Someone who implements AI for companies said on my podcast that this is upending the nature of what a billable hour looks like in white-collar work. If you're a junior lawyer, bench research and case law that used to take four hours now takes fourteen minutes. How do you log that?
You may not run a law firm. You probably still run on hours.
Hours logged were a proxy. The proxy just failed.
In manufacturing, logistics and engineering & construction, a lot of internal professional work still gets managed like billable time: estimating hours, claims packets, engineering change orders, submittal reviews, contract redlines. Leaders watch utilization. People protect their logged hours because that's how the week gets scored.
When AI collapses the research and first-draft part of those jobs, the clock no longer tells you who did good work. It tells you who typed longer.
The tension: speed vs. the scoreboard
On the one hand, you want the fourteen-minute version of the work. On the other, your systems, reviews and sometimes your customer invoices still assume the four-hour version.
I kind of land in the middle. Keep time tracking where you need cost history. Stop treating hours as proof of value the moment AI can shrink the same task by an order of magnitude. Measure the package that left the building: accuracy, cycle time, rework, customer acceptance.
Where this shows up outside law firms
- Estimating. Takeoff research and comparable-job pulls that used to fill a morning.
- Claims. First-pass narratives and document assembly that used to fill an afternoon.
- Engineering change orders. Precedent search and draft language that used to fill a day.
- Internal "professional services." Anyone whose worth was tallied in hours on a timesheet.
Same pattern. Different job titles.
An example
Say a mid-size E&C firm asks junior staff to log eight hours a day against project codes. An estimator who uses AI finishes a first-pass package in forty minutes that used to take half a day. If the scoreboard is hours, that person looks idle or underutilized. If the scoreboard is complete packages with low rework, they're the model. Keep the old scoreboard and people will slow-walk the tool — or hide how fast they got — because looking busy still pays.
(This post is about the economics of time. The separate problem of rewarding AI winners with more work is a culture issue for another day.)
Signs your time model is already broken
- Utilization looks "healthy" while cycle time on estimates, claims or ECOs doesn't improve.
- People pad hours after AI did the heavy lift, because empty time looks bad.
- You can't explain to a customer why a task that took minutes still billed like hours — or why internal chargebacks still do.
- Managers praise long days and ignore first-pass quality.
What to do this week
Pick one hours-logged workflow: estimating, claims, ECO drafting or contract review. For the next ten packages, record three numbers: time to first draft, time to approved package, and rework after approval. Compare to your old mental model of how long that work "should" take. If the gap is large, rewrite what "full contribution" means for that role before you expand the tool.
Common mistakes
- Keeping hours as the hero metric. The clock stopped being a fair proxy.
- Billing or chargebacks that ignore the collapse. Customers and internal partners will notice.
- Punishing the fast worker. Empty calendar time after a finished package is not laziness if quality held.
- Confusing this with "give them more pie." Fix measurement first. Workload redesign is a different conversation.
Who should own this
Finance and ops own how time and utilization get reported. Function leaders own what "done well" means for each package type. The CEO should ask whether the company is still paying people to look busy when the work no longer takes the hours the spreadsheet expects.
Change the unit before the math lies to you.
Start by measuring draft time, approve time and rework on ten real packages. When four hours become fourteen minutes, the honest move is to stop pretending the hour was the product.