Most teams measure AI by how often it’s used. The ones worth watching are asking a different question: what can we do now that we couldn’t six months ago?

Capability is the new productivity metric

Most organisations are measuring AI by speed. Tasks completed, hours saved, outputs generated. It's a reasonable instinct, but it's tracking the wrong thing.

The real shift is scope. AI is expanding what jobs can include: the analysis that used to need a specialist, the deliverable that lived permanently on someone else's plate, the decision that always needed a senior sign-off first.

Take JPMorgan's LLM Suite, an internal AI platform rolled out to over 250,000 employees. The platform lets analysts produce a full investment banking deck in 30 seconds. Work that previously took hours. That recovered time didn't get filed away as a win. It got reinvested into reasoning and judgment calls that previously sat a level above them.

Faster at the job is one thing. A bigger job is another.

The ceiling has quietly moved

JP Morgan is one data point. The UC Berkeley study zoomed out to ask a harder question: what happens to the room when enough people level up?

Researchers tracked how AI actually changes work at a technology company over eight months. What they found cut against the simple productivity story: AI didn't reduce workload. It expanded scope. Workers proactively took on more tasks and broader variety, even without being told to.

And as they did, expectations quietly recalibrated around them. What was once considered going above and beyond became standard performance. The ceiling rising feels like progress from the inside. But it also means the baseline has shifted for everyone in the room, including those who haven't caught up yet.

What looks like a productivity win at the individual level becomes a new performance floor at the team level.

What you should be tracking

The capability shift doesn’t happen invisibly. Here’s where to start.  

Audit what you escalate. Map the tasks your team regularly hands upward or outward. Those are the first candidates for AI to absorb — not to automate them away, but to give your team the capability to handle them directly.

Assign one stretch deliverable per cycle. Pick one output per month that previously required a different function, a senior sign-off, or a specialist. Use AI to attempt it. Treat the first version as a capability test, not a final product.

Track what's new, not what's faster. Keep a running log of work your team produced this month that didn't exist in their remit three months ago. That delta is your real productivity signal.

Capability expansion is what’s really changing. Speed is just the side effect.