The Best Work in Your Company Loses 90% of Its Content on the Way Up

Here is something that happens every day in every company with more than a hundred engineers.

A backend engineer builds a data pipeline. It processes customer events in near real-time, deduplicates across three source systems, handles schema evolution without breaking, and cuts downstream latency by 60 percent. It solves a problem the company has been stuck on for a year. It is genuinely elegant work.

The engineer explains it to their manager, who wrote code five years ago and follows most of it — but the deduplication strategy and the schema evolution approach go past them. They nod in the right places and write it up for the director.

The director reads the summary. Specifics have already blurred. "Near real-time event processing" becomes "faster data pipeline." The 60 percent sticks, because numbers travel well. The deduplication strategy — the hardest and most valuable part of the work — becomes one bullet.

The VP reads the director's version between two meetings and extracts what they need: new pipeline, faster, done. It goes on a slide as Data infrastructure modernization — Phase 1 complete. 60% latency improvement.

The CTO glances at the slide, says "good progress on the data side," and moves on.

Four hops. The substance is gone. And nobody in that chain did anything wrong.

The Layer Existed for a Reason

The instinct is to call this bureaucratic waste. It wasn't. The chain was solving a real problem, and understanding what it solved is the only way to see why it's now in trouble.

Executives cannot hold the details of forty engineering projects. Engineers cannot spend their week producing status in five different formats for five different audiences. So organizations built a translation layer — people whose job is to compress technical reality into business language going up, and business priority into technical direction coming down.

That is genuine work, and it was expensive because it required humans who understood both vocabularies. The compression loss was the accepted price of the executive being able to function at all.

The layer grew for a second reason too: coordination. Someone has to know what four teams are each doing, notice that two of them are building the same thing, and route the dependency. In a large organization that is a full-time job, and for thirty years the only thing capable of doing it was a person.

What Changed

Take an honest inventory of what an average manager — not the best, not the worst, the middle of the distribution — actually does in a week.

Aggregating status from direct reports. Maintaining priorities across a backlog. Writing summaries for the layer above. Reformatting the same information for different audiences. Sitting in meetings to stay informed. Tracking who is blocked on whom. Producing the artifacts that let other people believe work is happening.

Look at that list and notice how much of it is information movement rather than judgment. Collecting, compressing, reformatting, and routing — done by a human because until recently there was nothing else that could read a technical update and produce a business summary from it.

That constraint is gone. The work that filled most of the calendar of most of the middle layer is now the exact category of task the tooling does well and cheaply, and it does it without the four-hop degradation, because it can hold the original detail and generate the summary appropriate to each audience from the source rather than from the previous summary.

This is the part that matters. The problem was never that summarization happened. It's that each layer summarized the summary. When a CTO can ask a question and get an answer derived from the engineer's actual description rather than from three intermediate paraphrases, the value of the intermediate paraphrases collapses.

What Doesn't Go Away

It would be easy — and wrong — to conclude from this that management is over. Several things in that job were never information movement, and none of them are getting automated.

Deciding what not to do. Prioritization looks like information processing and isn't. It requires knowing which customer relationship is fragile, which executive will fight this, and which of two credible technical arguments to back when the evidence is genuinely ambiguous. That is judgment under political constraint, and the inputs are mostly not written down anywhere.

Absorbing risk on behalf of a team. Someone has to say "we're doing it this way, and if it's wrong it's on me." That is not an informational act. It's an accountability structure, and a team without one either freezes or routes every decision upward.

Developing people. Knowing that a particular engineer is six months from being ready for a bigger problem, and constructing the assignment that gets them there. This requires sustained attention to a specific human being over years.

Handling the situations that are actually about people. Conflict, underperformance, someone going through something difficult, two strong contributors who cannot work together. None of this is a coordination problem.

The uncomfortable observation is that these were always the hard parts of the job, and in many organizations they were the parts that got squeezed out — because the coordination overhead expanded to fill the week, and status aggregation is more legible and less unpleasant than telling someone their performance is a problem.

So a manager whose week is mostly the first list is in a genuinely difficult position. Not because they are bad at their job, but because the job they were doing is being commoditized, and the residual is the part they have had the least practice at.

The Reorg Nobody Planned

What follows from this is not a decision anyone announces. It happens through hiring that doesn't get backfilled, spans of control that quietly widen, and layers that stop being replaced when someone leaves.

The people who come out well are the ones whose contribution was never the translation. Builders who ship, and can now explain their own work upward without an intermediary, because the tooling handles the register-shifting. And leaders whose value was judgment, accountability, and developing people rather than being the pipe that information travelled through.

The people who struggle are the ones who — often through no fault of their own, in organizations that rewarded exactly this — became very good at being the pipe.

If that describes your week, the question worth asking is not whether the tools can do your job. It's which parts of it were ever the job.

AI and the Leadership Illusion: Why the Engineers Who Ship Will Outlast the Managers Who Meet works through the whole shift — how the layers accumulated, what the coordination collapse looks like in practice, why judgment doesn't automate, and what to do whether you're the builder or the leader.