{"slug": "youre-the-executive-now", "title": "You’re the Executive Now.", "summary": "An engineer argues that working with AI agents at scale mirrors the challenges of executive leadership, where a manager cannot directly see or understand all the work being done and must rely on trust, process, sampling, and key metrics. The author contends that directing collections of AI agents requires the same coordination, accountability, and vision as running a large human organization, and that this shift makes team collaboration harder as individual nodes become more capable and self-contained.", "body_md": "Working successfully with AI at scale feels a lot like being an executive leading a large organization.\n\nIf you lead any kind of larger organization, particularly a technical one, you quickly realize that it’s not possible to know all of what is being done in it. For a while, it can be possible to stay fairly close to the work, but eventually, enough is being done that you can’t even read or see all of it. This isn’t surprising - if one person could do everything, you wouldn’t need the team in the first place. So it’s an inevitable part of leadership, particularly executive leadership. There are all kinds of ways to compensate for this - trusted lieutenants, random checks, key metrics, specific roles that are meant to keep an eye on some particular behavior (like a CFO watching finances), etc.\n\nSome of these mechanisms run on trust, some on judgement and experience and some on process. All are necessary to run a successful large organization, and the role is very hard - you have to steer something larger and more complex than you are, filled with dynamic agents (people) who have their own complex incentives, and with imperfect information. As a bonus, the larger the team is, the harder the problems are that eventually make their way to you - because ever more people have failed to solve them on the way up. It’s a hard job and not everyone can do it.\n\nUnfortunately, it’s now a job that everyone has, or will have. Managing a collection of AI agents feels a lot like managing a large group of people. By definition, you can’t see or understand all of the work directly (you can, but if you are working at that scale, someone else is working at the larger one, getting more done - the pressure is very much in the direction of doing more, with more agents). You have to develop the same kind of mechanisms - trust, process, sampling, tooling, critical metrics, etc.\n\nWhat does an executive actually contribute? A mixture of things: accountability, coordination, vision/ambition, and broader context. All things that are valuable, and needed if you are going to scale work up.\n\nDoing work by building a larger system of agents (whether human or AI) that you can only partially understand, is a fairly foreign concept for most people. Most engineers (like me) go into leadership just because they have to - someone needs to do it, we need more hands on the problem, and they have to be coordinated. I’ve written in the past about the humility needed to do the role well.\n\nI think people are starting to understand that this somewhat unnatural posture, being an executive, is being brought forward by AI. This makes some people excited to use it and some nervous or even unhappy. Often, it has to do with whether you are comfortable or enthusiastic directing a collection of agents on your behalf or want to do the work “by hand”. There is a lot of value in hands-on work, of course, but there is also a lot of advantage and leverage from being able to direct larger systems to do work.\n\nHow does this fit with the idea of a “packet-switched organization”? Isn’t this the opposite? Not really. But it does make it harder.\n\nIn a more “packet-switched” world, work is routed to the right person (or agent) to do it at the right time. AI in the network is doing meaningful processing on “packets” of work to help with that routing, but there are also hints in the system - the equivalent of routing tables etc.\n\nBut “nodes” in this system are much more capable now. Each person is the equivalent of their own small company. This makes individuals more productive but makes it much harder to work in teams now. To work together well, boundaries, responsibilities, and domains of trust have to be even more well defined - think about how much harder it is to work with someone outside of your company or corporate culture. You have to know very precisely what the agreement is, what you can and can’t trust, how to adjudicate disagreements, and more. That friction is the cost of aligning two large, complex cultures to a common task. That will start to happen (is starting to happen) as individuals scale more work with AI.\n\nIn the packet switched world, responsibility and standards can still be stable, just like a consulting company has a fixed role. But assignments and working relationships can be much more fluid. The two are compatible, even if more complicated.\n\nThe real risk is one that every executive is familiar with - you can’t let yourself be the bottleneck. Every person (or agent) in a 1000-person team having “just one question” is overwhelming for the boss. The system has to optimize for, and be considerate of, the limited attention of the operator. In a real org, typically that means offloading responsibility to other people you’ve hired. In an agentic one, it means building automated ways to check, prove, and test answers. If every agent is waiting on you all the time, they’ve DDOSed their “control node” - this is a bad design.\n\nOne big challenge here is that it’s easy to reduce that attention load by “manufacturing” confidence. Less work needs to reach your attention, but if the tests, filters and mechanisms you’ve developed are poor proxies for reality, you won’t get good signal. Every executive knows not to just believe the presentation, but to ask probing questions. And they usually know that their teams are prepared for these and will have easy answers to the obvious ones. A good executive is able to ask unanticipated questions or ones with hard-to-fake answers, to get a real sense of what’s going on. It’s not antagonistic per se, but it’s not entirely trusting either.\n\nThis happens with AI too - they can cheat, assert tests are working when they aren’t, have generous interpretations of the work to be done, just like people do. If you’re never hearing about problems from your large engineering team, it’s likely that you aren’t getting a good signal, you’re getting sycophancy. This is true of AI as well.\n\nWe are still at the early stage where most people are just directly managing a single AI worker. Some, but not many, are starting to develop the kind of organizational leverage that I’m talking about here. But that time is coming. It used to be very hard to be a leader of a 100-or-1000-person team - you had to spend a lot of time working your way up, building confidence and skills. Now, having 100 agents working for you is a simple matter of dollars and prompting - but that doesn’t mean you are instantly effective at it, any more than dropping a new college graduate into a CEO role is likely to be successful.\n\nYour capacity to generate and direct work now is growing much faster than your ability to judge it. If you’re not careful, it’s easy to get a lot of activity but little progress, or to be fooled into thinking something works or is correct, when it isn’t. You have to build systems to judge that work, and you have to understand enough about them and the domain to be able to judge if those systems themselves are working.", "url": "https://wpnews.pro/news/youre-the-executive-now", "canonical_source": "https://sundaylettersfromsam.substack.com/p/youre-the-executive-now", "published_at": "2026-09-20 23:01:44+00:00", "updated_at": "2026-09-20 23:23:35.207426+00:00", "lang": "en", "topics": ["ai-agents", "artificial-intelligence", "ai-tools"], "entities": [], "alternates": {"html": "https://wpnews.pro/news/youre-the-executive-now", "markdown": "https://wpnews.pro/news/youre-the-executive-now.md", "text": "https://wpnews.pro/news/youre-the-executive-now.txt", "jsonld": "https://wpnews.pro/news/youre-the-executive-now.jsonld"}}