Disclosure: I'm an advisor to Reejig. This post builds on a conversation we had with Reejig CEO, Siobhan Savage, on The Meg & Amy Show.

Try writing the job description.

You want someone who knows how the work in a function actually runs. Not the process map — the real thing, including the workaround everybody uses. You also want someone who can hold a work architecture in their head: the taxonomy of roles and tasks and skills that keeps a global company compliant, payable, and legally defensible. And you want someone with honest judgment about what an AI agent can do right now, at what cost, at what reliability, inside a regulated process.

Post that req. See who applies. See who hits the mark.

That job description is where work redesign is actually stuck. At this point, we know we need to redesign work, but who actually does it — where do they sit, how many of them do you need, and what are they on the hook for?

The pressure is real, and it's coming from the board. And the usual relief valve is broken; there's genuine fatigue with consultants who run a bunch of workshops and accomplish nothing. 

The data finally exists. This is the genuinely new part. You can now see work at the task and sub-task level, compare what a task costs when a person does it against what it costs when an agent does it, and see where capacity could actually come from.

And there is no structure to receive any of it. Pressure plus data plus no operating model is how you get a bunch of pilots and/or analysis paralysis. 

The truth is: before you can redesign the work, you have to design the work of redesigning the work. Yes, it's totally recursive: there's no job architecture for the people who fix your job architecture.

On the show, Siobhan named three patterns from across her customer base — and unlike most of what's being said about AI and work right now, these come with some evidence attached.

It’s about Workflow, not agents. The unit organizations care about now is the workflow, not the agent. Technical teams are already swapping models in and out based on cost and performance. The agent is a component. The workflow is the thing that has to be designed. So we know the target.

This isn’t a short-term rental. You cannot outsource this to a vendor or a consultancy, because it is not a project. It's a permanent, iterative change to how work gets done. Siobhan’s customers are all building internal AI work design teams, and they're overwhelmingly landing under the CHRO. It’s the new HR center of excellence.

There are key roles that have emerged. 

Two of those roles are solved. One isn't.

Go back to that job description. The reason nobody can fill it isn't that the market is tight. It's that we've collapsed at least six different jobs into one title.

Consider what a work designer actually has to do, end to end:

Diagnose. Read the data and pick which workflows are worth redesigning. This is analytical work, with big picture business context.

Get to the real work. Extract actual current state from the people doing it, including the undocumented workarounds no one admits on the first pass. This is interviewing, observation, facilitation. 

Split the work between humans and agents. Decide what moves, what stays, where the handoffs sit, what happens when it fails. The hardest call is where a human genuinely has to stay in the loop — speed on one side, governance on the other. This is a genuinely new craft.

Spec it for the builder. Boundaries, acceptance criteria, a definition of "working." That's product management.

Land it. The SOP, the enablement, the actual behaviour change on the floor. Skip this and everything above it is a write-off. This is change management and human care.

Carry the job consequence. Work out what the redesign does to roles and org structures, and build the case for whoever decides. This is planning, relationships and business case development.

Look at that list again. These are not six facets of one temperament. Getting to the real work and landing it are people-facing and emotionally intelligent. Splitting the work and speccing it are systems-thinking. Diagnosing is analytical. Carrying the job consequence is political. Asking for all of it in one person is asking for a unicorn, and unicorns don't scale to the fifty workflows you need to redesign next year.

The unit of capability here is not an individual. It's a dynamic team that connects into a single business owner. 

The good news is that it is possible to assemble the team(s) with people already in your company or in the market.

Process excellence, Lean, business operations. The people in the enterprise who have professionally practiced work design as a discipline. They have done process mapping, they have done redesign. I am not hearing anyone talk about this talent pool and I don't understand why.

Product managers. The AI work designer role is internal-facing product management. Discovery, spec, working with builders, adoption, measurement. Same loop. If you have PMs who are good at discovery, you have work designers who don't know it yet. You may just have these folks in your HRIT or HRIS department.

UX researchers. Their entire craft is figuring out how work and journeys actually happen versus how the org thinks they happen. These folks are findable - outside your organization, if not inside.

HR business partners. Strategic HRBPs hold the political map of a function — who decides, who blocks, what's already been tried. There is absolutely a place for them on this dynamic team.

Your own SMEs, on rotation. The people who know how the invoice actually gets paid, borrowed for a tour and taught the method. Pair them with a UX researcher. Nobody can reliably narrate their own work. 

AI and solutions engineering. They understand what a model can actually do reliably, at what cost, and where it falls over. They are also the mirror image of the problem. They know exactly what an agent can do and nothing at all about how the work really runs.

The magic is in the combination. Each of these people brings a piece of the puzzle, but some of the pieces will still be missing and need to be invented (like figuring out the human-agent split). And they'll have to learn together, because no one of them arrives with the whole picture — the business, the work, the technology, and the people it lands on.

There really isn’t a curriculum for this (yet) and even if there was, I wouldn’t start there. Every transformation I've been part of that tried to build capability through training produced a lot of completed courses and not much of an outcome. 

You build a capability that doesn't exist yet the way people always have. You do the work, badly at first, preferably with collaborative colleagues who learn fast.

That said, we don’t have endless time and resources, so some structure is necessary. In particular, some clear goals and parameters.

There's precedent for that combination, and it's worth a short detour into Six Sigma. Belts are earned by completing projects, not by taking certification classes.

Real work, with a business owner who wants it. Not a sandbox, not a demo, not a workflow the COE picked because it was convenient. Someone in the business has to actually want this one changed.

Bounded scope and a clock. One workflow, one function, a fixed window. Small enough to finish, big enough that finishing matters.

A finish line agreed before you start — and it isn't just "redesigned." It's adopted, in use, and measured. Which means capturing the baseline before you begin. 

Someone watching who isn't doing. Early on, nobody has done one of these, so that's a peer with a second set of eyes and a shared willingness to say this isn't working. By the third or fourth, you have people who have been through the whole loop — and that's when it becomes real apprenticeship.

Verification by someone who isn't them. An independent read on whether the result is actually real. This is the mechanism that keeps the whole thing from becoming theater. Six Sigma understood this better than most — it's what makes the craft verifiable rather than self-declared.

Two before they're on their own. The first one is adrenaline and luck. The second is where the method sticks.

That's the training program. It's a job, with guardrails.

All three of these seem like they can wait. None of them can.

Pair constantly, but expect the pairs to change shape. Researcher with SME, work-knower with model-knower, designer with builder, HRBP with business leader. The constant isn't any particular pair — it's that nobody does any part of this alone. Which is where I'd part company with Six Sigma, where the certified individual was the unit and the team assembled around them. That worked because the method was written down. Ours isn't yet. It lives in the space between a few people working it out, so the team is the thing you're building.

Rotate the borrowed members. Hold the core. Your SMEs are on loan and should go back — a workflow or two, not a permanent posting — and every one of them who returns becomes an informed customer for the next redesign. That's how you get at scarce domain knowledge without trying to own it. But don't rotate the people carrying the craft, at least not yet. That's the group the capability is accumulating in, and while the method is still being invented, churning them resets you to zero. Once there's actually something to teach, rotation through the core becomes a way to spread it. Not before. And one thing to plan for now rather than discover later: within two years, these will be the most poachable people in your company.

One boundary worth naming. Designing a job change and having the authority to make one are different things. Combining roles, moving headcount, changing who signs off — that's comp, works councils, employment law, and in some places real liability. No work designer should be carrying that, but somebody has to. They need to be named and in the game before the team produces its first redesign.

Go back to that job description one more time. You still can't fill it. The difference is that now you know why, and you know what to do instead.

But I opened with four questions and I've only really answered two of them.

How many of these people do you need? I don't know, and nobody I've asked does either. The complication is that redesigns don't finish; the work keeps changing as new opportunities keep emerging. Start small and iterate but recognize that the future you are building for is not a one-and-done project but an ongoing reinvention muscle. And, it’s going to need to spread (and quickly).

And what are they on the hook for? Also unresolved. "Workflows redesigned" is a bad number — it rewards volume and says nothing about whether anyone works differently afterward. "Hours unlocked" is worse, because it measures the supply of capacity and stays silent on where the capacity actually went. I have opinions forming. I don't have an answer.

So my bet here is deliberately a small one. Don't start by designing the organization — the COE, the headcount plan, the reporting lines. Start by assembling enough of a team to get one workflow all the way through, with a business owner who genuinely wants it changed. Run it end to end — see it, redesign it, build it, land it, measure it. Then build the operating model out of what broke, because something will.

We go deeper on all three patterns with Siobhan Savage on The Meg & Amy Show. If you're building one of these pods, tell us in the comments: what did you try first?

Prepared by Amy Wilson, former tech executive and current product strategy advisor. For more insights on leadership and the future of work, subscribe to The Meg and Amy Show.