Quick Answer: AI agents for operations take on the coordination work operations teams carry every day: chasing suppliers and contractors, assembling reports and evidence, triaging internal requests and surfacing exceptions. They work best where the work recurs, mistakes are caught before they land and the rules can be explained. Because much of this work is never written down, the list of where agents fit has to come from the team itself, and the hours saved only count if they go somewhere deliberate.
In a survey of 9,615 knowledge workers carried out in November 2022, Asana found that coordinating work, rather than doing the skilled work people were hired for, took 58 per cent of the working day. GWI ran it for Asana, which sells work management software, across six countries including the UK.
Operations teams sit in the middle of that coordination. They are the people everyone else chases, and the people who chase everyone else. Yet most guidance on AI agents for operations treats operations as a supply chain, a factory floor or an IT service desk. That advice suits those settings, but says little about the scheduling, chasing and exception handling that fill most operations teams’ weeks.
AI automation for operations teams pays off when it is pointed at the work those teams actually do. That work is hard to see from outside, so the list of where AI agents for operations fit has to be built from the inside.
What Operations Work Looks Like
Operations means different things from one organisation to the next. In a manufacturer it may include the production line; in a professional services firm it may be closer to office management. The common thread is keeping work moving between people, sites, suppliers and systems. Most operations teams handle some mix of the following:
- Scheduling and resourcing: rotas, absence cover, and bookings for rooms and equipment.
- Suppliers and contractors: orders, follow-ups, renewals and performance.
- Internal requests: the steady flow of asks from other teams or sites.
- Reporting and status: weekly packs and leadership updates built from several systems.
- Compliance and audit evidence: certificates, logs and records gathered ahead of an inspection.
- Incidents and exceptions: whatever does not go to plan, and the follow-through afterwards.
That is also where AI is already landing. Eurostat figures for 2025 show that among EU enterprises with ten or more employees that used AI, 31.05 per cent used it to organise business administration processes or management, while logistics was the least common purpose, at 6.08 per cent. Large enterprises were far more likely than small ones to use AI for administration, at 43.38 per cent against 28.49 per cent. AI agents in operations, in other words, are arriving mostly in coordination and administration rather than on the warehouse floor. It mirrors the wider pattern for AI agents across the business, where repeatable tasks move first.
Why Operations Work Is Hard to See
Good operations work tends to show up as an absence: nothing missed, nothing late, no one left waiting. In a 1999 paper, the researchers Susan Leigh Star and Anselm Strauss argued that no work is inherently visible or invisible. People see work through indicators, such as a finished product or a changed state of affairs, and as working practices change, seeing the work behind those indicators takes longer chains of inference. In operations, the indicator is that things ran smoothly, and the effort behind it rarely appears on any list.
The knowledge behind that effort sits mostly with the people doing it: who to call at which supplier, which site always needs a reminder, which system has to be checked by hand. This is know-how that lives in people, and much of it is judgement that is hard to put into words, so it rarely reaches a procedure document.
Two things follow. Most published lists of AI agents for operations are organised by industry, so they describe a generic operations function rather than the one in front of them. And some of the most useful automation in an operations team may already exist as an automation someone built for themselves, working well but known to one person.
Where AI Agents for Operations Fit
Grouped by the kind of work rather than by industry, AI agents for operations tend to fit three areas. Not every step inside them needs an agent: where the inputs are predictable and the rules never change, fixed automation rather than an agent is often the better tool.
Chasing and coordinating
Much of an operations week is spent following things up. An agent can track open items with suppliers, contractors and sites, send reminders on an agreed schedule, read the replies and escalate when an answer does not come or does not fit. The person keeps the relationship: deciding when to push, when to give ground and what to do when a supplier simply cannot deliver.
Assembling reports and evidence
Weekly performance packs and audit evidence files take hours mainly because the material sits in several places. An agent can pull figures from the systems that hold them, draft the commentary, gather certificates and logs into one file and flag what is missing. The person checks the numbers, adds the context only they know and signs the result off.
Triaging requests and exceptions
Requests from other teams and sites arrive by email, form and phone. An agent can log each one, route it to the right person or contractor with the context needed to act, and chase it until it closes. Exceptions can be raised early rather than discovered late. Where a request crosses several teams, the handoffs matter more than any single step, and the process those handoffs belong to needs designing as a whole. The agents already running in production mostly sit at points like these: high volume, repeatable and easy to check.
What Should Stay With a Person
Knowing where AI agents for operations should stop matters as much as knowing where they start, and the deciding factors are practical rather than technical. A few questions about each task usually settle it.
| What to look at | Suits an agent | Better kept with a person |
| How often it happens | Daily or weekly, many times | Rarely, or once a season |
| What starts it | An email, a form, a date or a system event | A conversation or a hunch |
| Cost of a mistake | Small, and caught before it lands | Costly, public or hard to reverse |
| Who spots an error | A reviewer, or the next step in the work | Nobody, until much later |
| What is at stake | A routine transaction | A negotiation, a complaint or a staff conflict |
| Where the rules live | Written down, or explainable in one conversation | In one person’s judgement, still forming |
| Exceptions | A known share, routed to a person | Most cases are exceptions |
Most operations work splits within a single task rather than between tasks. The agent does the chasing; the person decides what to do when the answer is no. Treating that split as a design decision, made task by task, is what keeps AI agents for operations useful rather than risky.
One Operations Team, Five Pieces of Work
Consider a group of 24 veterinary practices supported by a central operations team of six. Asked where AI agents for operations might help, the team’s first answer is the weekly report. Asked to walk through an actual week, they describe five pieces of work.
| Work | What the agent takes on | What a person keeps | Settle first |
| Weekly practice performance pack | Pulls figures from the practice and finance systems and drafts the commentary | The operations manager checks the figures and adds context | Which system is the source of truth for each number |
| Locum and rota cover | Matches gaps to the availability list and contacts locums within agreed rates | The rota coordinator approves bookings and resolves clashes | The availability list lives in one coordinator’s spreadsheet |
| Supplier back-orders | Chases suppliers and proposes substitutes from an approved list | A clinical lead approves any substitute medicine | Substitution rules written down with clinicians |
| Equipment and facilities requests | Logs each request, routes it to an approved contractor and chases it to closure | The team handles disputes and any spend above a set limit | The spend limit and the approved contractor list |
| Inspection evidence | Gathers logs, certificates and training records into one pack and flags gaps | The practice manager signs the pack off | What complete means for each standard |
Two of the five things to settle first, the rota spreadsheet and the substitution rules, surfaced only because the coordinators were asked. Neither appears in any system the group runs. The spreadsheet matters for another reason too: the group’s cover arrangements depend on it today, and they would be at risk when the coordinator leaves.
Making AI Agents for Operations Pay Off
Time saved is not the same as value created. Pooling surveys from February to August 2025, estimates from the St. Louis Fed put self-reported time savings from generative AI at 1.6 per cent of all US work hours. The researchers caution that measured gains will be smaller if people spend the saved time on lower-value tasks, or on leisure at work.
In an operations team, freed time is easily absorbed by more of the same chasing. Deciding where the hours from AI agents for operations will go, whether that is the backlog, site visits or supplier reviews, belongs in the plan from the start. A few things are worth settling before the first agent goes live:
- Measure how long the work takes today, so the hours saved can be shown rather than estimated.
- Give each agent a named owner in the operations team, accountable for what it does.
- Agree who reviews the agent’s output, how often, and what triggers an escalation to a person.
- Limit the access each agent is given to what its task actually needs.
- Decide in advance where the freed capacity will go.
Whether each agent is built or bought is a separate decision, made piece by piece.
Mapping the Operations Work First
GrowthNation starts with the work rather than the agent. Every person in the operations team, and the people at sites who rely on them, is interviewed by voice or text, so coordination work surfaces with both sides of each handoff described. The AI agents for operations built from those interviews are served centrally, so an agent like the rota cover one does not live in one coordinator’s account. Hours saved are tracked for each agent and rolled up across teams, so freed capacity is visible and can be pointed somewhere deliberate. Some of what surfaces will suit fixed automation or an existing tool instead, and knowing which is part of mapping the work.
Start With One Operations Team
Start with one interview. GrowthNation will map your operations team, show you the scorecard and the work it surfaces, and hand over a working agent. No deck.
Frequently Asked Questions
What are AI agents for operations?
AI agents for operations are software agents that take on the recurring coordination work of an operations team, such as chasing suppliers, drafting reports, gathering audit evidence and routing internal requests. People keep the judgement, relationships and sign-off.
What can an AI agent do for an operations team?
An AI agent can follow up open items with suppliers and contractors, draft weekly reports and evidence packs from existing systems, log and route requests from other teams or sites, and raise exceptions early. It works best on work that recurs often, starts from a clear trigger and is checked before anything goes out.
Which operations tasks should not be given to an AI agent?
Tasks that happen rarely, where a mistake is costly or hard to reverse, or where most cases are exceptions are better kept with people. So are negotiations, complaints and sensitive staff conversations. In many tasks the split is partial: an agent prepares or chases, and a person decides.
How do you find the right operations work to automate?
Ask the people who do it. Much operations work lives in personal spreadsheets, inboxes and habits that no system records, so a list built from the top down tends to miss it. Walking through a real week with each person surfaces the recurring work, and the rules behind it, worth handing to AI agents for operations.
How do you measure whether an operations agent is paying off?
Record how long the work takes before the agent goes live, then track the hours it saves, counting the time people still spend reviewing its output. Then check where the freed time went. Hours saved only add value if they go to work that matters more.
How much oversight do operations agents need?
AI agents for operations need a named owner, a regular review of their output, clear rules for when to hand a case to a person, and access limited to what each task requires. Oversight can be relaxed gradually as an agent proves reliable, and tightened again if errors appear.
