Every workforce plan I have ever reviewed starts from the same place. Org charts. Job families. Headcount ratios by function. Span of control benchmarks lifted from industry surveys.

‍

These are not unreasonable things to look at. The problem is that they are outputs. They are the residue of how work got organised historically, not a map of what work actually is. And when you try to plan the future of work from a picture of how the past got structured, you are navigating forward by looking in the rear-view mirror.

This is not a new failure mode. It has been quietly distorting workforce planning for years. But AI has made it urgent — and visible.

‍

When generative AI arrived at scale, most organisations asked the same question: how many jobs will this eliminate? Boards wanted numbers. CHRO presentations started featuring replacement percentages. Consultancies published league tables of at-risk roles.

‍

I understand why. Headcount is legible. It sits in a spreadsheet. It has a budget attached to it.

But the question is still wrong.

‍

AI does not eliminate job titles in the way that question implies. What AI does — what any disruptive force does — is redistribute the activities that sit inside those job titles. Sometimes it automates a handful of tasks and barely touches the role. Sometimes it collapses six months of analyst work into an afternoon. The container stays the same. The contents are completely different.

‍

A "financial analyst" in 2024 and a "financial analyst" in 2026 may share a job title, a grade, and a salary band. They may be doing fundamentally different work. Your headcount plan will not tell you that. Your org chart will not tell you that. Only an activity-level decomposition will tell you that.

‍

Here is the discipline that is missing from most workforce planning processes: mapping the work itself.

Not the org structure. Not the job architecture. The actual activities that create value — broken down with enough granularity to ask meaningful questions about them.

‍

When you operate at the activity level, different questions become possible. Which of these activities require human judgement, and which are essentially pattern recognition running on historical data? Which depend on relationships that cannot be digitised? Which are already partially automated but nobody has formally acknowledged it? Which are duplicated across teams because the org structure never reflected how value actually flows?

‍

These are not HR questions. They are strategy questions. And most organisations have never had a systematic way to answer them.

‍

The activity-level view also changes what you do with AI.

‍

Instead of asking "will AI replace the financial analyst?" — which is an unanswerable question because financial analyst is a container with thirty different activities inside it — you ask: which of those thirty activities is AI capable of performing, at what quality threshold, under what governance conditions? Now you have something you can model. Now your workforce plan is not a guess dressed up as a forecast.

‍

This is not a hypothetical reframe. Organisations that have done this work find that the AI impact on any given role is rarely uniform and often surprising. Activities you expected AI to handle turn out to require more human oversight than anticipated. Activities that looked core to a role turn out to be automatable with modest effort. The activity map is where the actual intelligence lives.

‍

This is the starting point for The Work [re]design Framework — the methodology I use with organisations navigating exactly these questions.

‍

It begins, always, with mapping the critical value streams in the business. Then decomposing those value streams into discrete activities. Only then does it make sense to model the forces of change acting on each activity — of which AI is one, but not the only one. And only after that do you design the optimal human-machine-organisation configuration, before finally deriving what roles, skills, and headcount you actually need.

That sequence matters. Most workforce planning processes run it in reverse.

‍

The question that changes everything is not "how many jobs will AI eliminate?" It is: "What is the work, at the activity level, and what is acting on it?"

‍

Start from there, and your workforce plan stops being a political document that makes everyone nervous and starts being a strategic instrument that tells you something true.

‍

Everything I write here is about getting to that starting point — and what you build from it.

‍

Next issue: AI is only one of five forces redesigning work. Planning for it in isolation is a category error.