From Job Architecture to Work Architecture: Designing Roles for the Age of AI

When TalentNeuron opened its webinar on AI and the future of work, the room, virtually speaking, was already packed. Attendees joined from Munich, Vegas, Orlando, Dublin, Toronto, Johannesburg, and dozens of cities in between, all with some version of the same question on their minds: if AI is about to touch nearly every role in the business, where do you even start?

The session, led by David Wilkins, CEO of TalentNeuron, opened with a number that reframes the whole conversation. A study out of the University of Pennsylvania's Wharton School looked at the distribution of employment by AI automation potential across the US and found that only 1% of roles today can be 100% automated. Set that 1% aside, and roughly a third of remaining roles have essentially zero automation exposure. But in the middle, between the 50th and 90th percentile of roles, AI is already disrupting up to a third of the work, and up to 15% more once generative AI is added in. That puts some degree of AI disruption within reach of nearly 90% of roles.

It's why the questions coming from executive leadership have changed. Boards and investors have moved on from "what are our competitors doing" to two more future facing questions: where do we have the most automation potential, and what will it cost, in dollars and in the work of building new capability, to get there over the next three to five years.

Why work architecture is replacing job architecture

That shift in questions is also driving a shift in vocabulary. For years, job architecture (functions, job families, roles, levels, salary bands) has lived with Total Rewards. It's a useful map of the organization, but it was built for a world where a job was a fixed bundle of duties assigned to a person. AI breaks that assumption, and TalentNeuron's answer is a broader concept it calls work architecture: understanding not just roles, but the tasks and skills inside them, and who is doing each one, whether that's a human, a bot, or an agent.

The pace of change around automation means thinking about work "as a collection of jobs is really no longer the right level" to plan from, as Wilkins puts it. That's the thinking behind work architecture: organizations need to atomize their understanding of work down to tasks and skills, and track the connective tissue between roles, tasks, automation, capabilities, and organizational structure, because a change at any one of those levels eventually ripples through all the others.

Five steps for designing work in the AI era

TalentNeuron's framework breaks the process into five steps: leadership sets the business outcomes and the capabilities needed to reach them; those capabilities get mapped to specific roles and normalized against the labor market; automation potential is analyzed at both the job family and task level; the resulting role redesign gets built into the broader transformation plan; and that, in turn, triggers a fresh look at organizational design.

Take a data scientist role, for example: roughly 32% of it could be automated through a combination of generative AI and business process automation. But the more interesting lesson sits in a cost analysis of an insurance company's roles. It's tempting to start automating the function with the largest number of automatable tasks. Those are often also the most complex to build and the slowest to pay off. A better approach is to start with the tasks that are simplest to automate but still carry outsized headcount impact. In one case, automating roughly 25 tasks within an administrative role freed up the equivalent of 200 roles' worth of capacity, a much faster win than tackling the flashier, harder function first.

That logic played out again in a redesign example close to Wilkins' own background in HR: the HR shared services advice and guidance role. Managing employee onboarding takes up about 13% of that role's time and is roughly 32% automatable with tools like BambooHR or Workday. The first step is straightforward automation. The more interesting step is what that frees up: as more of the role's routine tasks are automated, the time saved can be reinvested into higher value redesign, from onboarding experiences built with augmented and virtual reality to a digital twin of the organization that tailors onboarding to each employee's role, learning style, and career path.

From role redesign to organizational design

Redesigning a single role rarely stays contained to that role. It raises immediate downstream questions: how will the work be distributed across the organization now that so much of it can be virtual or centralized into a hub? Does the person filling this role have the right skills for what's left after automation? Could this role be combined with another, given the capacity that's been freed up? These are the organizational design questions that role redesign inevitably triggers, and they're exactly where TalentNeuron's own model of work design is built to help, connecting market intelligence, strategy, workforce planning, and organizational structure into a single view.

Built for this: how TalentNeuron's platform gets you there

None of this works without data to normalize roles against the market and detect where automation is actually happening. TalentNeuron's platform currently includes close to 4,000 normalized roles across industries and occupations, each with associated skills, tasks, and automation potential data attached.

TalentNeuron also offers a capability that estimates automation potential for any existing job architecture an organization already has, benchmarked against TalentNeuron's own baseline data. The platform's strategic workforce planning module also includes driver models, a framework connecting the nature of a job to its expected work outcome, which enables true scenario based planning. As a simple example: if an airline plans to add flights between two cities, driver models can automatically translate that into the corresponding increase in pilots, baggage handlers, flight attendants, and mechanics, each scaling at its own ratio.

That same scenario-based thinking came up again in the Q&A, where Wilkins compared the approach to how finance teams operate: planning within a range of best case and worst-case scenarios rather than a single fixed plan. TalentNeuron's platform lets organizations build demand scenarios (What if we expand meaningfully into EMEA? What if we pursue M&A?) and supply scenarios (Where do we expect high or low attrition?), then combine them into a gap analysis, effectively a digital twin of the organization under different sets of assumptions.

As for keeping pace with how quickly the technology itself is moving, Wilkins noted that TalentNeuron tracks automation across three distinct categories, generative AI, business process automation, and industrial automation, rather than treating "AI" as a single trend, and validates its predictions against real client outcomes, including several Fortune 1000 companies.

As organizations increasingly manage a mix of humans and AI agents, leadership itself needs to evolve too. The same coaching instincts that make someone a good manager of junior employees, setting clear goals, giving iterative feedback, and correcting course when something isn't quite right, translate directly to managing AI agents. As Wilkins puts it, this may be the first time in human history that people are managing something other than other humans, which is exactly why leadership development deserves a place at the center of any AI transformation plan.

Want to hear the full conversation, including the audience Q&A? Watch the full webinar here.