Missing translation layer: they built what you asked for

Try this first

Three months of warnings, and the one that mattered looked the same.

In February a commercial director asked for something reasonable. A weekly list of accounts at risk of leaving, so his team could get ahead of it. Two years ago that request would have gone to a person who knew the business, and that person would have come back with a question before writing a line.

This time it went straight into a tool. The list appeared on the Monday, exactly as asked: accounts ranked by decline in usage. Clean, sorted, delivered every week without fail.

The trouble is that three of their five largest customers are seasonal. Usage drops every February and comes back in April, every year, and everybody who has been there more than two years knows it. So the list opened with the same three names for three months running, the team learned to skip the top of it, and in May a genuinely troubled account appeared at number two and got skipped along with the rest. That one left in July.

Nobody built the wrong thing. Somebody asked for the wrong thing, correctly, and there was no longer anybody between the asking and the building. Paul Musters

Look at what was actually lost there. Not a skill and not a headcount. What went missing was a specific moment that used to exist, in which somebody read a request and said, before anything got made, that is not quite what you want.

How it shows up

Three months of warnings, and the one that mattered.

None of these is dramatic while it happens. All three are a request going through untouched.

Someone at a desk working through a request with their agent — **The brief that got built the same day** You asked for a dashboard on Tuesday and had it by Wednesday. Fast, and nobody asked what decision you were going to make with it, which is the question that used to change half of these requests before anything was built.

Two colleagues going over a piece of work at a screen — **Two teams solving the same thing differently** Both went directly from their own need to their own build. There is no longer anybody sitting where both requests would have passed, so nobody noticed they were the same problem until the two answers disagreed in a meeting.

A meeting where a decision is being worked through — **The strategy that everybody agreed with and nobody could apply** It was clear at the top and it arrived on the floor as a slide. The layer that used to convert it into what a team should do differently on Thursday was cut last year, and the conversion did not happen anywhere else.

The missing translation layer is a company where requests go straight to build with nobody in between. Seventy in a hundred leaders say they have an AI strategy. Thirty-nine in a hundred of their employees agree.1

The bones of it

Korn Ferry surveyed more than 15,000 professionals across 15 markets. 44% of US employees reported cutbacks in manager levels at their organisation.1
40% of US employees say they now feel a lack of direction at work.1
72% of US senior executives say they are stretched beyond their capabilities, against 47% of their global peers.1
70% of leaders believe they have an AI strategy. 39% of employees agree.1
Deloitte adds the part that makes it worse: managers are not prepared to be supervisors of AI, and many executives lack the AI literacy to oversee it either.2

This one usually arrives as

Something got built fast and answered the wrong question
Two teams solved the same problem in two different ways
A strategy everybody agreed with changed nobody's week
You are doing work you used to delegate and cannot say when that started

What you take away

What the middle layer was actually doing, which was rarely what its job title said
A test on one thing you had built recently, in ten minutes
The research, with the sample and what it does not cover
Two habits that put the missing question back without rebuilding a layer

What it is

The routing was the visible part. The rewriting was the job.

This is where most companies get the arithmetic wrong. If you describe the middle layer as routing, then removing it looks free, because a machine routes faster and never goes on holiday. The routing was the visible part of the job and the smallest part of it.

What that layer actually did was take a request in one language and hand back a different request in another. A director says at risk of leaving. Somebody who has been there four years hears that and thinks about February. What comes out the other side is a narrower, better question, and the person who asked usually never knows it was changed.

You did not remove a relay. You removed the place where a request got improved before anybody built it. Paul Musters

There were three things happening in that layer and only one of them is easy to replace. Routing: genuinely automatable, and good riddance. Filtering: deciding what does not need to go up, which AI can partly do. And translation, which is the one that needed somebody who knew both the request and the reality, and which nothing has replaced.

What sat in the middle, and what it was doing

LeadershipSets direction, asks for things, sees the summary

The layer that has been thinningRouted, filtered, and translated. Only the third one was hard to replace.44% of US employees report cuts here1

The workBuilds what was asked, now much faster and with fewer questions

Three layers, and the middle one is where a request used to be rewritten before anything got made.1

And here is the part that makes this specifically an AI problem rather than a reorganisation problem. Before, a badly framed request was slow, and slowness gave somebody time to notice. Now it is fast, and it arrives finished. The friction that used to catch the error was a side effect of things taking a while, and nobody ever designed it.

The evidence

The layer thinned, and both sides felt it in different ways.

Korn Ferry's Workforce 2025 research surveyed more than 15,000 professionals, from entry level to CEO, across fifteen markets including the United States, the United Kingdom, France, Germany and Japan.1 Three findings from it belong together.

44%

of US employees report cutbacks in manager levels at their organisation

1

40%

of US employees say they feel a lack of direction at work

1

72%

of US senior executives say they are stretched beyond their capabilities, against 47% globally

1

Read those three as one movement. The layer gets thinner, the people below it lose direction, and the work that layer was doing reappears on the desk of somebody senior who was already full. All of that work still exists. It moved up, to the most expensive place it could have landed.

And the translation gap, measured directly

The same survey has one number that is this whole page in a single line. Seventy in a hundred leaders globally believe their organisation has an AI strategy. Thirty-nine in a hundred employees agree.1

A strategy that thirty-nine in a hundred of your people can recognise still exists, and it is probably good. What has gone is the step that used to turn it into what a particular team does differently on a Thursday.

Deloitte's 2026 research adds the uncomfortable second half. Among the things it lists as unresolved in AI decision-making are managers not being prepared to act as supervisors of AI, and many executives lacking sufficient AI literacy to contribute to oversight.2 So the layer is thinner, and the people left on both sides of the gap are not yet equipped to do the translating themselves.

Two limits on this evidence. The sharpest Korn Ferry figures are American, and the global numbers are consistently milder, so a Dutch company of two hundred should read the direction rather than the size. And Korn Ferry sells organisation design work, which is worth knowing when reading a finding about org design.

The mechanism

The layer came out for a reason that made sense on paper.

The valuable part of that job was invisible on paper

A job description says coordinates, reports, aligns. It does not say catches badly framed requests before anybody spends a month on them, because nobody ever wrote that down and the person doing it would have struggled to describe it. So when the layer gets costed, what shows up is the part a machine can do, and the decision looks obvious.

Removing the friction removed the catch

Two years ago a vague request sat in a queue, and while it sat there somebody thought about it. The waiting was not the point and it was where the thinking happened. Now the request is answered before the person who asked it has stood up, and the answer is confident and formatted. There is no longer a gap in which to notice anything.

The delay was doing something. Nobody designed it to, and it was still the only quality check some requests ever got. Paul Musters

Nobody was given the job afterwards

This is the ordinary one. The layer went, and the translating was not assigned to anybody, because it had never been written down as a task. It partly lands on executives who are already stretched, seventy-two in a hundred of them past their capabilities in the American data,1 and partly it stops happening at all.

Where does the company sit on this?

Three minutes, anonymous. Your level, the cost of staying on it, and what changes above it.

Do the Culture Level scan

The test

Take one thing you had built last month and ask three questions about the request.

Not about the output. About what was asked for. Pick something recent that got built quickly and works, because the failures already have your attention and the ones that quietly answer the wrong question do not.

1

What decision was this for?

If nobody can name a decision, you built a thing that gets looked at. That is the most common finding and it is usually the point at which somebody in the room says they had wondered about that.

2

What would somebody who has been here five years have asked?

Ask a person who fits that description, and ask them about the request rather than the result. Half the time you get the seasonality problem, or its equivalent, in under a minute.

3

Who could have sent the framing back?

Sending the framing back is a different act from refusing the work, and it needs somebody with the standing to do it. If the honest answer is nobody, you have located the gap, and it is the same gap for every request arriving next month.

Run it on three things rather than one. A single example is a story somebody can explain away, and three is a pattern that changes how the next request gets handled.

One thing to avoid. Do not do this on something that failed. The room will start allocating blame and stop looking at the request, and the request is the whole subject.

On the ladder

Somebody in your company is already doing this translating. Probably not the person you would name.

In every company I have looked at, two or three people are quietly still rewriting requests before anything gets built. They are rarely the most senior, they never have it in their job description, and they are usually invisible until they leave.

An Operating Profile in use. Personality type and AI level in one profile, with the agents that fit it.

The profile holds two things: how somebody thinks and works, and how far along they are with AI. Translating is a recognisable combination. Somebody who reads a request and instinctively asks what it is for, who is comfortable enough with the tools to know what is easy and what is expensive, and who has been around long enough to hold the context. Those three together are rarer than any one of them.

Once you can see who those people are, the fix stops being structural. You do not rebuild a layer. You give three named people ten minutes on every significant request, and you say out loud that changing a request is their job rather than an obstruction.

About emaho

emaho measures one Operating Profile per person: personality type and AI level in a single profile. On that we build a personal set of AI agents that fit how that person works, inside the tools they already use. Fifteen minutes to complete, first profile free, built for companies between 20 and 500 people.

See your own profile, free Find your AI Culture Level

Fifteen minutes per person. No credit card, no strings.

What actually helps

Put the question back in the room.

Two habits, both small, and neither of them requires a new role or a layer coming back. What they do is reinstate the moment that used to happen by accident.

One line at the top of every request

Anything anybody asks to be built carries a sentence saying what decision it is for and who will make it. That single line does most of the work, because a request that cannot be finished in that form is a request that needed a conversation, and now you find that out before the building rather than after.

Expect resistance to be mild and the effect to be immediate. Roughly a third of requests in the first month will stall at that sentence, and every one of those is a month of work you did not spend.

Ten minutes with somebody who has been here a while

Name two or three people whose job on any significant request is to read it before the work starts and say what it is missing. Give them a real slot and say publicly that changing a request is the job rather than an obstruction, because the thing that stops people doing this is the fear of looking like a blocker.

Pick them on instinct rather than on seniority. You are looking for the person who reads a brief and asks what it is for, which is a habit some people have and most do not, and which has almost nothing to do with how long their job title is.

And one thing to stop. Stop treating speed of delivery as evidence that a request was good. Fast is now the default for everything, including the requests that should have been refused, so the interval between asking and receiving has stopped carrying any information at all.

What was the request actually for?

If you run the three questions on something you had built, send me the answer to the first one. What people find when they try to name the decision is usually the most interesting thing on their desk that week. One message back, and no calendar link in it.

Message me on WhatsApp

What it costs

You get exactly what you asked for, every time, forever.

The cost that arrives first is work that was done well and should not have been done. Nobody can be blamed for it, which is why it never gets counted: the request was answered, the output was correct, and the whole thing was pointed at the wrong question from the start.

Then there is where the work goes. Seventy-two in a hundred American senior executives say they are stretched beyond their capabilities.1 Some of that is the translating, arriving at the top because nobody was ever given it. That is the most expensive hour in the company doing a job that used to belong to somebody two levels down.

And there is the slower one, which is direction. Forty in a hundred employees already say they lack it.1 A strategy that only thirty-nine in a hundred recognise is arriving in a language nobody converted. Every quarter it stays that way, the distance between what you decided and what is happening gets a little wider.

Questions

Questions about the layer that went missing

What is the missing translation layer?

It is what a company loses when the people who turned a request into the right request are no longer in the middle. Their visible job was routing and reporting. Their valuable job was reading a request, recognising what the person actually needed, and handing back a better question before anything got built. Only the first of those has been replaced.

How many companies have cut management layers?

In Korn Ferry's Workforce 2025 research among more than 15,000 professionals across 15 markets, 44% of US employees reported cutbacks in manager levels at their organisation. In the same survey 40% of US employees said they feel a lack of direction at work, and 72% of US senior executives said they are stretched beyond their capabilities, against 47% of their global peers.

Why do AI tools make a thin middle layer more dangerous?

Because a badly framed request used to be slow, and the waiting was where somebody noticed. Now the request is answered before the person who asked has stood up, and the answer arrives confident and formatted. The friction that used to catch the error was a side effect of things taking time, and nobody designed it or replaced it.

How do you find out whether requests in your company still get translated?

Look at whether anybody has a written place in the process to change a request before work starts. emaho places companies on five AI Culture Levels, and a written way of working, including who reads a request first, is Level 3, Blueprint, where about one in ten organisations sits. Below that it depends on whoever happens to care that morning. The scan is six questions, three minutes and anonymous.

Do employees recognise their company's AI strategy?

Often not. Korn Ferry found that 70% of leaders globally believe they have an AI strategy while 39% of employees agree. A strategy that four in ten people recognise usually still exists and is probably sound. What has gone is the step that turned it into what a specific team should do differently this week.

What did middle managers actually do that AI cannot?

Three things, and only the third resists replacement. Routing, which is genuinely automatable. Filtering, deciding what does not need to go up, which AI can partly do. And translation, which requires somebody who knows both the request and the local reality well enough to change the question before work starts.

How do you find the people who are still doing the translating?

They are rarely the ones a leadership team names, and they never have it in a job title. emaho measures one Operating Profile per person, personality type and AI level in a single profile, and translating shows up as a combination: somebody who instinctively asks what a request is for, is comfortable enough with the tools to know what is cheap and what is expensive, and has been around long enough to hold the context. All three together is rarer than any one of them.

Are managers ready to supervise AI?

Deloitte's 2026 Global Human Capital Trends research lists this among the unresolved problems: managers are not prepared to act as supervisors of AI, and many executives lack sufficient AI literacy to contribute to oversight. So the layer is thinner and the people on both sides of the gap are not yet equipped to do the translating themselves.

Where does the work go when a management layer is removed?

Some of it upwards, into the most expensive hours in the company, and some of it nowhere. In the Korn Ferry data, 72% of US senior executives report being stretched beyond their capabilities. The translating in particular tends to arrive at the top, because it was never written down as a task and so was never reassigned when the layer went.

What does a missing translation layer cost?

Three things. Work that was done well and should not have been done, which nobody can be blamed for because the output was correct. Senior time absorbed by a job that used to sit two levels down. And direction, which 40% of US employees already say they lack, widening every quarter a strategy stays in a language nobody converted.

Where this goes next

The same gap, from another angle

Each one is what happens to work when the layer between asking and doing gets thinner.

23

Managers absorb it

Where the work goes when a layer thins, and why the people carrying it are the last to say so.

Read this one 03

No juniors to develop

The other end of the same pipeline. Translating was learned by doing, and the doing has gone.

Read this one 13

Good advice, wrong company

What a correct answer to a badly framed question looks like when it arrives confident and formatted.

Read this one

All 25 AI challenges, and what to do about each

Twenty-five things that break inside a company once people start using AI, each with the research behind it and the level where it starts to bite. This one starts to bite at Blueprint, and 7 of the others start there too.

Open the full set

From here

Who is still translating, and how long will they stay?

They exist, they are not in your org chart under that name, and fifteen minutes per person is enough to see who they are. After that this stops being a structural problem and becomes three short conversations.

Name your translators out loudThey exist, they are not in the chart under that name, and everybody knows who they are the moment you ask.
Find out how they do itTranslating is a recognisable combination of how somebody thinks and how far along they are, and a profile picks it up.
Make it a role before they leaveRight now it is a favour on top of a job, which is exactly how it walks out of the door.

Start with your own profile, free Ask me anything first Plan a 30-minute call

First profile free · no credit card · built for companies of 20 to 500 · you decide what your team gets to see

Not ready to put your team in anything yet? Start with the level of the company instead. The Culture Level scan is six questions, three minutes, and asks nothing of you.

Paul MustersFounder of emaho

Paul Musters

Fifteen years of leadership and team development in Dutch scale-ups. That practice now sits in software: one Operating Profile per person, with agents that actually fit. He writes these pages from what he runs into with clients, not from a research summary.

LinkedIn · paul@emaho.world · WhatsApp

Sources and numbers used on this page

Korn Ferry, Workforce 2025 Global Insights Report, released 17 April 2025. Annual survey of more than 15,000 professionals worldwide, from entry-level positions to CEOs, across 15 major markets including the United States, the United Kingdom, France, Germany, Brazil, the United Arab Emirates, Saudi Arabia, Australia, Japan and India. Source of: 44% of U.S. employees reporting cutbacks in manager levels at their organisation; 40% of U.S. employees saying they feel a lack of direction at work; 72% of U.S. senior executives saying they are stretched beyond their capabilities against 47% of their global peers; and, globally, 70% of leaders believing they have an AI strategy while only 39% of employees agree. The press release phrases that last figure as “more than three-quarters (70%)”, which is internally inconsistent; the 70% is used here. The report also finds 48% of employees globally concerned their job will be replaced by AI within three years, and 59% among technology industry employees. The sharpest figures quoted here are American and the global equivalents are consistently milder. Korn Ferry sells organisation design and talent consulting.
Deloitte, 2026 Global Human Capital Trends, chapter 3, AI and the future of human decision-making, by David Mallon, Julie Duda, Stefano Besana and Maya Bodan, published 4 March 2026. Research conducted with Oxford Economics among more than 9,000 business and human resources leaders in 89 countries, with separate worker, manager and executive surveys and more than 50 interviews. Source of: the finding that managers are not prepared to act as supervisors of AI and that many executives lack sufficient AI literacy to contribute to oversight, both listed among the unresolved challenges of AI-enabled decision-making.

Numbers are quoted as published. The February incident, the three scenes and the quotations attributed to Paul are composites drawn from client situations rather than transcripts, and the company described is not a real one.

Composite, assembled from more than one company.