Know what the promise will cost.
The contractor was the first version of staff augmentation. The agent is the next.
Contingent labour let firms add capacity without adding a permanent colleague, but the work still needed supervision, boundaries and a route home. Agents promise a cleaner version of the same flexibility while making the hidden management work easier to deny.
What it feels like
The subtle indignity of being told a role is temporary when the work it carries plainly is not.
You are not being replaced by a tool. You are being asked to notice which employment promise the tool makes easier to avoid.
They told you the role is temporary. The work plainly isn’t.
The desk is there. The queue is there. The release that has to be tested, the leave that has to be covered, the customer history somebody still needs to understand.
All of it will be there on Monday.
Only the person holding it is provisional.
That isn’t a complaint about contractors. It isn’t a complaint about taking a contract either.
You’ve got rent, obligations, and a calendar full of ordinary things that don’t pause while a company decides whether a need deserves permanence.
The manager who asked for help has a backlog.
The team has a deadline.
Somebody has to make the work move.
Flexible capacity can be a decent answer to an immediate problem.
There’s no machine to be afraid of here.
Notice the mechanism instead.
Organizations learned long ago how to add capacity without extending the employment promise that usually comes with it.
The contractor was one version.
The agent is the next.
Both arrive as relief. Both make one question easier to postpone. Who’s responsible for this work when the convenient extra capacity goes away?
2000 — A contractor gets a desk for the project.
2011 — The agency timesheet arrives on Friday.
2023 — The copilot is added to the team licence.
2026 — The agent does the task. Somebody still has to know when it should stop.
Real capacity, real boundary
Staff augmentation starts with your manager turning an absence into a requisition.
The request isn’t supposed to be “find a developer” or “make the release less late.”
It states your dates, your hours, the rate ceiling, the authorization, the equipment, the manager, and the reason a permanent hire or a supplier-owned service isn’t the better vehicle.
Then the distinction that follows is unfashionably clear.
If you join stand-ups, use client tools, take priorities from the client lead and sit alongside employees, your client bought staff augmentation.
If it needs defined testing coverage, reports, acceptance measures and a supplier manager, it should have bought a statement of work.
US federal acquisition rules say a performance work statement describes required results, not the manner of performance or the number of hours.
Three different purchases. Hours. A named deliverable. A service level.
That difference decides who carries the awkward part when your work doesn’t go to plan.
In augmentation, your client directs you and holds the risk that the work takes longer than hoped.
In a statement of work, the supplier prices management, planning, quality assurance, substitution, coordination, rework risk and profit into a milestone or a fixed fee.
That may cost more per visible hour. It can also be the price of delivery risk leaving the client, rather than an expensive version of the same worker.
The agency then turns the requisition into an assignment it can actually staff.
UK guidance requires at least six items from the hirer. Identity and location. Start and duration. Responsibilities and hours. Required experience or authorization. Known health-and-safety risks. The steps to control them.
An agency can’t safely fill your manager’s request by sending whoever looks available.
And before you start, the terms have to match your access.
Somebody confirms a bill rate, a purchase order, a start date, an approved supervisor.
For people operating through an intermediary, a medium or large UK private-sector client makes the status decision before first work and issues a Status Determination Statement with reasons.
That record travels through three parties. Client, you, agency.
A staffing supplier can carry the paper. It can’t make your actual working practices disappear.
On day one you get access, equipment, security instructions, team context, working hours, an escalation route, and the manager who sets your daily priorities.
In Great Britain, agency workers get shared facilities and vacancy information from day one.
That isn’t sentimental inclusion. It’s a recognition that contingent capacity has to be inside enough of the workplace to be useful, while the employment relationship stays somewhere else.
That’s the old bargain in its most honest form.
The agency employs, sources, screens, pays, redeploys. The client integrates, directs, reviews, approves. You supply the capacity and carry the uncertainty of the end date.
In one US snapshot, 54.8% of temporary-help agency workers were also contingent on that same main job. Worldwide, 62% of agency workers held full-time jobs in 2023.
Temporary has never reliably meant incidental.
Usually it means your work is continuing while the promise isn’t.
Friday was the control surface
That Friday timesheet was never just a small administrative insult.
It was the record from which the agency calculated your pay and the client received its invoice.
The client consumed labour time. You got paid through the supplier. The agency reconciled the gap.
So your manager had to approve accurately and promptly, because that approval was operational evidence and payroll input at once.
Tickets closed, test cases run, handover notes and accepted shifts could all show useful work. None of them turned a time-based assignment into a supplier-owned outcome.
Which is why the bill rate always revealed more than your wage.
It can fund statutory employer contributions, paid leave, benefits, insurance, recruitment, screening, payroll, unbillable administration, programme fees, bad-debt risk, overhead and profit.
It can also carry pass-through items, taxes, approved expenses, and shift premiums.
US median weekly earnings for full-time temporary-help agency workers were US$818 in July 2023, against US$1,132 for workers in traditional arrangements.
Neither of those is your bill rate. That’s exactly the point.
The margin wasn’t simply somebody’s greed hiding in a spreadsheet.
It was the price of a supplier finding you before the client knew whether the need would last, employing you while the client directed the work, and taking you back into a pool when the release ended.
The private-employment-services industry placed just over 60 million people into temporary and permanent jobs in 2023. Scale makes machinery easier to miss.
And your invoice concealed work that belonged to the client.
Manager time. Procurement. Interviews. Security review. Access provisioning. Equipment. Induction. Training. Supervision. Quality review. Knowledge transfer. Offboarding.
A lower hourly rate becomes a higher cost per useful output if you take longer to learn the systems, need rework, or leave before the knowledge settles.
The useful comparison was never employee wage against agency markup.
It was the lifetime cost of capacity against the duration and uncertainty of the need.
At the end, the bargain had a route home.
Your manager could end, extend, materially change, convert, or recruit directly. A material change in job, location, hours, manager or risk called for a new assignment record. Not a casual email.
A genuine end meant handover, access removal, equipment return, final time approval, and feedback you probably never saw.
You couldn’t stay on the badge, in the account and in the work by inertia after the purchase order stopped.
You’re not cynical for seeing the continuity here.
Flexible capacity has always depended on somebody else holding the uncertainty.
You held it in the employment channel. The client held it in supervision, integration and delivery risk. The agency held it in sourcing, payroll and the prospect of redeployment.
The timesheet made those obligations visible because it had to.
The new timesheet is a trace
An agent offers you a more flattering picture.
No empty desk. No awkward introduction. No Friday request sitting in your manager’s inbox.
It reads a case, retrieves a contract, assembles a draft and returns a clean sentence before anybody decides whether a person is needed.
Which makes it more important, not less, to ask what capacity got bought, what it may do, and where its work ends.
The payables example starts from a narrow request. Prepare a supplier bank-detail change. Not “fix this supplier.”
That isn’t an attempt to make software look like an employee. It’s the digital equivalent of refusing to let you start because a manager wants help tomorrow.
Read-only retrieval, a reversible proposal, and a commit are different effects. The first can expose data. The second creates a draft. The third changes a business, legal or physical state.
The employee who asked for the work doesn’t need to donate every permission they accumulated over nine years just because the agent received a polite instruction.
So each agent service needs its own machine identity. Separate from your identity, and separate again from the identity of the tool it calls.
Its credentials should be narrow, short-lived and audience-bound.
And its version of time approval has to be an actual gate.
A message saying approval was requested isn’t a control if the model can carry on regardless.
Then comes the part the friendly interface most wants to hide.
A model saying the change is complete isn’t proof of completion.
A release path isn’t pessimism. It’s how capacity stops without leaving its access and its decisions behind.
The promise in the room
You never proved that people were interchangeable.
You proved something more limited and much more useful to your employer.
A bounded slice of work could be obtained through a requisition, made productive through supervision, paid through approved time, and ended through a formal release.
Your work might continue. Your role might be renewed.
But your employer had a mechanism for treating the employment promise as optional.
The agent makes that mechanism cleaner.
No recruiter builds a pool. Nobody has a shift differential or a waiting period. No conversion clause asks whether the temporary solution quietly became the permanent function.
The vendor relationship is a model, a platform, a connector, a team licence.
Your management work remains. Identity, authority, tool scope, approval, evidence, change control, recovery.
It’s just easier now to call all of that configuration instead of supervision.
That’s the correction. You’re not being replaced by a tool.
You’re being asked to notice which employment promise the tool makes easier to avoid.
What feels indecent isn’t that every agent will take a job.
It’s that the old vocabulary admitted you were being brought in without the full promise of belonging.
The new vocabulary lets an organization add capacity without admitting it added anyone at all.
Faster than a requisition. Quieter than a timesheet.
And still dependent on somebody knowing what the capacity may do, how its result gets accepted, and when it’s allowed to stop.
Put this to work
- Staff augmentation buys supervised time under client direction, while a statement of work buys an accepted outcome under supplier accountability.
- An agency bill rate funds employment, sourcing, payroll, risk, programme, and supplier costs as well as worker pay.
- Requisitions, onboarding, client supervision, accurate timesheet approval, and formal end-of-assignment controls make contingent capacity governable.
- An agent requires its own machine identity and narrow credential and tool scope rather than inherited employee authority.
- An approval gate binds a named approver to a stable proposed act, evidence, policy version, and expiry.
- A committed downstream result and a release or recovery path matter more than an agent's completion message.
Questions
- What is the difference between staff augmentation and a statement of work?
- Staff augmentation buys a person's time at an agreed rate while the client directs priorities, supervises the work, and bears the risk that it takes longer than expected. A statement of work buys a milestone, deliverable, service level, or capacity outcome. The supplier then carries defined management, planning, quality, substitution, coordination, and rework responsibilities and prices that delivery risk.
- Why doesn't calling someone a contractor turn the work into a supplier service?
- The label does not determine the arrangement. Staff augmentation exists when a person joins client meetings, uses client tools, takes daily priorities from a client lead, and works alongside employees. A supplier-owned service requires defined coverage, reports, acceptance measures, and supplier management. A fixed deadline alone does not create a statement of work or transfer delivery accountability.
- What does an agency need before it sends someone to a client?
- UK guidance identifies six core items from the hirer: identity and location, start and duration, job responsibilities and hours, required experience or authorization, known health-and-safety risks, and steps to control them. The agency also needs the pay or bill-rate parameters, workplace conditions, security checks, and the client contact authorized to approve hours. A vague request leaves suitability and accountability unresolved.
- Why is timesheet approval more than an administrative chore?
- In staff augmentation, the client consumes labour time, the worker is paid through the agency, and the agency invoices from client-approved time. Accurate, prompt approval is therefore both an operational record and a payroll input. Tickets, test cases, handover notes, or shift acceptance can show work occurred, but they do not turn a time-based arrangement into a fixed-result supplier service.
- Why can an agency bill rate be much higher than a worker's pay?
- The bill rate funds more than pay. It can cover statutory employer contributions, paid leave, benefits, insurance, recruitment, screening, payroll, unbillable administration, programme fees, bad-debt risk, overhead, and profit. Pass-through costs, taxes, expenses, and overtime or shift premiums can be separate items. The gap between take-home pay and the bill rate is not, by itself, a measure of agency margin.
- Why isn't the cheapest hourly contractor automatically the cheapest capacity?
- The client also pays for manager time, procurement, interviews, security review, access, equipment, induction, training, supervision, quality review, knowledge transfer, and offboarding. A lower hourly rate can therefore produce a higher cost per useful outcome when ramp time, rework, or turnover rise. The comparison must include duration, uncertainty, expected output, and which party carries delivery risk.
- What has to happen when a contingent assignment ends?
- The manager decides whether the need ends, extends, changes materially, or becomes a direct hire before the expected end date. A genuine closure includes handover, access revocation, equipment return, final time approval, and performance feedback to the agency. A material change in job, location, hours, manager, or risk needs a new assignment record rather than an informal extension.
- Why does an AI agent need its own identity when an employee asked for the task?
- An agent service needs a machine identity separate from the employee identity and from the identity of every tool it calls. Authentication establishes who makes the request; authorization establishes whether that identity may act on a resource for a purpose at that time. The employee request remains audit context. It does not give the agent every permission accumulated by that employee.
- What makes an agent task bounded instead of just automated?
- A bounded agent task begins with a specific purpose, such as preparing a supplier bank-detail change, and receives an immutable run identifier tied to requester, record, tenant, policy version, and time. In the example flow, three deterministic checks confirm an open case, assigned requester, and in-scope supplier before planning. Named tools and runtime-validated state transitions constrain the work further.
- Why is an approval request not the same as an approval gate?
- An approval gate ties an authorized person's decision to a stable proposed action, supporting evidence, policy version, and expiry. It records four outcomes: approve, reject, request further evidence, or cancel. The runtime checks that the approval still applies immediately before the tool runs and fails closed when it cannot. A notification can describe a requested approval while allowing an action to continue.
- What proves that an agent actually completed a change?
- The authoritative proof is the downstream system's committed result, such as a confirmation number, resulting record version, and timestamp. The agent's natural-language completion message can be wrong, stale, or incomplete. A useful trace links request identity, evidence, policy decision, credential scope, approval, tool arguments, external result, and recovery event, so a reviewer can distinguish a completed message from a completed transaction.
- What happens when an agent cannot verify the evidence or the permission?
- The run enters a failed or awaiting-evidence state, escalates, or parks according to policy; confidence does not become authorization. If a tool call times out, the runtime queries the downstream system with the idempotency key before recording completion or retrying. A 400 malformed request, 401 absent or invalid authorization, and 403 insufficient permission are three different conditions with different recovery paths.