Your workforce already includes non-human workers. None of them show up on your org chart or workforce plan.

Somewhere in your company this week, work got done by something that isn't a person, isn’t on the payroll, and doesn’t show up in an employee directory.  For years, that’s been happening in the manufacturing space with the ever-increasing pace of industrial automation, but now it’s happening in the arena of “knowledge work.”

Maybe it was an agent drafting a first-pass at contracts or an RFP response. Maybe a model quietly triaging a support queue that used to take a team of twelve. Or maybe – if you've been watching the last couple of weeks – it was Claude, sitting inside a Slack channel, getting tagged by name like any other colleague and handed a task to run while everyone else moved on to something else.

With the launch of Claude Tag, the thought experiment of a “generalist, AI teammate” just went from the world of sci-fi to a v1 implementation.

The teammate with no employee ID

Anthropic just shipped the next wave of AI disruption, and I have seen very little about it across my HR / SWP network, but I think it’s big. You add Claude to your Slack workspace, give it access to the channels and tools you choose, and from then on anyone on the team @-mentions it the same way they'd loop in a coworker. It keeps context across the channel, picks up where the last person left off, and can work a task in the background for hours or days. It can even proactively alert you when someone else contributes an update that is relevant to your workstream or interest areas. Anthropic says its own product team now generates something like two-thirds of its code through an internal version of Claude Tag.

Set the number aside for a second. Every company will drive a different value stream from this tech.  The point is the overall shape of it: a generalized teammate, with a name, a memory, and a queue of work, interacting independently across time and across diverse workstreams and your HRIS and your HR team doesn’t know it exists or what it’s working or what recommendations it’s making or how well it’s performing.  So much for the HRIS as the “system of record...”  

And it isn't just the emergence of AI generalists. Specialized agents are multiplying even faster – one for reconciliation, one for sourcing, one for interviews, one for pre-op medical conversations, one for ticket triage, one for first-draft analysis – each quietly absorbing tasks that used to live inside a human role. Some you bought. Some were stood up on a corporate card.  Some live inside your enterprise or specialized software.  Some were built by your IT team.  And chances are, nobody in HR has ever seen any of them nor been involved in deployment, HR impact analysis, workflow discussions, hiring implications…

Two years ago, an emerging question was “What is the workforce we’re going to need over the next three to five years to meet our strategic objectives as a business?”  Now a new more fundamental question is emerging:  “What is a workforce? And how do we account for the non-human actors now doing meaningful work?”

You're already running a hybrid workforce. But you’re only planning for the part that is human.

Some of you, most of you, are already running a hybrid workforce – humans and non-humans, side by side, doing the work. Most of your are only planning the human side of it. For now, given the percentage of non-human work happening, that’s probably ok, but what about the future? World Economic Forum analysis suggests that 15% of work currently being done by humans will be done by non-human actors by 2030.  McKinsey thinks up to 50% of human work will be augmented by automation and therefore require fundamental redesign of the work and workflows.

The whole agent conversation has been stuck on one question: how many people will they replace? That's an important question, but the bigger one – the one that actually changes how you run the organization – is this: "When does an agent stop being a line in the IT budget and become a unit of capacity you plan around? With a cost. A capability and skills profile. A set of responsibilities and defined tasks. A manager. A line in the workforce demand model, right next to the human roles."

The moment you take automation seriously at the task level – which most of us now claim to be doing – your traditional demand forecast quietly breaks. A model that only counts humans is structurally incomplete. It's forecasting one side of a multi-sided workforce where the non-human sides are growing at an accelerating pace.

Planning for a workforce that isn't all human

This is not "headcount, minus an automation discount." That's a starting point but it’s insufficient as an end game. Modeling a hybrid workforce means holding two kinds of capacity in the same plan, with the same rigor:

• What does each category of worker actually cost? For AI automation, it’s token costs + the human supervision + the quality analysis and rework an agent still sometimes needs. For industrial automation, it might mean hardware + maintenance costs + human supervision + repair costs + depreciation. For humans, it’s salary + benefits + tax burden.

• What is each genuinely good at, and where does the handoff between them happen?

• What work should remain human because it’s linked to company differentiation, brand promise, brand identity, or highly regulated or higher-risk legal landscapes?

• Against the future productivity or outcome metrics the company’s strategy requires, what’s the optimum mix of human and non-human labor required to generate required outputs?

• Where does agentic or robotic capacity reduce human demand – and where does it create new human demand, for the people who design, supervise, and correct the bots?

• What's the right span of control for a manager whose team is half people and half agents?

• How do role change and therefore role definitions, expectations, pay etc… when automation takes over xx% of the duties and tasks of the role?

• How can you leverage excess capacity created by automation to move talent into emerging areas of need or to fill long-standing areas of talent shortage and therefore decrease dependency on the external labor market?

The management problem nobody owns

And it runs past planning into something more basic. If an agent is doing a third of a team's throughput, who manages it? Who sets its objectives, reviews its output, owns its mistakes, decides when to widen its scope or pull it back? We've spent a century building the discipline of managing human performance. We have almost nothing for managing a blended team, and "the vendor's dashboard" is not an operating model.

This is the part that should keep a CHRO up at night, and it has nothing to do with the technology. The technology works already - at least in pockets. The question is whether the organization has any structure to hold it.  

Even the concept and nomenclature of HR is now hopelessly dated. Terms like “Human Capital Management” and “Chief Human Resource Officer” – now only describe part of the role. We need new language and conceptual frames – terms like “Workforce Management” and “Chief Work Officer.”

That said, the “human” side of this matters now more than ever. Automation doesn't only change the headcount math – it challenges the people who remain, at a time when engagement is already fragile. Gallup's 2026 report puts global engagement at just 20%, costing the world economy an estimated $10 trillion a year in lost productivity – and engagement tracks strongly with business-unit productivity, profitability, and sale. That's a real bottom-line business impact, not a soft one. Layer job displacement or role redefinition on top and the strain multiplies. In a recent EY survey, 75% of employees expressed concern that AI would make some jobs obsolete; 65% feared for their own job.

That anxiety isn't just a mood: higher awareness of automation-driven replacement predicts increased job stress and lower wellbeing – a known precursor to disengagement and performance decline.

The downstream effects hit exactly the levers a workforce plan depends on. Job insecurity undermines job satisfaction, organizational commitment, and task performance – the very productivity automation is meant to unlock. It also threatens something a capacity model rarely accounts for: identity and belonging. Work supplies recognition, status, belonging, and self-concept, so a role that's hollowed out or reshuffled can register as an identity shock, not just a reassignment. Reskilling that fixes the "what" while ignoring the "who" only deepens the disconnection.

None of this is an argument against automation; it's an argument for designing the human transition with the same rigor as the capacity model. The payoff is measurable: organizations that redesign workflows around optimal human–AI task allocation, rather than simply automating, achieve business outcomes two to six times better than purely automated approaches.

Fortunately, the mitigation strategies are well understood: employment security, reskilling opportunities, and worker protections blunt the stress and job-displacement fears that otherwise erode performance. The plan that adds agents as a costed, managed line needs a human column beside it: not just how many roles change, but what happens to engagement, belonging, and trust when they do – and what we're doing about it.

If you have no idea how many agents are running inside your company, you're normal

If you're reading this thinking we couldn't tell you how many agents are running inside our four walls, let alone what they cost or who owns them – you're not behind. You're normal. Almost nobody has this wired yet.

But normal is exactly the opportunity window. The organizations that start treating agentic capacity as a planned, costed, managed part of the workforce – rather than a surprise that shows up in the cloud bill – will plan more accurately and move faster than the ones still pretending that the workforce consists of nothing but “giant sacks of mostly water.” (ST:TNG - Season 1: "Home Soil") When you can model the humans and the agents in one place, you can finally answer the question everyone is actually asking: not "will AI take these jobs," but "given what agents can now do, what is the right shape of this workforce, and how fast can we get there?"

A big part of what we've been building toward with clients is exactly this – humans and agents as members of the same workforce model, sitting on a shared foundation, so demand, cost, skills, and automation exposure account for both at once. If that's a conversation worth having, I'm always glad to walk through what we're seeing. But the software is step two. Step one is easier: open your current workforce plan and ask where the non-human part of the workforce is actually represented. For almost everyone, the honest answer today is nowhere.  And that’s exactly your opportunity – the first companies that actually start planning this way will plan smarter, invest more deliberately, and have a framework to evaluate outcomes and ROI.  Which drives competitive advantage.

The org chart is going to have non-human workers on it whether we plan for them or not. The only real choice is whether they show up as a managed line in the plan – or as a surprise in the budget.

So I'll ask: is anyone actually modeling agentic capacity in their workforce plan yet – cost, capability, ownership and all? Or is it still living over in the IT budget, where no workforce planner can see it? Genuinely curious where people are on this one.

‍