AI Workforce Transformation: How Roles, Skills and Headcount Change

On 27 August 2026, the U.S. Bureau of Labor Statistics published its employment projections for 2025 to 2035 alongside new AI exposure categories. Data scientists are projected to grow 34.6% and information security analysts 21.0%, against an all-occupation average of 3.5%. Bookkeeping and auditing clerks are projected to fall 5.6%, and CNC tool operators 9.4%. Most of the roles we charted across corporate and domain functions keep growing, and the ones that shrink share a common trait: routine tasks make up most of their workload.

We set those projections against task-level data from Etter and the Draup Talent platform to see where the change actually happens. The answer matters for anyone leading workforce transformation, because AI exposure turns out to be a property of tasks. A function label tells you surprisingly little about how a role will change.

The deck works through that change in sequence: which roles grow, which tasks automate first, how a role empties out, which skills to invest in, and how to translate all of it into a headcount plan.

Which roles grow and which shrink in the 2025 to 2035 projections

Specialist and technical roles lead the growth. Logisticians are projected to grow 17.6%, industrial machinery mechanics 17.8%, and computer and information systems managers 15.8%. Six of the roles we charted are projected to shrink: CNC tool operators by 9.4%, computer programmers by 7.3%, human resources assistants by 6.2%, bookkeeping and auditing clerks by 5.6%, network and systems administrators by 4.1% and database administrators by 0.1%.

The more useful pattern sits inside each function. In IT, information security analysts grow 21.0% while network and systems administrators shrink 4.1%. In data, data scientists grow 34.6% while database administrators hold roughly flat. In HR, training and development specialists grow 10.8% while human resources assistants shrink 6.2%. Each pair sits in the same function, and the two roles still move in opposite directions.

Why a role's task mix sets its AI exposure

Etter scores every role on an AI spectrum from 0 to 100. The score combines automation potential, the share of work AI can perform with little or no human involvement, and augmentation potential, the share where AI assists a person who stays in the loop. Legal specialists score highest in our set at 68.3, and accounts payable specialists carry the highest automation score at 49.2.

Roles in the same function separate clearly on this measure. A QA automation engineer scores 63.3 and a senior QA engineer scores 51.9. A financial analyst scores 63.8 and a senior financial and data analyst scores 51.6. A business intelligence analyst scores 44.7, and it is the one role in the set where augmentation (28.3) outweighs automation (16.4).

The task-level scores explain these gaps between roles in the same function. Data entry and processing carries the highest automation score at 44.4, followed by review and approval at 44.1 and documentation and record keeping at 40.5. These are the rules-based steps that repeat in every business process. At the other end, administrative tasks score 34.3 and code development scores 32.7. No task category scores below 32, so every role carries some exposure, and the size of that exposure depends on how much of the day goes to the high-scoring categories.

How a compressing role empties out task by task

We mapped the tasks of bookkeeping and auditing clerks, the role behind the 5.6% projected decline, into four layers. Routine posting, tallying debits and credits, and checking figures move to RPA and accounting AI. Drafting invoices, reconciling accounts and running payroll become AI-assisted work with a person checking the output. Applying tax and compliance rules, compiling budgets and resolving flagged discrepancies move up to senior finance staff and controllers. The residual work, such as banking cash, raising purchase orders and filing, is absorbed elsewhere or outsourced.

The role contracts gradually as each layer removes work from it, so the 5.6% projected decline reflects tasks moving out of the role over time. The same movement reshapes the organization above it. As coordination work automates, our models expect management layers to fall from 9 to 11 today to 6 to 7 in payments and transaction networks, from 8 to 10 to 5 to 6 in software and SaaS platforms, and from 7 to 8 to around 5 in data and analytics providers. Spans of control widen by 40% to 65% depending on the organization type, and each remaining manager oversees more people as well as more agent output.

Four investment plays for every skill

Across more than 1B global job descriptions in software, finance and HR, we plotted each core skill by its current demand and its demand growth from 2020 to 2026. The result sorts skills into four plays. Skills with high demand that keep rising, such as Kubernetes, Terraform and AI-assisted development, call for sustained investment. Skills with small talent pools and the fastest growth, such as RAG, LLM fine-tuning, AI governance and AI output validation, call for early capability building. High-volume skills that are automating away, such as journal-entry accounting and manual documentation, are the best candidates for reskilling. Small and declining skills, such as data entry, manual admin and legacy ETL and BI, call for a planned exit through attrition and redeployment.

Our projections show how quickly the mix shifts. In software and tech, AI-assisted development rises from 3% of job postings in 2022 to an estimated 45% in 2030, while manual documentation falls from 22% to 6%. In finance, AI output validation rises from 2% to 34% and journal-entry accounting falls from 28% to 7%. In HR, AI governance rises from 1% to 28% and manual HR admin falls from 26% to 6%. Judgment and human-interaction skills grow alongside them: architecture trade-offs reach 29% of software postings, business partnering 29% of finance postings and people analytics 23% of HR postings.

How to calculate the future workforce

Workforce planning needs a simple equation to work from. Start with today's base of FTE capacity, level mix and spans of control in each function. Subtract the capacity that automation realistically frees, which depends on AI adoption, workforce readiness and change management. Add the new work AI creates, such as verifying outputs, maintaining data architecture and governing data. Add the capacity that business growth requires. The result is the future workforce, sourced through Build (in-house capability, agents and skills), Buy (tools) or Borrow (partners).

In an illustrative Etter scenario with 20% annual demand growth and a balanced automation pace, total demand ends roughly 11% lower after 12 quarters while output holds. The freed capacity is redirected across Build, Buy and Borrow according to each role's future requirement.

What the healthcare model shows about attrition and headcount

We ran the same logic for a healthcare workforce over 36 months, with 10% annual demand growth and a balanced automation pace. AI frees about 24% of today's capacity. Attrition and retirement account for 34% of the starting team over the period, and replacement hiring and new capacity add 41%. Headcount ends at 83% of the starting level. Meeting the same demand without AI would take 113%.

That 30-point gap is the number to plan against. In this model, the 83% is reached without layoffs because attrition absorbs most of the change, and rehiring, reskilling and redeployment are planned against the gap from the start. The automation pace setting, from cautious to maximum, changes how quickly the shift happens. It leaves the total amount of automatable work unchanged.

Where to start a workforce transformation plan

The first step is to describe your largest roles as task mixes, using categories such as data entry, review and approval, documentation and coordination, because that is the level at which AI exposure shows up. The second is to give every core skill an explicit play: scale, build, reskill or planned exit. The third is to run the base, freed capacity, new work and growth calculation for each function and compare the result with your current hiring plan. The healthcare case suggests that a gap identified early can be closed largely through attrition, rehiring and redeployment.

I would welcome your view on which roles in your organization are changing fastest at the task level, and how your current planning model accounts for them.

‍