Managing (AI) Intelligence: When AI Adds Work, Not Time

WEBVTT
00:00:00.330 --> 00:00:03.947
If your team just got AI tools and you think you're about to get some breathing
00:00:03.947 --> 00:00:05.862
room back, you might be wrong.
00:00:06.340 --> 00:00:11.321
New research on 200 employees at a real tech company found the opposite happened.
00:00:11.808 --> 00:00:16.034
People worked longer, took on more tasks, and burned out faster.
00:00:16.400 --> 00:00:19.425
And most of them did it to themselves without anyone asking.
00:00:19.964 --> 00:00:22.942
So if you're rolling out AI and hoping it lightens the load,
00:00:23.238 --> 00:00:26.543
this episode might change how you manage the next six months.
00:00:34.821 --> 00:00:42.594
Every leader has heard this pitch. AI drafts the email. AI summarizes the report. AI debugs the code.
00:00:43.093 --> 00:00:46.988
So your people get back their time for the work that actually matters.
00:00:47.389 --> 00:00:48.260
Well, here's the problem.
00:00:48.672 --> 00:00:53.658
A researcher at Berkeley just spent eight months inside a real company testing that promise.
00:00:54.030 --> 00:01:00.058
And what she found should worry every manager who thinks rolling out AI is a workload fix.
00:01:01.219 --> 00:01:06.617
So the authors here followed about 200 employees at a U.S. tech company for eight months.
00:01:07.030 --> 00:01:11.424
Engineering, product design, research, operations.
00:01:11.999 --> 00:01:15.818
They weren't watching them from a spreadsheet. They were on site two days a week.
00:01:16.248 --> 00:01:21.165
They ran more than 40 interviews, and they tracked how work actually moved through the company.
00:01:21.612 --> 00:01:25.664
Nobody was forced to use AI. The company just made it available.
00:01:26.453 --> 00:01:31.868
The result wasn't less work. People move faster, yes, but they use that speed
00:01:31.868 --> 00:01:37.028
to pick up more tasks, work more hours, and expand what they were responsible
00:01:37.028 --> 00:01:39.979
for, often without anyone telling them to.
00:01:40.466 --> 00:01:47.223
And that's the headline for managers. Speed doesn't become free time. It becomes more scope.
00:01:48.058 --> 00:01:51.140
The researchers found three specific mechanisms doing this.
00:01:51.576 --> 00:01:55.652
And if you manage people, you've probably already seen all three without naming
00:01:55.652 --> 00:01:58.262
them. Number one is task expansion.
00:01:58.871 --> 00:02:04.816
When AI makes something feel achievable that used to feel out of reach, people just take it on.
00:02:05.263 --> 00:02:09.712
A project that sat on the back burner for months suddenly feels doable in the
00:02:09.712 --> 00:02:11.793
afternoon, so it gets started.
00:02:12.211 --> 00:02:15.415
Multiply that across a team and your headcount hasn't changed,
00:02:15.683 --> 00:02:17.685
but your total workload just did.
00:02:18.425 --> 00:02:23.814
Number two, blurred boundaries. AI lowers the friction of starting something.
00:02:24.168 --> 00:02:28.754
That sounds good until you realize it means work starts leaking into lunch,
00:02:29.004 --> 00:02:35.174
into meetings, into the evening, because firing off a quick AI-assisted task
00:02:35.436 --> 00:02:37.206
feels effortless in the moment.
00:02:37.723 --> 00:02:40.591
People aren't clocking extra hours because you demanded it.
00:02:40.962 --> 00:02:46.233
They're doing it because starting is easy now and stopping never got easier.
00:02:46.893 --> 00:02:52.295
Three, more multitasking, not less. This is the one that surprised me most.
00:02:52.910 --> 00:02:57.112
People weren't doing one thing well with AI's help. They were running multiple
00:02:57.112 --> 00:03:03.220
threads at once, writing code by hand while an AI agent drafts an alternative version,
00:03:03.684 --> 00:03:09.322
running several AI agents in parallel, reviving old, deferred projects because
00:03:09.322 --> 00:03:13.309
now AI can, quote, handle part of it in the background.
00:03:13.826 --> 00:03:15.288
It feels like momentum.
00:03:15.822 --> 00:03:21.099
It's actually constant context switching, constant checking of AI's output,
00:03:21.372 --> 00:03:23.874
and a growing pile of half-finished things.
00:03:24.600 --> 00:03:27.589
Put those three together and you get a self-reinforcing loop.
00:03:28.129 --> 00:03:34.222
AI speeds up the work, speed raises what's expected, expectations pull in more
00:03:34.222 --> 00:03:38.220
work, and the cycle repeats. Well, until it doesn't.
00:03:38.963 --> 00:03:43.508
The researchers found this pattern is unsustainable. Cognitive fatigue builds,
00:03:43.752 --> 00:03:48.922
decision quality drops, and the early productivity win gets erased by burnout,
00:03:49.206 --> 00:03:51.145
mistakes, and turnover.
00:03:51.668 --> 00:03:53.862
And here's the part that really should get your attention.
00:03:54.402 --> 00:03:58.521
In follow-up reporting on the same research, the people absorbing most of the
00:03:58.521 --> 00:04:00.329
strain weren't senior leaders.
00:04:01.421 --> 00:04:06.061
It was associates and entry-level employees, the people with the least power
00:04:06.061 --> 00:04:07.985
to push back on a growing workload.
00:04:08.560 --> 00:04:11.683
So if you're a manager and you're not seeing this on your own desk,
00:04:11.939 --> 00:04:14.888
that's exactly why you might be missing it on your Teams.
00:04:15.636 --> 00:04:19.636
So the instinct here is to tell your team, hey, use good judgment.
00:04:19.636 --> 00:04:23.247
Don't overdo it. The research says that doesn't work.
00:04:24.416 --> 00:04:28.576
Asking employees to self-regulate against a tool that makes overdoing it feel
00:04:28.576 --> 00:04:34.143
effortless and rewarding is asking them to fight incentives you built into the job.
00:04:34.793 --> 00:04:39.234
Output gets noticed. Burnout doesn't show up until it's expensive.
00:04:39.541 --> 00:04:43.808
So self-regulation isn't the fix. Structure is.
00:04:44.220 --> 00:04:46.843
And so here's what managers should actually do.
00:04:47.522 --> 00:04:52.202
The researchers call this building an AI practice, quote unquote.
00:04:52.689 --> 00:04:59.835
Not a policy document nobody reads, but real norms your team operates by in three practical moves.
00:05:00.172 --> 00:05:02.639
First, build in intentional pauses.
00:05:03.231 --> 00:05:08.716
Don't let speed compound automatically. If AI cuts a task from four hours to
00:05:08.716 --> 00:05:12.630
one, that save time is not automatically more work.
00:05:12.967 --> 00:05:16.531
Say so explicitly. Normalize stopping.
00:05:17.132 --> 00:05:20.284
Second, sequence the work. Don't just accelerate it.
00:05:20.835 --> 00:05:25.906
If AI lets someone finish task one faster, that doesn't mean task two,
00:05:25.906 --> 00:05:29.612
three, and four should all get pulled forward into today.
00:05:30.054 --> 00:05:33.310
Decide deliberately what moves up and what waits.
00:05:33.954 --> 00:05:37.246
Third, add human grounding back into the loop.
00:05:37.832 --> 00:05:43.057
The more parallel threads AI lets someone run, the more you need real check-ins,
00:05:43.248 --> 00:05:49.441
not status updates, actual conversations to catch fatigue before it becomes a resignation letter.
00:05:50.116 --> 00:05:54.063
The teams that win with AI aren't the ones who adopt it fastest.
00:05:54.435 --> 00:05:55.703
They're the ones who manage the
00:05:55.703 --> 00:06:01.848
capacity it frees up on purpose instead of letting it default to do more.
00:06:02.465 --> 00:06:07.968
So here's the main takeaway. AI doesn't create free time. It creates capacity.
00:06:08.281 --> 00:06:15.526
And capacity gets filled one small, quote, just one more thing decision at a time.
00:06:15.996 --> 00:06:21.395
Unless someone in charge decides otherwise. and that someone has to be you.
00:06:22.057 --> 00:06:26.904
So if you're rolling out AI tools this quarter, don't just measure whether people adopted them.
00:06:27.269 --> 00:06:31.136
Measure whether their hours, their scope, and their stress went up.
00:06:31.455 --> 00:06:37.029
Because right now, for a lot of teams, the answer is yes, and nobody planned it that way.
00:06:37.824 --> 00:06:41.406
Well, that's it for today's Manager Lab. If this hit close to home,
00:06:41.748 --> 00:06:46.423
send it to a manager on your team who just rolled out a new AI tool with a straight
00:06:46.423 --> 00:06:50.415
face and said, hey, this should save everyone time.
00:06:51.762 --> 00:06:56.023
Well, until next time we meet in the manager lab, do good work.