The strange thing about AI at work is that optimists and doomers keep reaching for movie plots.
In one version, the machine politely helps with email and then sits in the corner like a very expensive stapler.
In the other, it storms the building, confiscates your laptop, and leaves you explaining your transferable skills to a recruiter named Brad.
Actual work is less cinematic and considerably more irritating.
In many knowledge jobs, AI does not immediately replace the human. It drags the human one level up the stack. You stop being the person who produces the first draft of everything. You become the person who decides what should exist, what good looks like, what can be trusted, and what is allowed to ship.
That is why I keep returning to the same line: AI did not take your job. It promoted you.
Not to “manager,” exactly. More to operator, editor, reviewer, systems designer, and person who still gets blamed when the output is wrong. So yes, a promotion. But in the traditional corporate sense, where the responsibility arrives before anyone remembers the pay band.
I find this framing more useful than the usual replacement panic because it resembles what is already happening inside real workflows. The ILO and NASK’s 2025 global exposure index estimates that 25 percent of global employment is in occupations potentially exposed to generative AI, with greater exposure in high-income countries, and argues that transformation rather than replacement is the more likely outcome. Full job automation, in their framing, remains limited because many tasks still require human involvement.[1]
That does not mean everyone is safe. It certainly does not mean everyone receives a pleasant strategic role, a larger title, and a commemorative mug.
It means “job” is the wrong unit of analysis.
A job is a lumpy bag of tasks. AI opens the bag and starts sorting.
Some tasks become cheaper. Some disappear. Some turn into review work. Some move upstream into planning and constraint setting. Others move downstream into verification, integration, exception handling, and accountability. The surviving human role is usually less about creating every line from scratch and more about governing the system that now produces much of the raw material.
That is the promotion.
The wrong debate is replacement
The wrong debate is “Will AI replace software engineers?” or “Will AI replace marketers?” or “Will AI replace analysts?”
The useful question is: which parts of those jobs are moving from direct execution to supervision?
This sounds obvious. Teams still miss it with impressive consistency.
For a developer, draft code, boilerplate tests, migrations, documentation stubs, and first-pass debugging suggestions become cheaper. For a product manager, turning a fuzzy heap of stakeholder thoughts into the first draft of a PRD becomes cheaper. In support, standard replies and retrieval-heavy responses become cheaper. For an operator, summarising a process, comparing options, or converting a meeting-shaped disaster into action items becomes cheaper.
Cheaper does not mean solved. It means scarcity has moved.
The bottleneck has shifted upstream.
When a model can produce ten respectable starting points in two minutes, typing speed is no longer the expensive part. The expensive part is framing the problem, supplying the right context, defining the constraints, identifying hidden failure modes, and deciding which of those ten drafts deserves to enter the world.
That is why work starts to feel managerial even when your title remains stubbornly unmoved.
OECD analysis points in the same direction. In occupations with high AI exposure, vacancies disproportionately demand management and business-process skills, alongside social, emotional, and digital skills. Across ten OECD countries, 72 percent of vacancies in high-exposure occupations demanded at least one management skill, while 67 percent demanded at least one business-process skill.[2] My reading is straightforward: a growing share of white-collar work now contains managerial motions even when nobody has had the courtesy to call them that. You specify work, route work, review work, and coordinate outputs across humans and systems.
Again, this is not glamorous. It is merely accurate.
It also helps explain why AI is reaching white-collar, highly digitised work first. OECD research on worker exposure says the occupations most exposed to AI are typically white-collar roles such as IT professionals, managers, and science and engineering professionals, while manual occupations generally have lower AI exposure.[3] For the people reading this blog — developers, leads, PMs, founders, CTOs — that matters. The ladder wobbling beneath you is not an abstract factory story. It is your ladder.
And yes, clerical work faces the harshest direct exposure in the ILO data.[1] Some jobs are under much more immediate pressure. Some task layers will be shaved away so aggressively that the “promotion” feels less like advancement and more like being appointed supervisor of your own replacement. I am not pretending otherwise.
I am saying that the dominant early pattern in knowledge work is not clean replacement.
It is messy reallocation.
Your new job description is not “use AI more”
This is where most teams become embarrassingly vague.
They tell people to “use AI” as though that were a workflow rather than a cry for help delivered through a company-wide memo.
What matters is whether the job has been redesigned around cheaper first drafts and more expensive judgment.
The new job description usually looks something like this:
You define the artifact before it exists.
You specify what good looks like.
You provide examples, edge cases, and constraints.
You decide what the model may do and what it must not touch.
You evaluate the output.
You integrate it into a larger system.
You own the result anyway.
That is not “prompting.” That is operating.
I notice this repeatedly in my own work. I begin fewer things from a blank page now. I begin with a brief, a list of likely failure modes, and some form of verification. The keyboard time has gone down. The responsibility absolutely has not.
This is why I keep arguing that acceptance criteria are becoming the real leverage point. If generation is cheap, specification matters more. A bad brief no longer produces slow bad work. It produces fast bad work.
That is worse.
Garbage at machine speed remains garbage. It merely arrives promptly and with better punctuation.
There is also a shift from judgment as theatre to artifact as evidence.
In the older version of white-collar work, a great deal of value was signalled socially. Could you sound intelligent in a meeting? Could you explain the plan? Could you bluff through ambiguity with enough confidence that nobody risked a follow-up question?
AI is rude to that style of competence. It can produce plausible language by the barrel. The emphasis therefore moves toward the artifact that survives contact with reality: the shipped feature, the tested process, the memo that withstands scrutiny, the forecast with explicit assumptions, the incident write-up that genuinely explains what happened.
Frankly, that is a better game.
It is also less forgiving.
The smartest people in AI-heavy environments are not those who can make the model look magical during a demo. They are the people who can construct a harness around it: context packs, templates, examples, validations, tests, rollback paths, review rules, and clear ownership. “Harness engineering” is less seductive than “prompt wizardry,” which is unfortunate for marketing but probably excellent for civilisation.
The best AI workflow is usually slightly boring.
Why juniors get lifted and seniors get rearranged
One of the most interesting signals in the early evidence is who benefits most from these tools.
In customer support, Brynjolfsson, Li, and Raymond found that access to a generative AI assistant increased productivity by 14 percent on average, with a 34 percent improvement for novice and lower-skilled workers and minimal impact for the most experienced workers.[4] In a separate experiment involving professional writing tasks, Noy and Zhang found that ChatGPT users completed their work 40 percent faster and produced output judged 18 percent higher in quality.[5] In software development, Cui and coauthors reported that, across three field experiments involving 4,867 developers at Microsoft, Accenture, and a Fortune 100 company, access to an AI coding assistant increased completed tasks by 26.08 percent, with larger gains among less experienced developers.[6]
That is not a minor pattern.
It is the shape of the change.
AI is very effective at collapsing parts of the apprenticeship curve.
It helps people reach a passable first version sooner. It exposes lower-skilled workers to patterns that previously lived mostly inside the heads of stronger operators. It narrows some performance gaps. It makes the middle of the quality distribution fatter.
Good.
Also dangerous.
Good, because more people can perform useful work more quickly.
Dangerous, because many career ladders were built around earning your way through the routine layers. When those layers are compressed, organisations must become far more deliberate about how people acquire judgment. You cannot manufacture seniors from thin air. You certainly cannot produce them by asking juniors to rubber-stamp machine output all day like exhausted airport screeners.
So the ladder changes.
Juniors can often reach competent output faster.[4][5][6]
Mid-level workers lose some of the value that came from being the reliable person who could grind through standard work.
Seniors remain essential, but the source of their value shifts. Less of it comes from being the fastest hands on the keyboard. More comes from task decomposition, exception handling, quality judgment, system design, and teaching others where the model is likely to betray them.
This is why some experienced workers misread the moment.
They see AI helping a junior write passable code or prose and conclude that seniority has stopped mattering.
Wrong.
Expertise usually does not disappear. It moves up a layer. When the easy 60 percent becomes cheaper, the remaining 40 percent becomes the entire game.
That 40 percent is where expensive mistakes live.
The jagged frontier is why blind delegation is stupid
If AI were simply a universal multiplier, this story would be easy. You would hand it everything, leave early, and congratulate yourself on the transformation programme.
Sadly, reality continues to insist on nuance.
Dell’Acqua and coauthors’ work on what they called the “jagged technological frontier” is useful here. In the BCG field experiment, consultants using AI completed tasks inside the model’s capability frontier faster and at higher quality. A Harvard summary of the study reports more than 25 percent greater speed, more than 40 percent higher human-rated performance, and more than 12 percent more task completion on those tasks. But on harder tasks outside the frontier, consultants using AI were 19 percentage points less likely to produce the correct answer.[7]
That is the management problem.
Your promotion is not simply “use AI more.” Your promotion is deciding where AI belongs, where it requires guardrails, and where it should be asked to wait outside.
Some tasks should be delegated cleanly.
Some should be AI-assisted but checked tightly.
Some should use AI to generate options while humans retain the final reasoning.
Some should remain almost entirely human because the cost of a subtle mistake is too high.
Making that routing decision is work.
Real work.
The lazy way to use AI is to ask it for everything.
The adult way is to map the workflow, identify the high-volume and low-novelty steps, keep humans close to consequential decisions, and build a verification loop capable of catching silent failure.
This is where “preference is not performance” stops being philosophical and becomes operational.
A tool may feel magical in chat and still be mediocre inside a shipping workflow. If it creates more review burden than usable output, you do not have leverage. You have acquired a remarkably cheerful source of rework.
The metrics that matter are boring, which is precisely why they are useful: cycle time, acceptance rate, rework, escaped defects, reviewer load, and the time spent clarifying requirements after generation. If those measures do not improve, your “AI strategy” is probably an expensive first-draft machine wearing a blazer.
Promotion without a raise is still a promotion
To be clear, I am using “promotion” to describe a functional change, not moral progress.
A company can absolutely use AI to widen spans of control, compress headcount, increase output expectations, and deposit more review work on the same number of people. The ILO is explicit that policy choices and implementation paths will shape both worker retention and job quality in AI-exposed occupations.[1]
So yes, the promotion can be rude.
You can gain leverage and lose comfort at the same time.
You can move into more judgment-heavy work while also being measured more aggressively, interrupted more frequently, and expected to cover a wider surface area. This is what partial automation looks like outside the keynote presentation. It rarely arrives with a brass band. It arrives as: “Can you also oversee the AI-assisted version of this process?”
That is why half-adoption is miserable.
When management automates generation but not evaluation, workers become janitors for machine sludge. They spend their days reviewing plausible nonsense, correcting avoidable mistakes, and cleaning up drafts that should never have existed. That is not leverage. It is a new flavour of administrative burden.
When management redesigns the workflow properly, the human moves toward the work that genuinely deserves a human: ambiguous trade-offs, exceptions, prioritisation, escalation, stakeholder judgment, and quality control.
Those are not the same future.
So when leaders announce that “we are rolling out AI,” the important question is not whether the model is capable. The important question is whether the workflow surrounding the model is sane.
CTOs, in particular, should treat this as organisational design rather than software procurement.
Buying a capable model is easy.
Deciding where it belongs in the workflow, how people learn to use it, what must be tested, what gets logged, what gets escalated, and what qualifies as done — that is the difficult part.
That is management work.
Which is exactly why I think “promotion” is the right word.
A concrete workflow: shipping a feature with AI in the loop
Let me make this less abstract.
Suppose a team needs to ship a modest internal feature: a permissions update, a reporting view, perhaps a boring admin screen. The sort of task that traditionally began with somebody staring into an editor while converting caffeine into intent.
In the old workflow, a developer might spend the first part of the task translating a fuzzy request into a plan, then drafting the implementation, then remembering the tests, then writing the supporting documentation and release notes because everybody else had suddenly become very busy.
In the promoted workflow, the sequence changes.
Begin with the brief, not the code. Write down the user outcome, constraints, non-goals, edge cases, and acceptance criteria. This is not paperwork. It is the control surface.
Use AI to turn the brief into options. Ask for an implementation outline, likely risks, missing requirements, test cases, migration concerns, and rollout questions. Make the model expose its assumptions.
Let AI draft the first pass where appropriate. That may include code, tests, documentation, and the change summary.
Run the boring machines on the machine output. Use linting, unit tests, type checks, security scanning, schema-diff review, and whatever else applies.
Spend the human review budget on the genuinely risky parts. Examine business logic, failure handling, unusual permission edges, naming that affects maintainability, and whether the thing should exist in this form at all.
Use AI again for downstream packaging. Draft support macros, stakeholder updates, release notes, runbook changes, and follow-up tickets.
Notice what changed.
The human did not disappear.
The human moved.
Less blank-page drafting.
More problem framing.
More evaluation.
More integration.
More accountability.
That is the promotion, demonstrated in one small workflow.
I use the same pattern in writing. I do not begin by asking the model to “write the article.” I begin by making the argument legible to myself, defining what would make it true, and deciding which evidence deserves to remain. The machine can then help with structure, counterarguments, alternative phrasing, compression, and cleanup.
The model is useful.
The model is not the author.
The job has moved upward, not outward.
This also explains why AI-heavy teams begin to care more about reusable scaffolding. Shared prompts are fine. Shared rubrics are better. Shared evaluation checklists are better still. The moment a team can generate work cheaply, consistency stops being a pleasant extra and becomes survival equipment.
What teams should actually change
If you accept the promotion framing, several practical implications follow.
First, train people in review and specification, not merely in tool features.
Most AI training is essentially a software demonstration with improved posture. That is not enough. People need to learn how to scope tasks, express constraints, inspect outputs, and recognise failure patterns. Otherwise, you are distributing power tools and congratulating yourself because the packaging looked expensive.
Second, redesign role expectations explicitly.
Do not tell people that “everyone should use AI” and then evaluate them as though the work itself has not changed. If first drafts are now cheap, clarity, judgment, and orchestration should be rewarded more directly.
Third, instrument real workflows.
Preference is not performance. Measure cycle time, acceptance rate, escaped defects, and rework across a handful of recurring processes. If the tool helps only during demonstrations, that is not adoption. It is theatre with a subscription fee.
Fourth, protect learning loops for less experienced people.
If juniors never have to think, they will not become seniors. Let AI accelerate them, but do not allow it to replace the reasoning repetitions entirely. Ask for explanations. Rotate responsibility for defining acceptance criteria. Make people compare outputs rather than merely consume them.
Fifth, stop fetishising the prompt.
The prompt matters. But the larger gain is almost always in the surrounding system: better context, cleaner data, stronger templates, reusable checks, and clearer ownership boundaries. The best AI workflow is usually slightly boring because reliability is usually slightly boring.
None of this sounds glamorous.
That is exactly why it works.
The slightly annoying conclusion
The useful question is not whether AI can perform your job.
It is whether you have moved your value one level up before the market moves it for you.
If your value consists mainly of raw drafting, raw formatting, raw summarising, raw boilerplate coding, or raw information rearrangement, AI is very rude news. Those layers are becoming cheaper, sometimes dramatically.[4][5][6]
If your value lies in defining the work, building the harness, judging the output, spotting edge cases, aligning people around decisions, and standing behind the artifact, AI is not removing you. It is expanding your span of action.
That may still be exhausting.
It may still be unfair.
It may still arrive with exactly zero ceremonial appreciation from management.
But it is a more precise description of what is happening.
AI did not take your job.
It promoted you.
The annoying part is that promoted people are supposed to know what good looks like.
Do you?
Notes / references
International Labour Organization and NASK, “One in four jobs at risk of being transformed by GenAI, new ILO-NASK Global Index shows,” May 20, 2025, summarising Generative AI and Jobs: A Refined Global Index of Occupational Exposure (ILO Working Paper 140, 2025).
OECD, Artificial Intelligence and the Changing Demand for Skills in the Labour Market, OECD Artificial Intelligence Papers No. 14, 2024. The report finds that, in occupations with high AI exposure, 72 percent of vacancies demanded at least one management skill and 67 percent demanded at least one business-process skill across the ten-country sample.
OECD, Who Will Be the Workers Most Affected by AI?, OECD Artificial Intelligence Papers, 2024. The executive summary notes that many occupations most exposed to AI are white-collar roles such as IT professionals, managers, and science and engineering professionals.
Erik Brynjolfsson, Danielle Li, and Lindsey R. Raymond, “Generative AI at Work,” NBER Working Paper 31161, 2023.
Shakked Noy and Whitney Zhang, “Experimental Evidence on the Productivity Effects of Generative Artificial Intelligence,” Science 381, no. 6654 (2023): 187–192. DOI: 10.1126/science.adh2586.
Zheyuan (Kevin) Cui, Mert Demirer, Sonia Jaffe, Leon Musolff, Sida Peng, and Tobias Salz, “The Effects of Generative AI on High-Skilled Work: Evidence from Three Field Experiments with Software Developers,” SSRN working paper, August 20, 2025. DOI: 10.2139/ssrn.4945566.
Fabrizio Dell’Acqua, Edward McFowland III, Ethan Mollick, Hila Lifshitz-Assaf, Katherine C. Kellogg, Saran Rajendran, Lisa Krayer, Francois Candelon, and Karim R. Lakhani, “Navigating the Jagged Technological Frontier: Field Experimental Evidence of the Effects of Artificial Intelligence on Knowledge Worker Productivity and Quality,” Organization Science 37, no. 2 (2026): 403–423. For the summarised speed, quality, and task-completion figures used above, see also Harvard Business School’s official summaries of the study from September and November 2023.