Field service automation: today, next, tomorrow

1. Today: where the job leaks

Most trade businesses already run a job system. Jobs get created, techs get a list, invoices go out eventually. The leaks are in the handovers between those steps, and they follow the job from the first call to the next service.

The cash path, from enquiry to paid, is the biggest single leak for most shops. The day’s operations (the right person, the right parts, the right hours, the next visit) are the next set. All of it sits on top of the job system you already use, and nothing here asks you to replace it.

2. The technology is not the blocker

Almost nothing in the “future” of field service is waiting on an invention. The large platforms already ship agents that act:

• ServiceTitan’s Atlas answers a technician’s questions from the job history, equipment and invoices, and is pitched for booking and dispatch recommendations too.
• Salesforce’s Agentforce for Field Service drafts the post-work summary from what the tech captured and reads a pre-work brief aloud before they arrive.
• Microsoft’s Copilot in Dynamics 365 Field Service summarises a work order from its bookings, notes, parts and asset history.

Those are enterprise platforms, priced and sized for enterprise service fleets. A 12 person electrical contractor on ServiceM8 and Xero is not moving to one to get a summary feature, and should not. And the general assistants (Grok Bot, Claude and the rest) can already read your systems and draft the work for the price of a subscription you may already pay. What each kind of agent can do in 2026, and why a typical trade business is not running any of it yet, is in The AI your trade business could be running in 2026.

So the question is not whether a model can do it. It is three plainer ones. Is the workflow written down? Is there one job id the whole chain can trust? And is the risky step (the live call, the write into your ledger, the record that has to survive a dispute) done in a way someone can check afterwards? Without those, an agent is guessing at speed.

3. Next: a standard for every handover

We publish each handover as an open build standard: the checks in order, the exceptions and who owns them, the step not to hand-roll, and a pass test you can stage on your own business. Here is the job’s path, with the standard that holds each step. The ones marked Next are the gaps we are writing now.

One standard holds the cash steps together: Cash Seam. The job id is written once, by the job system, and every later step reads it rather than matching on a name or a phone number. That one rule is what lets a call, a quote, an invoice and a chase be one record instead of four guesses.

4. Run it your way

A standard says what has to be true, not what you have to buy. The same checks and the same pass test hold whichever way you run it, so you can pick the way that suits each job and change your mind later without starting again.

Mixing is normal. A business can let Get Paid ring the overdue invoices, run Invoice Out from a Google sheet, and ask Grok Bot for Monday’s call list, all against the same Xero. The rules in Cash Seam are what keep three tools one system: each reads the id the last one wrote. Here is what is ready today, standard by standard.

Every standard can also be built in your own code from its page. “Build guide” means the platform has a general guide for that standard but no ready-made template yet. This table reads the same lists the standard pages do, so it changes when they change.

5. Tomorrow: agents acting inside the rules

Once the standards are running and the job id is clean, an agent can act rather than just suggest, because the standard already says what it may do, what it must refuse, and who gets the exception. We have not built these agents. What we publish is the set of rules each one would act inside.

The office’s half of tomorrow is one queue. Every standard already parks what it cannot decide, each with a reason and an owner. Exception Desk, the last of the standards we are writing, puts all of them in one place: every item a standard parks, from an unpriced quote line to a job nobody can be sent to, in one queue with a reason, an owner and a clock, closed only by a person.

The tech’s day becomes a brief in the morning, the work, and a close on site. The office’s day becomes that queue. The customer gets a window that holds, a call if it moves, and an invoice that arrives with the job. It is the same job system you run today. The routine decisions move from a person clicking through screens to an agent working inside written rules.

6. The order that works

1. 1

   Cash first

   One job on the phone, written once into the system you already run, then the quote, the invoice and the chase on the same id. The money it brings back pays for what comes next.

2. 2

   Then the day's operations

   Whichever of dispatch, parts, timesheets or the next service costs you the most admin time or the most second visits.

3. 3

   Then the agents

   Only on top of clean ids and written rules. An agent that jumps to sensors or a fleet of agents before the cash path is reliable is building on a leak.

The same order, with every rung named, is on the roadmap. If you are already building your own agents, the shape each step needs before it can be supervised is in Building Agents That Run the Work. Still choosing a job system? Start with the comparison.

7. What to do this quarter

Run the eight-minute diagnosis. It maps the leaks you describe to the standards that already exist and names your first move.

Then start with the cash path. The rest of this roadmap is the same method carried further down the job: checks in order, named exceptions, one record, and the step you do not hand-roll. The connected, agent-run business is what you have when those standards are running, on whichever tools you chose.