MING Labs FieldNoteby MING Labs
Past Releases Subscribe minglabs.com

NO. 01FieldNote

Everybody got faster with a copilot. But nothing got quicker.

2 September 2026

Hi there,

Everyone I talk to has rolled out a copilot by now, and everyone reports the same thing: people are visibly faster, and nothing ships any sooner. We ran the same experiment on ourselves. Individual output doubled in a month. End to end, delivery moved ten percent.

Where did the rest go? Into the handovers.

A tool knows one person. The context, what the client wants, why the last version came back, lives in that person's head. It does not transfer. The next person spends a morning rebuilding it, and that morning shows up in nobody's numbers.

Everyone's stretch got faster. Nobody owns the gaps between the stretches. And the gaps are where the time went.

Tools raise personal productivity. Agents raise organisational performance.
WHAT WE DID INSTEAD

We kept the copilots. They save real time. What we changed is that we stopped handing out tools and started handing over jobs.

We hired our first agent in February. There are fifteen now. Each one has a job description on file, a name the team uses in conversation, a place on the org chart, and a person who answers for what it produces.

Hybrid organisation
People and agents on one org chart, on the same terms. Mostly it is a staffing decision.
Six things, and you can write all of them before a line of code.
What makes it work
1.Job description. What it owns end to end, and what it may decide alone.
2.Skills. The written procedures it follows.
3.Tools and data. Scoped access to what it needs, never to all of it.
What makes it responsible
4.A hiring manager. One named human who answers for its output.
5.KPIs and review. Measured like a hire, with weekly debriefs while it is onboarding.
6.An escalation path. The cases it must hand back to a human, written down before it goes live.
The rule: the first three are the ones everybody does. The second three are the ones that decide whether it lasts. Miss any of the six and what you have is still a tool.

None of that is engineering. It is writing down what lives in people's heads.

One of ours runs the order inbox at a distributor client: two to three hundred orders a day, 60,000 SKUs, every format including handwritten scans. Three people used to do that job. Now one handles the exceptions and the agent does the rest, at €0.015 an order, all in. Nobody there got faster. The job got quicker, and somebody owns it.

One number worth knowing
Across roughly 300 enterprise deployments, 95% of AI pilots produced no measurable effect on profit or loss (MIT Project NANDA, 2025). Almost everything runs. Almost nothing reaches the accounts.
Free download
See what an org chart looks like with agents on it.
What a hybrid organisation actually is, one real agent’s job description, and the six boxes you fill in before anybody writes code.
5-minute read.
Download the guide →

What we learn running agents. Sent to you.

Lessons from inside real agent rollouts — ours and our clients’ — with the wins, the failures, and what they actually cost. Written by the people doing the work, readable in minutes, useful the same day.

Just your email — nothing else. Unsubscribe in one click, any time.

Browse past Releases →
FieldNote Every two weeks

Fresh from the field. We put AI agents to work inside real companies every day — hundreds of projects so far. You get what worked, what failed, and the playbooks we built along the way — so you learn from our mistakes instead of making your own.

The Benchmark Quarterly

We read every major piece of AI-agent research, then hand you the few numbers that actually matter. Your next decision on agents — backed by data, not hype.

Best Practices Coming soon

The most impressive AI agents running inside major corporations right now — shown as real demos, videos, and snapshots, each with the one insight worth copying. Twice a year, and worth the wait.