Workforce
A standing team of AI agent seats that work your cloud estate on a rota — managed-service coverage priced per seat-hour instead of per person.
At a glance
| Route | /workforce, with sixteen sub-pages |
| Section | AI & Automation |
| Page permission | workforce |
| Billing | Per seat-hour, from its own balance |
What it is
Workforce lets you staff a standing team of AI agent seats — a Service Delivery Lead, a Chief Cloud Architect, SREs, database administrators, security and finance seats — that work your estate on their own rota rather than waiting to be asked.
The difference from Apex, in one sentence
Apex answers a question when someone types one. A Workforce seat comes on shift, looks at the estate, decides whether anything needs doing, and hands the work to the seat above it when it cannot finish alone.
Apex is a colleague you ask. Workforce is a team that turns up.
Who it is for
Organisations that want managed-service coverage without a managed-service contract.
It suits two people in particular:
- A platform lead who knows what should be watched but has nobody to watch it at three in the morning.
- A finance owner who wants the cost of that coverage stated per seat, per hour — rather than buried inside a retainer nobody can break down.
How it works
The catalogue: 22 roles across seven functions
You staff the team seat by seat. Every escalation eventually reaches the Service Delivery Lead, which is the one coordinating role — so there is always a single place where unfinished work surfaces.
Grades set price, capability, and cadence together
Each seat is graded 1 to 5. The grade decides three things at once: what it costs per hour, how capable a model it reasons with, and how often it comes on shift.
| Grade | Hourly rate | Reasoning capability | Baseline cadence |
|---|---|---|---|
| L1 | $0.25 | Lightest | Every 30 minutes |
| L2 | $0.50 | Light | — |
| L3 | $1.00 | Mid | — |
| L4 | $2.00 | High | — |
| L5 | $3.00 | Highest | Every 12 hours |
Higher grades think harder but look less often; lower grades look frequently but reason more simply. That is deliberate — a cheap seat checking every half hour catches obvious things fast, and an expensive seat reviewing twice a day catches subtle things well.
Cadence is a baseline, not a timetable
Some roles override the baseline because the job demands it: a NOC Analyst looks every 15 minutes; a Capacity Planner looks once a day.
Each seat then paces itself inside a band running from four times faster to three times slower than its baseline:
Critical finding
Pulls the seat to the fastest end of its band — it starts looking much more often until the situation resolves.
Quiet estate
Stretches the interval — nothing is happening, so it stops burning money checking.
Approaching your monthly cap
Stretches it further, so the team slows down rather than silently overrunning your budget.
Blocked and waiting
Stops re-reporting entirely and waits — then returns immediately when the thing it was waiting on moves.
Why 'blocked' behaviour matters
A seat waiting on you for an answer, or on a colleague to pick up a hand-off, stops repeating itself. This is the difference between a team that reports the same blocker forty times overnight and one that raises it once and waits.
Reads are free-hand. Writes never are.
Reads run on the seat's own authority against a read-only capability allowlist. Writes never do — a proposed change goes to the Changes queue and runs only once a person approves it.
Three limits that hold regardless of grade
- High-risk work is staffed behind mandatory approval whatever the seat's grade. A senior seat does not earn the right to skip the gate.
- Seats deliberate with each other inside a cycle, and a senior seat can endorse or reject a colleague's conclusion — but can never approve a change into execution. Only a person does that.
- Everything a seat concluded carries the reads behind it, so you can check its working.
How billing works
This is the part most people want explained precisely, because "AI agents working overnight" sounds like an unbounded bill.
Billed per rota-minute
Not per scheduler tick. A seat that is off shift costs you nothing at all.
Quiet cycles bill at a quarter
A cycle that found nothing bills at the standby fraction — 25% of the seat's rate.
AI passthrough is separate
Metered separately at a 1.5x factor, so the model cost is visible rather than hidden in the seat rate.
Its own balance and ledger
Workforce draws on its own funds, NOT the Apex AI credit pool. A busy team can never drain the credits Apex chat needs.
Answering 'you charged me for a quiet night'
Because a quiet cycle bills at a quarter rate and every invoice line traces back to a cycle you can open and read, that question is answerable from the row itself — not from a support ticket.
Features
How to use it
Start with one or two seats, not a full team
Pick the coverage gap that actually hurts — usually out-of-hours reliability or unwatched spend. Staffing a full team on day one makes it hard to tell what any single seat is worth.
Check the projected monthly cost before launching
The Teams page shows it before the rota starts. Set the monthly cap at the same time.
Watch the Live floor for the first few shifts
You are checking two things: that the seat's conclusions match what you would have concluded, and that its pacing looks sensible for your estate.
Work the "Needs you" list daily
This is the page that determines whether Workforce is useful. A team blocked on a missing permission or an unanswered question produces nothing — and correctly stops asking.
Approve changes deliberately at first
Read the reads behind each conclusion. You are building the same trust you built with Autopilot, by the same method.
Set Missions once the team is trusted
Give it outcomes to work toward rather than leaving it purely reactive.
Reconcile Spend monthly
Seat hours by role against what the team delivered. Retire seats that are not earning their rate — the per-seat pricing is what makes that judgement possible.
Why it matters
Managed service is the product most FinOps customers actually want and almost none can afford, because it is priced as people.
Workforce prices it as seat-hours, bills a quiet night at a quarter rate, and shows the working: every conclusion carries the reads behind it, every change carries an approval, and every invoice line traces to a cycle you can open and read.
Connects to
- Complements Apex — Apex answers when asked; Workforce works a rota
- Changes route through the same approval discipline as Approvals
- Funded separately from Apex AI Credits
- Reads the estate through Cloud Accounts
- Shares the trust-building path with Autopilot
Apex
The AI command center — a natural-language control plane over the entire platform, wired into 135 real capabilities with full permission and approval enforcement.
Autopilot
An autonomous cost optimisation agent that acts within explicitly defined guardrails, with three graduated trust modes so you adopt autonomy at your own pace.