XO Docs
The Future of WorkPhase 1: Agentic Workforce

Overview

Businesses hire agents as workers. Skill unbundles from time, and output stops being capped by headcount.

Phase 01
Agentic Workforce

Businesses hire agents as workers, not copilots. Skill unbundles from time, and output stops being capped by headcount.

The shift

For all of history, buying someone's skill meant buying their time. Agents break that link.

The old deal
Brain
judgment, framing
Skill
practiced craft
Bought together, rationed by time
The unbundling
Brain stays human
scarce, valuable
Skill becomes runtime
on demand, parallel
Bought separately, at marginal cost

Two things change here. First, businesses stop renting a person's hours to get their skill and start renting the skill itself, as an agent: available on demand, in parallel, at any hour, for marginal cost. You hire the coding, the research, or the analysis directly, without hiring the forty-hour week it used to come bundled with.

Second, the unit of work changes. Today you prompt a tool and stitch the pieces together yourself, so the prompt is the unit of work. As agents take on whole jobs, the prompt stops being the unit and the work becomes the unit: you hand over an outcome to own, not an instruction to run. The brain that decides what is worth doing stays human; the skill that executes it becomes runtime.

What a business wants

Businesses do not want to manage tokens or prompts. They want to state intent and get an outcome, the same way they delegate to a team today, only faster and cheaper.

Today
Intent
A team or agency
Outcome
weeksfixed headcountbilled by the hour
With XO agents
Intent
A fleet of agents
Outcome
hourson demandper unit of work

A business does not buy effort, it buys outcomes, and it wants them at a fixed, predictable cost. Tokens and prompts are the wrong unit for that: they price the machinery rather than the result, and they swing with how verbose a model happens to be. So as the work itself becomes the unit, pricing follows. Agents get charged by the unit of work delivered, a resolved ticket, a reviewed contract, a shipped report, the way you scope a contractor, not the way you read a utility meter. That gives a business the efficiency and predictability it already expects from every other line item. XO does not provide the agents or the models; you bring your own. XO only meters the token spend against each outcome so the price stays tied to the result, not the machinery.

What is a unit of work

A unit of work is a complete, verifiable outcome an agent owns end to end, not a step along the way.

Resolved ticket
A support issue triaged, answered, and closed.
Reviewed contract
A document read, risks flagged, redlines proposed.
Shipped report
Data pulled, analyzed, written up, delivered.
Merged change
A fix written, tested, reviewed, and merged.

A prompt is an instruction and a token is a billing artifact, but a unit of work is the job itself. What makes something a unit of work is three things: a clear definition of done, a result you can actually check, and a scope small enough for one owner to be accountable for. If you can describe what finished and correct looks like, you can hand it to an agent, verify it when it comes back, and price it as one item. That is the unit businesses plan, budget, and pay against, which is why it becomes the natural thing to charge for.

Definition of done
You can describe what finished and correct looks like.
Verifiable result
You can check it when it comes back.
A single owner
Scoped small enough for one party to be accountable.

How we calculate a unit of work

It needs no new accounting. We measure it the way we judge work today: did the state change, and what did it cost?

01
Create an action item
Name the outcome and the end state you want.
02
Assign a budget
Set what reaching that state is worth.
03
The agent works
It runs against the item until the state is reached.
04
Compare and reconcile
Check the state changed, then budget against AI spend.

The trick is that we simply compare the state, the same way a manager does today. You create an action item, describe the end state you want, and assign it a budget, exactly like scoping a ticket or briefing a contractor. The agent runs against it. When it reports done, there are two comparisons: the state, did the world actually change the way you asked, the ticket closed, the contract reviewed, the report delivered; and the cost, what you budgeted versus what the AI actually spent in tokens to get there. The budget is the value of the outcome, and the tokens are where the AI spends, on the model you bring. XO does not run the model, it only enables this tracking, and the gap between budget and spend is your efficiency. That is the whole calculation.

This lives inside the product. Each XO workspace can hold many sessions, and you open a separate session per intent: one to clear the support backlog, one to review contracts, one for the weekly report. Every session comes with the same machinery enabled by default, the action item, the budget, the state check, and a meter on what the AI spends. Because the boundary is the session and the session is the intent, cost and outcome are tracked per intent instead of smeared across a single token bill. That is what lets you price on intent and outcome rather than tokens.

Did the state change?
State before
State after
Compare the two, the same check a manager makes today. If the world moved the way you asked, the work is done.
What did it cost?
Budget (the price)
AI spend
efficiency
The budget is what the outcome is worth. The tokens are where the AI spends, on the model you bring. XO only enables the tracking; the gap is your efficiency.

Who does what

Agents
  • Execute the skill
  • Work in parallel, any hour
  • Produce the first pass at marginal cost
Humans
  • Frame the problem
  • Decide what good looks like
  • Own the outcome

In plain terms

Phase 1 is the near-term shift already underway. For all of history, hiring a skill meant renting a person's hours, so expert work was capped by how many people you could afford and how many hours were in their day. Agents package the skill without the hours: a coding agent, a research agent, or an ops agent delivers practiced craft on demand, in parallel, at marginal cost. The work that was never economical to staff, the analysis a small business skipped or the research a solo founder could not justify, becomes affordable. That is why this expands the amount of knowledge work rather than shrinking it.

What does not change is the need for a human in the loop. Someone still has to decide which problem is worth solving, what a good result looks like, and who is accountable when it ships. Phase 1 moves people from doing the work to directing it, and XO's job is to make an agent safe to direct in production: a known identity, a spending limit, a full record of what it did, and the option to run it inside your own cloud.

The honest signal that this is real, rather than a demo, is how often a human has to step in to fix or redo an agent's output. When that intervention rate keeps falling, agents are genuinely doing the work. When it stays high, they are still just tools with a chat box.

On this page