems is the only viable route, and how we do it.
Shadow mode: going live without anyone carrying the risk
The system runs but doesn't decide. Why this is the most useful six weeks of a deployment, what to measure during it, and when you can move on.
Which model should you use? Wrong question.
Model choice is a measurement, not a belief. How to build so the model stays swappable, and when switching is actually worth the effort.
Token maxing: how to burn a year's budget in three months
Handing everything to the model is more expensive, slower and less reliable than code would have been. Where the money goes, and what to do instead.
How to measure AI ROI so finance accepts it
Three groups of metrics matter: cost savings, risk reduction and revenue impact. How to turn each into defensible numbers without cheating.
A 30-day plan to become a forward deployed engineer
Four weeks, week by week: build a real agent, make it survive reality, measure it, and learn to defend it in business language. With concrete checkpoints.
When not to use AI — six cases from practice
One of the most valuable sentences an FDE can say is “no model belongs here”. Six situations where AI is a worse answer than a rule or a better form.
The two sides an FDE has to master
An FDE isn't the average of a consultant and a developer — it's the best of both. What that means in skills, and how to build the weaker side.
Prefer a reader? RSS feed.