To Improve, Not Prove
Watch enough of the vlogs and the pattern is hard to miss. Translators, copywriters, designers, consultants. Senior people with twenty years of practice, told their position has been eliminated. The reason cited, more often than not, is AI.
The retrenchments are real. The impact is real.
But the announcements skip the more useful question:
The position was eliminated. Was the task actually replaced?
In most cases, the answer is no. The position is gone. The task got handed to a tool the company subscribed to last quarter. The work still happens.
The Position Is Not the Job
A senior copywriter is not a copy-producing machine. They are also the person who reads a brief and says it is wrong before typing a word. The one who pushes back on legal. The one who notices that the brand voice has drifted three campaigns in a row.
A senior translator is not a translation engine with a salary. They are the one who knows which phrase will land badly in Mandarin even though it reads fine in English. The one who pauses production and says the paragraph is technically correct but commercially dangerous.
A senior consultant is not a slide generator. They are the one who redirects the workshop with a question nobody saw coming. The one who tells the client what the internal team cannot say.
The task is the visible part. The practice is invisible. Communicating with the team. Asking the right question. Criticising the output that looks fine but reads wrong. Refusing the obvious answer. None of this was ever written into the job description, but it was the reason the senior person was paid like a senior person.
When the position is removed and the task is automated, the practice is not reassigned to the tool. It evaporates.
Workflow vs. Headcount
The honest sequence for any responsible AI adoption is straightforward. Map the workflow. Identify what AI can usefully do inside it. Then decide what changes about the team.
That sequence is rare.
What we commonly see is the opposite sequence. Cut headcount. Announce the AI transformation. Figure out the workflow afterwards.
Restructuring headcount is a quarterly event. It is visible, measurable, and reduces the operating cost immediately. Restructuring workflow is a six-month project. It requires honest conversations about how work actually happens, who does what, and which parts of the process were never documented because they lived in someone’s head. It does not produce a press release.
Boards reward the first. They rarely ask about the second.
So the workflow stays unmapped. The headcount gets cut. The AI tool gets deployed into a process nobody fully understood to begin with. Six months later, the work is faster, cheaper, and noticeably worse, and no one can say exactly why.
Best Practice Is Slow
A best practice is not a document. It is an outcome.
It is what survives after years of conversations, briefs that did not land, decisions that turned out wrong, decisions that turned out right for reasons no one expected, the rejected drafts, the arguments with the client, the calls to the regulator. The best practice is the residue of all of it.
Can a best practice be autonomously executed by an AI agent? In a narrow, stable task, yes. The agent can repeat the steps. What it cannot do is generate the practice in the first place. It cannot have the conversations. It cannot sit in the rejection. It cannot be the one who learns from being wrong.
The practitioner is not just the executor of the best practice. The practitioner is the source of it.
Replace the practitioner and you can keep running the existing practice for a while. You cannot keep developing it.
What Does “Learn AI” Actually Mean?
The public response to all of this is a slogan. Learn AI.
It sounds responsible. It is repeated by governments, schools, employers, and LinkedIn. Look at what it produces in practice and the answer is more specific than the slogan admits.
Schools hand out free AI tool subscriptions and call it literacy. Staff outsource the thinking part of their jobs to a chat window and call it productivity. Governments release national chatbots and call it citizen access. Each of these is real. None of them is what the slogan implies.
What is being trained, in every case, is operators. Training everyone to operate whichever model launched this week is not improvement. The tools change every quarter. Being an operator of a moving target is not a career. It is a treadmill. And the operators will be the ones chosen to let go.
Prove or Improve
There are two reasons an organisation adopts AI. Most are not honest about which one applies.
The first is to prove. To the board, that the company is doing AI. To investors, that costs are coming down. To the market, that the company is not behind. Headcount cuts get announced before workflows are mapped. KPIs measure adoption rate, not work quality. The pilot exists mainly so leadership can mention it on stage.
Proving is a defensive posture. It treats AI as a verdict to deliver, not a capability to develop.
The second reason is to improve. Improving keeps the human accountable, the judgment in the room, and AI in service of the work. Success is measured by whether the work got better. Not by how many people were removed.
These two paths produce two very different organisations in the future.
What Is Actually Being Cut
The retrenchment vlogs are instructive. They show, in real time, what happens when leaders use a powerful technology to prove their decisiveness instead of improve their organisation.
The companies that will look strong in the future are not the ones with the leanest org charts today. They are the ones quietly building the practice of working alongside AI, with judgment intact, and with their best people still in the room.
Improve, not prove.
