This week, I gave a talk at LDX3 New York. When they asked me to speak and asked me for a topic, I said that it was probably time I finally give a talk on what AI means for management. Of course, this was back in April, when I figured surely I’d have it all figured out by September. Then the conference organizers did me an unexpected favor: they wrote the title of the talk, forcing me into a framing that narrowed the scope enough to write something coherent. Below is not exactly a transcript of what I said.
Press enter or click to view image in full size
Framing: A lot of people want management dead
The talk started with 3 quotes from tech CEOs that have proclaimed the massive reduction and likely end of “management.”
“There is no need for a permanent middle management layer.” — Jack Dorsey
“I don’t think people managers will have any value in the future.” — Brian Cheskey
“No pure managers: Every leader at Coinbase must also be a strong and active individual contributor.” — Brian Armstrong
To me, these are ideas from sci-fi futures. They don’t represent a realistic understanding of the world we live in today. There are many sci-fi futures people are thinking about: massive reduction of people employed by the tech industry, the death of knowledge work overall, cybernetic AI panopticon overlords, post-scarcity, data centers in spaaaccee, and of course, AI literally and completely replacing the need for any human managers.
I can’t write a talk about sci-fi futures, because I’m not a futurist, and there is no manager’s path if management is dead anyway, so instead, I will focus on the actual work of 2026.
The Manager’s Path in 2026
I don’t think I’ve ever given a talk that is just about “The Manager’s Path”, partly because there’s a lot of stuff in the book, and partly because, well, I started a new job right after it was published and stopped speaking for a bit. This talk forced me to boil the book down to a few recurring themes of focus areas for managers; namely, Technical Skills, Signal Processing, and People Skills. What are these things and how have they started changing with the current state of AI?
Technical Skills
Why do we care about technical skills in management? Because a lot of what we call “management” is in fact a risk management function. The manager is responsible for, among other things, making sure the work gets done, at reasonable quality, in reasonable timeframe, without causing so much personnel strain that people constantly quit. When building software systems, it’s hard to really know how long something should take, and what good looks like, if you don’t have any experience building this sort of system yourself. It’s not impossible, sure, but it’s hard. It’s easier for someone with background to manage a lot of the risks than it is for someone who has to delegate that judgement entirely.
But of course the act of creating software has changed a lot, very quickly. Nowadays engineers increasingly direct agents to produce the software rather than directly writing it line by line themselves. And the speed at which you can end up with a seemingly-complete piece of software is consequently significantly increased. This has led to a lot of weird outcomes that we’re trying to manage:
- People are really obsessed with productivity, and not just CEOs/Execs; individual engineers are also really addicted to the new speed and impatient to make it work for them
- The amount one person can do by themselves is a lot higher, so we have to completely revisit our intuitions about how to break up work to account for this increased capacity per individual
- Other risks are looming thanks to this, some related to speed, others related to the technical capabilities of AI: security risks is a big one, but also the question of managing the quality of the systems being built this way over time to ensure they are maintainable.
This change is the most obvious and the one that I suspect the industry is most well-equipped to figure out over time, even though it will be bumpy in the interim. A lot of the conference was dedicated to people sharing lessons about managing some of this transition, and I suspect most people reading this are thinking about this actively. Sorry, I don’t have magical solutions to offer here except that we all need to keep experimenting and sharing what we learn through the transition.
Signal Processing
Ok so this is a big one. First of all, what do I mean? I mean things like diagnosing problems, debugging teams, making decisions in the face of ambiguity. How good are you at not just getting information but getting the right information in order to make the best decisions?¹
We start with another Jack Dorsey quote:
Companies move faster or slower based on information flow. Hierarchy and middle management impede information flow.
The idea that many of these management doomsayers are building on is that AI can provide essentially a cybernetic mechanism for overseeing organizations that allows information to pass through computers, create signals, and inform far fewer decision-makers than current organizations because (waves hands).
And it is true that in some ways AI is a leap forward in visibility. If you produce digital artifacts of your work (aka stuff that can be consumed by the AI), someone with access to those artifacts can ask “what has Jane been working on the past week” and get a reasonable starting point answer. (Aside: I urge you all to turn your favorite internal AI on yourself and review the answer, to level set in the quality of the response.) I can believe that this visibility helps managers of individual contributors get a valuable starting point for their 1–1s that could allow them to spend less time on pure status updates and more time going deeper into what is happening.
But how well does it actually improve understanding? My experience is that when you ask these questions about things that span more than an individual, the answers start to get less useful, to the point that they lead to questions and clarifications with individuals that take as long as it would’ve taken to ask the humans for updates in the first place. And while perhaps those who try harder can make this work better than I have, I also know that my asking the leaders is training them to think about what is important, and focus their own attention on understanding the work of their teams and the justifications for that work. This is quite important in keeping a well-running and well-aligned organization.
Furthermore, we have a problem that reliance on AI tooling may be inhibiting the development of situational awareness. You know, your ability to read the room? A lot of information is carried in non-text formats. Body language, tone of voice, camera on, camera off, the things said and the things left unsaid. Maybe AI will read our minds someday (creepy), but I suspect many of us are starting to lose those subtle situational awareness signals because we are tuning out in meetings where we are relying on AI to do all of the note taking and transcription for us. Similarly, relying on summaries even of digital artifacts means that we are trusting that the algorithm has highlighted the important details we have to know, but it is necessarily stripping out some amount of information that might actually encode useful signals we might want to be aware of.
All of this culminates in the challenge of AI and decision-making. You can use AI to get visibility to assist decision making; when writing a performance review, giving the AI a competency matrix and a person’s name and asking for work they’ve done over a time period against this matrix can be useful in reminding you of things you may have forgotten, combatting recency bias. But the minute the AI starts to write that performance review, it’s starting to decide for you whether you think it is or not. AI writing is compelling, it draws your thinking in a direction, and it’s hard to fight that pull. And this undertow is dangerous.
Why? Machines cannot be held accountable, so they cannot make management decisions, to paraphrase the immortal words of IBM circa 1979. I once got a performance review from a boss who basically just copy-pasted my own self review in and called it a day. Now, I wrote myself a nice review, I got a good score on this review, but was I happy? No! I was terribly disappointed that my boss showed I was worth so little time and attention, that he had basically nothing to add. It made me lose trust in him immediately. When you make management decisions, and yes, your performance reviews are management decisions, you need to be able to stand firmly behind them and own their consequences. Could you defend every word the AI wrote in the review? If so, ok. But for most of us, to get to that point of certainty, we need to do the writing ourselves, to force the thinking, analysis, and decision-making.
People
Finally, we come to people. This one is just depressing. Has AI made anything easier about the people management/people skills era? If it has, it has been swamped by all of the new interpersonal challenges we’re hitting thanks to AI.
To start, there’s the increased expectations. Many companies think that that increased visibility and waves hands AI productivity means managers can start to manage far larger teams, with no help coming as we reduce managers and don’t hire new ones. And of course, some companies are also expecting managers to write code because that is apparently the only individual contribution that matters (the idea that managers don’t make valuable individual contributions to companies is so mind-boggling I could write a whole additional piece on it). And while you’re doing this, you have to manage the AI transition: make sure your team is using AI enough, not too much, and soothe their anxiety, and deal with their impatience, and you too have to use it but no one is bothering to teach you how to use it well or soothe your anxiety, you are just expected to figure it the f* out.
So is it any wonder 1 in 3 managers are considering going back to individual contribution roles? Hell, if you’re going to ask me to write code and manage a huge team of people, I might just prefer to say, thanks, writing code is enough for me so I’ll go back to doing that.
And these are not all the people challenges! Have we mentioned the politics? Oh my god, the politics. Who is supposed to build what? Do organizational boundaries matter? “Working code wins” was a useful mantra for a long time; if someone wrote code to solve a problem, it was a sign they were really willing to put in some elbow grease and effort, they cared a lot, they had a strong opinion. This didn’t always lead to great outcomes, but it was a useful signal that whoever wrote that code cared. Nowadays? Code is cheap. Someone gets slightly annoyed that they might have to think about another pre-existing system they didn’t build and decides to just build their own, and btw, now this should be our team’s responsibility and ownership. Managing through these politics, bad feelings, and frustrations is no fun at all.
AI has corrupted the cultural channel at work. When humans write code, they think about who owns what, where the systems they want to use reside. Conway’s Law is both an observation and something useful for humans: I know that if I work on a team, I work on certain systems, and I work closely with teams nearer to me in the org chart, and if I want to do something with work from teams farther away that may require more thought, conversation, etc. AI cares nothing for this cultural information. AI will take dependencies on anything it can see, and can quickly misuse those dependencies, without the initiating team realizing they are doing anything wrong.
And worse, AI undermines the small day to day activities that help build trust across organizations. It’s great that AI can quickly answer simple questions, and I would never suggest taking that away. But when you talk to glean, slackbot, claude, chatgpt, you aren’t talking to your colleagues. You aren’t asking them for help, and exchanging the kind of low-stakes favors that form connections; you aren’t learning their personalities and building your network.
Depressing Interlude About The Future
I worry that all of this gets worse before it gets better.
That our productivity obsession will combine with our general habit of rewarding performative “first to do a new thing” behavior and lack of communication.
This will cause a lot of “almost real” 80% systems to pop up everywhere.
And let’s not even talk about the potential chaos of Citizen Developers.
We’re heading towards a chaos of ungovernable fragmented systems and logic that is missed by overburdened managers, driven by the cult of productivity and the inability to differentiate visibility from understanding.
So What Do We Do?
If you still want to stick with management despite all of these challenges what do you do?
Don’t regress to easy
It’s my favorite cliche, stay outside your comfort zone, but in this era it is truer than ever. For some of you this means that you need to stop spending all of your time playing around with new ways of coding and using AI to “manage” your teams. It’s fun to learn this new stuff but your job is as much with your people as it is with the new tools, and you do no one any favors by forgetting that.
For some of you this means get your head out of the sand and get engaged. You can’t just hold this back through sheer inertia. If your company is using AI, you need to get comfortable with it so that you can help your team navigate the transition, and to avoid becoming irrelevant.
What should we be doing?
The important work is, of course, to make AI work with and for our teams. A lot of attention is going into the changes to workflows and processes that will make AI more successful: fixing code review, thinking about evals and new ways of testing, etc. And that’s certainly one of the most interesting technical things to spend time on and pay attention to.
But a lot of other things need to change as well, and managers need to remember that the cultural changes around policies and standards are just as important. An example of this is the Respectful AI Policy work I wrote about a few months ago. Teams need guidance on how they will work together with AI, and show up to work when they have AI available, and this is something that managers and leaders should provide.
I mentioned earlier that I’m concerned with our ability to develop situational awareness in the face of all of these note takers and summarizers that encourage us to tune out and favor getting the cliff notes over the details. The way to combat this is the same as it always was: you have to follow up on surprises. Writing that Respectful AI Policy came from me talking to leaders in one of my teams and hearing their frustrations. I will admit I was surprised! I figured surely people wouldn’t just send slop over to their colleagues or tune out all the time, but of course that was me overestimating human nature a little bit. No one catches every important detail, and you may risk missing more if you rely on AI, but if you follow up on the surprises that come out of this you have the chance to both fix the mistakes and identify areas to follow details more closely.
Remember the Goal
My final idea for managers navigating this new world is to remember that we’re trying to take advantage of tools that provide incredible individual productivity while maintaining and even improving team coherence. This means cooperation and trust matters; how can we help pre-empt and navigate the inevitable turf wars coming from cheap code everywhere?
My suggestion is one that I suspect some will dislike: I think you actually need to spend more time talking to people. You’ve got all of these productivity enhancers, and they are driving everyone in the organization to move very fast. It was never easy to keep up with the pace of work, but now, one person on the team can produce as much code as several people, in a very short timeframe. When you have multiple people you’re trying to keep track of (let alone multiple teams!) I am not sure that even the best summarizations are going to catch the ways these individuals and groups can drift off in different directions and start to lose alignment and coherence.
We should be spending some of our productivity gains on things that seem like slowing down: meetings, design reviews, conversations. These ways of maintaining alignment, sharing knowledge, and creating human connections are, in my opinion, more important than ever to creating good outcomes at scale.
Coming Back to 2026 and Concluding
Back in 2026, what’s happened recently?
Well, Meta just (as of ~9/13/26) rolled back some of its most aggressive management reductions, recognizing that there were a lot of bad outcomes that came from going to a normal number of managers to ratios of 40:1 or 50:1.² So I think that perhaps we are not in fact ready to drastically reduce management and maintain high performing organizations.
I wanted to end the talk with a different quote, from my friend Harper Reed who spoke recently to a group of high school students and among other things said this:
Always ask: “Who has power in this system?”
When someone tells you how a system works, ask who benefits from you believing that version.
Now, the quotes I started this talk with were from incredibly successful tech CEOs, and they clearly are trying to sell us a future where management doesn’t matter, where many of us may even basically work for computers. There are many powerful people in the world who want to sell us the idea that humans don’t matter in the future, not really. But I want everyone listening to this talk to remember this:
You have power too. You are part of the system, you are building the system, you can make things better or worse based on your choices and your behaviors and how you show up as a leader.
You must reject the hopeless messaging that paints the demise of you role as inevitable, that paints us as powerless to shape the future.
And, importantly: this will only work if you talk to humans more than you talk to Claude.
So I hope that everyone listening to this talk will join me in staying curious, engaged, and committed to making AI work for humans. Working together, we will chart the next path.
Thank you.
¹There’s a lot I didn’t say here; I really wanted to quote a lot of things from AI as Social Technology by Henry Farrell and Cosma Shalizi, various bits of wisdom from Peter Drucker’s writing, and a few things from my friend and occasional collaborator Tim Danford. Somehow only Tim made it to the final edit, contributing the ideas about AI corrupting the cultural channel, and even that was me indulging myself a little bit.
²https://laughingmeme.org/2026/09/14/managers-are-catalysts.html for more tidbits and related good ideas
I’d be remiss not to thank a laundry list of people for contributing indirectly to this talk, but it’s literally so long I fear leaving too many people out. If I talked to you about AI and management in the past 6 months, know that you were in my mind as I was writing, and I genuinely appreciate it.
Enjoy this post? You might like my books: The Manager’s Path, and Platform Engineering: A Guide for Technical, Product, and People Leaders, available on Amazon and Safari Online