When I was an individual contributor, I used to wonder what my manager did all day.
I wrote code. I shipped features. I could point to what I had made. At the end of the week there were commits, tickets, artifacts with my name on them. My manager, by contrast, seemed to float from meeting to meeting, answer emails, and occasionally appear with a question or a decision. Whatever they were doing, it did not look like work you could print.
When I became a manager, the question changed. I would leave exhausted, having done almost nothing that showed up in any system. I had turned three vague requests into one coherent plan, quietly killed two ideas that would have broken the team, brokered a truce between functions, rewritten a roadmap in my head three times as new constraints arrived, and spent the rest of the day making sure none of that turbulence reached the people who actually built things. My to-do list remained intact. My calendar was wreckage. The work lived almost entirely in the white space between other people's tasks.
Good managers, like good system administrators and security teams, seem to do almost nothing at all. You only notice them when they are gone. When they are present and effective, there are no visible fires. Deadlines slip less than they should. Bad ideas die quietly before they hurt anyone. The absence of catastrophe looks, from the outside, like the absence of effort.
That invisibility is the first part of the problem. The second is that managerial work lives at the intersection of two worlds that cannot see each other directly.
At the top of a company, there is too much context to act on directly. A founder or CEO carries investor pressure, cash runway, market shifts, regulatory risk, competitor moves, internal politics, and a long tail of prior commitments, all in their head at once. These forces do not line up politely. They arrive as overlapping tensions, half-formed fears, glimpses of opportunity that may or may not be real. To live in that role is to be bombarded by meaning.
At the bottom, there is too little context to act without guidance. An engineer knows a codebase and a toolchain. A salesperson knows their accounts and pipeline. An operator knows their systems and procedures. Their knowledge is real but local. They do not sit in board meetings. They do not spend evenings scanning analyst reports. Their job is to make something work, not to carry the weight of the company's entire cognitive load while they do it.
Between these two spaces, something has to happen. The CEO's raw thought-stream cannot be poured directly onto the people doing the work. Nor can raw status updates be poured back into the CEO's mind and produce coherence. The two sides inhabit different coordinate systems. Someone must perform the transformation.
We give that role many names. Manager. Lead. Director. Whatever the title, the work is the same: interpret, translate, project, and regulate.
The manager takes the company's strategic direction—often expressed in fragments, contradictions, and implicit assumptions—and converts it into something actionable for people whose jobs require focus rather than synthesis. They take ground-level reality—which is granular, partial, and fast-moving—and compress it into something the executive level can actually use to make decisions. They do both of these things simultaneously, adapting the message for each direction, filtering what would paralyze and amplifying what would clarify.
This is not coordination. Coordination is scheduling. This is not reporting. Reporting is data transfer. What managers do—when they do it well—is closer to translation, and translation requires fluency in both languages. It requires enough context at both levels to know what must be preserved, what can be simplified, and what must never be said to whom.
The organization concludes that management is mostly coordination and reporting. It believes both can be automated, and that whatever remains can be absorbed by the layers above and below. It assumes that if everyone hears the same words from the top, everyone will carry the same understanding. It then removes or overloads the only layer structurally positioned to hold both worlds in mind at once.
For a time, this looks like efficiency. Then the distortions appear.
Companies that tried to operate with no managers found that once they crossed a certain size, product decisions stalled, strategy drifted, and informal cliques took over the work of deciding what mattered. Companies that hollowed out their middle to "empower ICs" discovered that their executives were suddenly drowning in decisions about expense reports and interpersonal conflict, and their ICs were suddenly expected to carry business context they never signed up for. Companies that layered management on management to be "rigorous" found that no one could move without running a gauntlet of approvals, and that the people with the most context were the ones with the least time to think.
The common failure is a misunderstanding of what lives in the middle.
This does not mean every manager earns their place. Some hoard information, create work to justify their existence, or use their position to avoid accountability. Those people are parasites on the transformation function, not instances of it. Removing them is not falling into the trap. Confusing them with the function, and concluding that the function itself can be eliminated, is.
Resisting the Context Trap means acknowledging that context is asymmetrically distributed, that it cannot be fully equalized, and that the work of translating between levels is itself a form of production.
It means recognizing that there is a maximum number of people whose work and lives one person can meaningfully understand at once, and designing spans of control with that limit in mind. It means treating the time a manager spends in conversation, in thought, in quiet judgment about what not to pass along, as part of the work. It means using automation to strip away the mechanical parts of the job so that more attention can go to the parts machines cannot touch.
It also means tolerating slack. If everyone in the organization is visibly at one hundred percent capacity on tasks that fit neatly into systems, no one has time to do the invisible work of seeing around corners. Slack is where interpretation, synthesis, and care happen. A company that insists on eliminating all slack has already chosen what kind of failure it will have.
You can remove the boxes labelled Manager from a slide. You can flatten titles, deploy agents, and speak to everyone at once from the stage. For a while, you may even feel more in control. But unless you have decided, deliberately, who will stand in the space between too much context and too little, that space does not go away. It fills with noise, and with people who were hired to do other things quietly trying to perform a job no one admits still exists.
Somewhere, in every functioning organization, there must be humans whose work is to hold two incompatible worlds in their head at the same time and make them intelligible to each other. When you erase that layer in the name of efficiency, the system keeps moving, and measuring, and reporting—for a while. What it loses first is the ability to understand itself.