The Mission Defines the Workforce

For 170 years the unit of workforce planning was a person in a box. With humans, agents and robots sharing the same work, that unit becomes the "Mission": a bounded outcome with an owner, evidence, and one budget.

We spend many hours each week with HR leaders at enterprises, and lately the same problem keeps coming up. They’re being asked to plan a workforce that no longer resembles the human one their planning tools were built for.

All these tools have a single starting point. Back in the 1850s, an engineer named Daniel McCallum drew a diagram for the New York and Erie Railroad, showing who reported to whom across five hundred miles of track. Most people call it the first org chart, and it answered the workforce coordination challenges of its time. It worked so well that organizations never stopped using it. Much of today’s planning, requisitions, and headcount review descend from that drawing. And all of it assumes the unit of workforce planning is a person in a box.

In the age of AI and agents, that assumption is coming apart, and HR is feeling it first. Close to $13 trillion in wages moves through the US economy each year, about $3 trillion of which pays for coordination and management. This includes the routing of work between people in status meetings, handoffs, approvals, and reporting. AI agents reach that layer first, before they touch anything else. McKinsey projects $3 to $5 trillion in agent-mediated AI commerce by 2030. Separately, Morgan Stanley’s forecast for physical AI puts a billion humanoid robots in the world by 2050.

Three kinds of workers are now arriving on three separate payrolls, and only one of them fits in a box on the chart.

The translation problem

When we ask HR leaders what’s hardest for them now, they tell us it’s translation more than supply. Their business objectives turn into portfolios; then the portfolios break down into projects that then become workstreams. But somewhere in that chain, there’s a disconnect with the employees. Enterprises ask for outcomes and the plan answers in headcount; the space between those two is where the value leaks out.

Now that digital capacity costs a fraction of what it did two years ago, your leadership has far more things it could fund and far more ways to do each one. What’s still scarce in companies is institutional attention: deciding what work matters, how tightly to bound it, and the proper mix of human judgment and agents that get authorized to do it.

Someone in your company is already making that call, and it’s probably not who you would guess. When a team configures an agent to handle specific tasks, they’re deciding which tasks a person no longer performs. That’s a staffing decision, made by an engineer or an operations manager working inside a vendor tool. It’s never connected into a workforce plan or a job description; not even a headcount review. Nothing is being purposely hidden, and there’s no software glitch going on. There’s just no HR discipline that catches these calls, which happen, in enterprises, up to thousands of times a quarter with no auditable record. 

Six levels, and one of them is missing

In an enterprise, work already moves through five levels:

• An objective the company must improve or protect. 
• A portfolio that allocates capital and capacity across competing investments.
• A project, which organizes a body of change with an owner and a delivery structure.
• A job to be done, which defines the work, decisions, handoffs, and controls requirements for the outcome.
• An assignment that connects a person or AI agent to authorized work and captures how it’s performed.

The level that’s missing between the project and the job is the Mission. A Mission is one “bounded outcome” the enterprise authorizes its workforce to deliver, and it includes an owner, a time-specific deadline, the evidence that counts as done, its economics, and its authority.

A project funds change, while a mission contracts for a result. In practice, the Mission record has twelve fields that must be completed before staffing starts. These include the unit of work, volume, SLA, quality threshold, the skills required, as well as the budget, risk, data class, and the date the authority runs out. Bring those together and you can answer for workforce design with discipline instead of habit.

The work decides the worker

Once the mission exists, it’s broken down into a series of underlying jobs rated on eight dimensions: judgment, empathy, accountability, repeatability, data access, physicality, novelty, and the consequence of getting it wrong. When the ratings are complete, leaders compare workforce options — human-led, agent-first, and blended — on cost, time, quality, risk, and the amount of oversight for each option.

Here's what that reordering does to headcount math. Let’s take for example, planning that produces demand for five full-time sales specialists. When looked at job by job, it turns out that four approved AI agents can cover about 40% of the activities in that role. So, this leaves 5 x 60% or three full-time employees, plus a redesigned role for the people who remain.

As a result, two requisitions never open. The named owner approves the plan and records the reasoning, and it’s that reason which is required when the choice differs from what the system recommended. This is how the decision stays defensible in a quarterly review nine months later.

The last two years of AI buying have taken something valuable away from HR leaders, which they can now get back: their judgment and say over how the workforce gets designed. When agents show up through a software purchase, the vendor designs the work by default, with no requisition, no job description, and no approval anywhere in the process. Bring them in through a Mission plan with a named owner, and the design happens on purpose, decided by the people trained to decide it.

Agents need work authorization too

Every person on your payroll works under an employment contract, a role description, credential checks, and a senior they report to. Do agents have the same equivalent? For most enterprises, agents have a login and a prompt.

Digital workers deserve the same treatment companies have given humans for more than a century, and it starts with stated authority. A bounded work permit binds every agent execution to:

• a Mission and a job
• a data scope
• permitted tools and actions
• spending limits in tokens and dollars
• escalation checkpoints
• a named sponsor
• an environment, and
• an expiry date

Before any work starts, the runtime checks the assignment, the approval, the permit, the budget, and the policy. Anything that’s prohibited is blocked by software instead of by a memo nobody reads. Where there’s a significant change in the agent’s version, scope, data, tools, or cost, the prior authority is voided until a person signs off again. A human, in the role of sponsor, can also stop a mission, pull an assignment, or cancel a permit outright, with the reason and time recorded.

Signal Labs recently modeled a scenario involving a health plan, where the Mission focused on resolving health provider inquiries in under two minutes. The related permit ran at $0.65 per case against a max of 250,000 cases, remaining valid for 90 days. It called for a human checkpoint any time health plan-provider contract language was ambiguous or the agent's confidence dropped below 0.92. Three candidate agents “auditioned” against the same workload and scored 76, 84, and 93. Only the last one became eligible for the bounded work permit.

None of this slows down work, any more than running payroll slows down employment. What it does is make a blended workforce of humans and agents answerable. When an executive, an auditor, or a regulator asks who authorized a piece of machine work, under what limits, and who owned the result, the answer exists in one record rather than in an engineer's memory.

One ledger, and a memory that outlasts the quarter

Making a blended workforce answerable involves two records and a single ledger. Human effort and agent usage keep their own units, hours on one side and tokens on the other, while both roll up to the same mission and project. The aforementioned health plan Mission totaled 1,840 human hours and 6.2 million AI tokens, with quality at 98.4 percent against a floor of 98. All of it read off one screen.

Corrections create adjustment events instead of overwriting the original, which is what makes the number trustworthy six months out.

The second record adds and never deletes. It holds the reasoning behind the workforce plan, the bounded permit that authorized the work, the evidence the work produced and the outcome someone accepted, as well as the gap between what was planned and what happened. Nobody using the product can remove an entry from it. That one constraint changes what a quarterly review is, turning it into a reading of the record instead of an argument about whose version is right.

I want to be clear about this point: the discipline does not optimize itself. It will not simulate a whole portfolio, rebalance capacity on its own, or move capital without a human decision. Open marketplaces where agents are bought and contracted without anyone signing are out of scope as well, and so is any robot doing physical work. What an enterprise can put in place first is the record, and everything else waits on it.

Why we anchored this in talent infrastructure

A record is only worth as much as the history behind it, which is why Signal Labs acquired ATS leader BrassRing in July. BrassRing has spent 25 years running hiring and onboarding for Fortune 500 enterprises across 172 countries and 42 languages, and it holds decades of outcome-tagged talent decisions: who was hired, who advanced, who stayed, and what happened next. No model generates that from training data. This allows recruiting to stay where it is for your team, where the change happens in the layer above. These are outcomes as evidence leaders can act on: which open roles are drifting toward a stall, which internal people look like the ones who became your best leaders, and where the next Mission's human capacity should come from.

When you read talent history that way, it changes what an applicant tracking system is for, and the change carries a cost worth naming. If the category is moving from time-to-hire toward time-to-outcome, the ATS becomes one component in a broader system for work rather than merely the system of record for workforce planning. Anyone whose scorecard runs on requisition throughput will find that uncomfortable. It is also the trade that puts work design back under HR's authority instead of IT's.

What to ask before the next planning cycle

Your 2027 workforce plan is being drafted somewhere in your organization right now, almost certainly rooted in headcount. Here’s five questions that HR leaders should be asking themsleves, to determine whether it is a plan for the workforce they actually have:

• Which outcomes is the plan contracted to deliver, and could you trace a single role back to one of them?
• Before a requisition opens, has the work been rated for judgment, data access, and the consequence of an error, or does last year's org chart decide it?
• Which individual is accountable, by name, when an agent and a person share a piece of work and it fails?
• Where do wages and token spend appear in a single view, so the cost of a blended team can be read off a page instead of assembled by hand?
• And when this year's decisions are reviewed down the road, will anything retain what worked, so next year's plan starts from evidence rather than memory?

McCallum's chart answered the coordination question of the railroad age, and it lasted 170 years because its planning unit stayed stable. Today, that unit has changed. Companies that answer those five questions first will spend 2027 deciding what work is worth doing, while the ones that don't will spend it explaining last year's headcount.