The middle of the org chart is where AI cuts first.
Why experience no longer guarantees safety and judgment is becoming the new seniority.The comforting story about AI and jobs is that it comes for the bottom first, the entry-level and the rote, and works its way up slowly enough that experience keeps you safe. The actual pattern forming in 2026 is…
Why experience no longer guarantees safety and judgment is becoming the new seniority.The comforting story about AI and jobs is that it comes for the bottom first, the entry-level and the rote, and works its way up slowly enough that experience keeps you safe. The actual pattern forming in 2026 is stranger and less comfortable.The pressure is landing hardest in the middle, on the large tier of competent, experienced people whose work is real but repeatable, while both the very top and, oddly, the irreplaceable get more valuable. If you are in that middle, and most people are, the safe-because-experienced assumption is the one that will hurt you.Jason Lemkin gave the clearest concrete version of this when he described replacing SaaStr’s entire go-to-market team. After his last salesperson quit, he stood up twenty AI agents managed by 1.2 humans, and they now do the work of roughly ten SDRs and AEs at equivalent business performance. His specific observations are the useful part.The agents respond instantly at eleven at night on a Saturday, never cherry-pick the good leads, and follow up on every opportunity, so response and qualification actually improved. His blunt prediction is that junior SDRs hired out of college to send emails should be extinct within a year, and his sharper claim is the one I want to build on: the biggest impact lands on the mid-pack, the transactional and repetitive roles, not the top and not just the bottom.Why the middle, and not the bottom, gets hollowedThe intuition that AI comes for the bottom first is built on the last century of automation, where machines took the least-skilled physical work and left the cognitive work alone. This wave inverts that, because the thing it is good at is exactly the competent, repeatable cognitive work that defines the middle of most org charts.The middle is where the largest number of people do work that is real, requires skill, and follows patterns stable enough that a system trained on enough examples can approximate it. That is the sweet spot for the current tools, and it is also, not coincidentally, the most populated tier of the knowledge economy.The bottom is partly protected for a boring reason, which is that the very bottom often involves the ambiguous, physical, or trust-laden tasks that models still handle poorly, and partly exposed for Lemkin’s reason, that the specific entry role of sending templated emails is pure pattern and dies fast. So the bottom is uneven.The top is protected because the top is where judgment lives, the deciding-what-matters that the tools cannot do because, as Edwin Chen of Surge AI put it, the models have no drive of their own. The middle is the tier with the least protection, because its work is too patterned to be safe from the tools and too far from the real decisions to be irreplaceable. It is the fat part of the distribution and the soft part of it at the same time.The squeeze is not limited to individual contributors, and the clearest early evidence is in what senior people are doing with their own careers. Jenny Wen was managing twelve to fifteen designers at Figma, a solidly upper-middle perch by any org chart’s standards, and left to become an individual contributor at Anthropic because she doubted the future of middle management in design. Read that move for what it is.A person with a direct view of where the tools are going looked at the coordination layer she occupied and decided the IC seat was the safer one. Middle management is patterned work too. Much of it is routing information, consolidating status, and translating between layers, and those are precisely the tasks agents and better tooling absorb first.When the people doing a job start exiting it on purpose, before any layoff forces the question, that is usually the most honest signal available about the job. There is an uncomfortable corollary in Lemkin’s numbers.The agents did not just match the human team, they improved on the parts humans are worst at, the eleven-p.m.-Saturday follow-up, the refusal to cherry-pick, the coverage of every opportunity. The middle-tier worker’s competitive advantage over a junior was reliability and coverage, doing the repeatable work consistently and at volume.That is precisely the advantage the agents erase, because tireless consistent coverage is what they are built for. The middle worker’s edge over the person below them turns out to be the edge the machine most easily takes.The top got more valuable, and that tells you where to goThe other half of Lemkin’s story is the part people skip, which is what happened to the people who were not replaced. The agents did not flatten everyone toward the machine’s level.The 1.2 humans running twenty agents became more valuable, not less, because their job changed from doing the transactional work to directing and correcting a system that does it. And the failure modes he documents tell you why the human on top matters. Unsegmented agents spam the same prospects and conflict with each other.Rolling out twenty of them is exhausting enough to burn out a team. Someone has to segment, orchestrate, catch the conflicts, and decide what the agents should be pointed at, and that someone is worth more than the ten people whose output the agents now produce.This is the shape of the escape, and it is worth being literal about it. The value moved from producing the repeatable output to judging and directing the systems that produce it. “Senior” is quietly being redefined. It used to mean you had done the work long enough to do it faster and better than juniors. It increasingly means you can evaluate delegated work, catch where the system went wrong, and decide what should be produced in the first place.Those are different skills from being an excellent individual producer, and the person who spent a decade becoming an excellent individual producer has been building exactly the capability the machine now supplies, while under-building the judgment-and-direction capability that is becoming the actual definition of seniority.The institutions are starting to write this into their org design, which is how you know it is more than a thought experiment. LinkedIn scrapped its legendary Associate Product Manager program and replaced it with an Associate Product Builder program that teaches coding, design, and product together, and its CPO Tomer Cohen introduced a formal Full Stack Builder title and career ladder that lets someone from any function take a product from idea to launch.His diagnosis was that product development had accumulated too many handoffs and AI collapses the roles. Every handoff that disappears is a middle seat that disappears with it, because the middle of the org chart is largely made of handoffs.Tal Raviv had already described the individual version of this in one of Lenny Rachitsky’s most popular guest posts: the super IC, a single contributor using AI tooling to do what previously took a pod, thriving in what he calls the great flattening of management layers. Cohen is building the org chart for the world Raviv described, and neither version of it has much room in the middle.You can see the same restructuring outside sales. Tom Orbach, writing about handing marketing work to Claude Fable 5, described what competitor analysis looks like now: the model assembles its own research operation, something like 150 agents working across 700-plus sources with fact-checking, and produces a result that would have taken a team. Notice what the marketer’s job becomes in that story. Not doing the research.Deciding what question is worth 150 agents’ effort, and judging whether the output that comes back is any good. The doing collapsed into the tool. The value relocated to the framing and the judgment, which is the top of the ladder, not the middle of it.The concrete skill that makes you hard to cutLemkin ends on a career claim that I think is the most actionable thing in the whole account. Anyone who can deploy, train, and iterate AI agents themselves is, in his word, hyper-employable. This is more specific than the usual “learn AI” hand-waving, and the specificity is what makes it useful.He is not saying learn to prompt a chatbot. He is saying be the person who can stand up a system of agents, point them at the right work, catch the failure modes, and improve them over time, because that is the 1.2-human role, the one that got more valuable while the ten roles disappeared.If you want a way to locate yourself, the eight-level adoption ladder that Mike Taylor and Laura Entis published at Every is the most useful map I have seen. It runs from chatbot at level one through copilot, agents with approval checkpoints, autopilot with post-hoc review, workflows, proactive assistants, multi-agent systems, and finally an orchestrator, a meta-agent managing teams of agents.Their finding is that most knowledge workers live at levels one through four while engineers reach five through eight, and their explanation for why people stall is practical rather than psychological: the output quality is either too low for the work they do or too expensive. They are also careful to say a higher level is not automatically better, and that the most sophisticated users they know operate at several levels at once.The career reading of the ladder is blunt, though. The mid-tier roles being absorbed are the ones whose daily work sits at levels one through four, and the roles getting more valuable are the ones comfortable running things at five and above. Knowing your level, honestly, tells you how much of your job is already inside the machine’s reach.The move out of the kill zone is upward into that role, and the direction is toward directing the work rather than producing it. Concretely, that means spending your reps on the parts of your job that are about judgment and orchestration, and deliberately learning to run the tools that do the repeatable parts, so that when the transactional layer of your job gets automated, you are the one who automated it rather than the one it replaced.The middle-tier worker who says “I deployed the agents that now handle the volume, and I run them” has moved to the top of Lemkin’s structure. The one who kept doing the volume by hand, better and faster than juniors, has optimized the skill the machine just made abundant.I want to be honest that this is easier to say than to do, because it asks people to abandon the very thing they are best at. The competent middle-tier professional is competent precisely at the repeatable work, and telling them the escape is to stop doing that work and start directing systems that do it is telling them to give up their identity as a skilled producer.That is a real loss, and the advice to just move up the value chain is glib about how much it costs to stop being good at the thing you are good at and start being mediocre at a new thing. But the alternative is worse, because staying excellent at the automated layer is investing more effort into the skill with the fastest-falling value.Where the picture is messier than the pitchThe neat version of this argument would say everyone should race to become the agent-orchestrating senior, and the messy truth is that not everyone can, and the pyramid does not have room for it anyway.If twenty agents plus one human do the work of ten, then the org needs far fewer of the middle-tier roles and only slightly more of the orchestrating roles, which means the middle does not all get to move up, it mostly gets smaller. That is the part the hyper-employable framing quietly elides.Being the person who can deploy agents protects you, but it protects you by being scarce, and scarce means most of the middle does not get there. I do not have a comfortable resolution to that. The individual advice, become the orchestrator, is correct and also cannot be true for everyone at once.There is a more optimistic counterargument, and it deserves a fair hearing because it comes from someone actually running the experiment. Dan Shipper’s “After Automation,” the most-read thing Every published this year, argues that AI creates more human work rather than less.His company has automated everything it can, code, email drafts, support, newsletter compilation, and is still around thirty people and still hiring, because when cheap competence floods in, the bar for good work rises with it, and agents are structurally built to rely on humans for direction.“There will always be a new frame for humans to hand it,” he writes. I half buy it, and I notice the half I buy is the half that concedes my point. Even in Shipper’s story, the new human work is judgment-shaped, deciding what matters, reviewing, directing. The work that got automated away is exactly the competent repeatable production the middle tier is made of.His future has plenty of humans in it. It just does not have many humans doing the job the middle currently does, and whether the displaced can become the directors he still needs is the question his essay does not answer.The other honest caveat is timing. Lemkin’s twenty-agent team is one company’s account, run by someone with an incentive to tell a dramatic version, and the failure modes he lists, the spam, the conflicts, the burnout, suggest the transition is rougher and slower than the clean numbers imply.The middle is not going to vanish next quarter. What I am confident about is the direction and the ordering. The pressure is landing on the competent, repeatable, mid-tier work first, the safe-because-experienced assumption is backwards, and the durable position is the one that judges and directs the machine rather than competes with it on the repeatable output it was built to produce. The middle is the kill zone.The way out of it is up, toward the work that decides and directs, and the ticket up is being the person who runs the tools rather than the person they replace.This story is published on Generative AI. Connect with us on LinkedIn and follow Zeniteq to stay in the loop with the latest AI stories.Subscribe to our newsletter and YouTube channel to stay updated with the latest news and updates on generative AI. Let’s shape the future of AI together!The middle of the org chart is where AI cuts first. was originally published in Generative AI on Medium, where people are continuing the conversation by highlighting and responding to this story.Source: Generative AI Pub — Published — Category: Image AI