Every few months someone declares that management is dead. Big Tech is in the middle of a Great Flattening, and there are companies that have a single Director managing 80 engineers and a fleet of agents with nobody in between.
I think it’s the opposite. Management is becoming more important as AI changes how businesses work, and the job itself is shifting faster than most managers have noticed. Let’s talk about what’s changing and how to deal with it.
Changes
Here are the biggest themes of what is going on:
- AI is both making people less collaborative, and driving more need for collaboration
- AI is making the focus on ROI and actual business outcomes stronger than ever
- AI is confusing managers about how to think about performance management
-
AI is requiring people to set new standards and culture
Collaboration
Two things are going on in the collaboration world.
First, people are becoming less collaborative. AI has this magical way of convincing everyone that they are auteurs who just need to be unblocked so they can realize their vision. Less sardonically - people spend all day collaborating with an LLM, and it’s making them less collaborative with humans; the sycophantic, optimistic nature of LLMs are megalomania fuel that even the most even keeled people are susceptible to.
Second, the need for collaboration is rising. Producing stuff got cheap. Reviewing, integrating and coordinating that stuff did not. A recent Harvard study of ~700 companies found that AI coding agents pushed pull request volume up 23%, and the time to get a PR reviewed and merged up 49%.
This turns your company into a factory where every station is running at full output and nothing ships. Everyone does their job, and nobody else is doing theirs. This leads to the “cleaning up the Airbnb after a party” effect, where each person individually feels like they’re doing more than their fair share, while everybody else isn’t pulling their weight.
The answer here is unsexy - you must have crisp, well-defined expectations of collaboration. That means things like target SLAs on response times for high volume interaction modes (like pull request reviews). Measure it, staff it, and make it somebody’s problem when it’s slow.
ROI
In the prior era, people moved so slow that you could course correct investments in real time. With AI, if you don’t know what your goal is in advance, you might spend $100k in tokens before realizing nobody asked for, or wants, this feature.
The answer is again unsexy - make sure everything has a clear line of sight to business value. Metrics should be coming back in a big way, after years of people cynically citing Goodhart’s law as a reason not to have any.
The catch is you have to measure outcomes and not activity. Amazon built an internal leaderboard ranking engineers by AI usage, and people promptly started pointing agents at pointless tasks to climb it. Goodhart would be proud. Measure things as viciously close to actual things that get reported on a 10-K as possible.
Performance Management
Managers are confused about how to manage performance in the age of AI. Should the bar be raised? Did you do it or did the AI do it? Etc.
The short playbook is:
- The bar must be raised, everywhere
- Who cares if it was the AI or not as long as the outcome was there
- You own what you ship, no matter who or what wrote it
- The ROI needs to be there, see the section above
What does a raised bar actually look like? For a senior engineer, it means the project that used to take three people now takes one person with a clear spec. It means “I’m blocked on the frontend team” stops being an acceptable answer for a small change you could make yourself. It means the design doc shows up with a working prototype attached.
Ownership cuts both ways. If you shipped it, it’s yours. “The AI wrote that part” is not an excuse for a bug, the same way “the intern wrote that part” never was.
People have gotten a lot of credit over the last stretch just for doing more stuff with AI. Before 2023, there was a strong correlation between volume of output and being good at your job. AI has completely shattered that correlation; it’s probably now close to 0. It’s time to rebaseline and be clear that more and higher quality work is expected, and set expectations that the baseline will rise with every model improvement.
Standards / Culture
Can docs be written with AI? What counts as slop?
New cultural standards are needed in the AI era, and there is strong debate on what is right and what is wrong. If you want one rule to follow it’s this - ignore the genesis of output, criticize the content of output.
“You used AI” is a tone deaf and fast dying criticism. Set a culture where criticizing outputs on their own merits, not on their origin, is the norm. This will help people actually understand and articulate what good/bad looks like, and also avoid an anti-AI bias that is increasingly ludicrous.
Wrapping up
AI multiplied everyone’s output. Management is the job of pointing that output at something that matters. None of the fixes here are exciting. SLAs, metrics, a higher bar and clear standards are mostly what good management always was. The difference now is that when you skip them, you find out faster, and it costs a lot more.