Trimble’s new AI agent comes with a 10-hour clock

Arc Agent is aimed at transportation and logistics teams with a familiar mess: an order arrives in an email, a rate contract lives in a PDF, somebody copies fields into a transportation management system, and the exception gets chased across another inbox. Trimble says the agent connects those unstructured inputs to systems of record instead of stopping at a chatbot answer.

The launch skills include order entry and contract intake in North America, truckload rate intelligence in Europe, customer support that can open a Jira or Salesforce ticket, and a personal assistant that pulls email, calendar, message and project data into one prioritized view. Available skills depend on the Trimble products a customer already uses.

Trimble is also making a product bet against a swarm of separate bots. Arc uses one agent with a catalog of ready-made and custom skills, one set of permissions and one action history. The company says teams can build custom skills through conversation without engineering help, and that skills are tested and monitored before use.

The launch release cites a Trimble and FreightWaves survey in which 70% of transportation and logistics organizations named manual, repetitive tasks as their leading pain point. The release does not include the sample size or survey method, so treat that figure as a vendor-reported signal, not an industry baseline. The paper pile itself is easy to recognize without stretching the statistic.

A seat is blunt, but everyone knows what was purchased. “Agent working time” can mean several different things, and other vendors already prove how wide the gap is.

Retool says its billable clock starts when an agent is triggered and stops when it returns an answer. Waiting for human approval, authentication or subagents is excluded, but failed runs still consume time and subagents accrue their own billable runtime. Amazon Quick uses another definition: time spent across research, automations, its desktop app, artifact generation and custom AI apps, metered to the second. Its professional and enterprise pricing pages list additional automation hours at $3 each, with research hours priced separately in some cases.

Neither definition is inherently wrong. They price different systems. The problem begins when a buyer hears “10 hours” and pictures 10 hours of finished clerical work while the vendor is measuring model runtime, tool waits, parallel workers, testing or some combination of them.

Trimble’s announcement says overage hours are available, but it does not publicly say what starts or pauses the meter, whether unused hours roll over, how multiple skills share the pool, whether failed runs count or what an extra hour costs. A demo cannot answer those questions by feeling fast. The contract and usage screen have to answer them in plain language.

For a small logistics office, 10 focused hours might be plenty if Arc handles one repetitive lane with clean inputs: extract an order, validate required fields, write the record and stop when something does not match. The same allowance could vanish quickly if the first month is spent tuning a broad personal-assistant skill across email, calendars, messages and project tools.

Start where Trimble says it typically starts: one high-volume, well-defined job. Pick a queue with an obvious before-state and an obvious finish. Contract intake is easier to judge than “help everyone be more productive.” Order entry is easier to price than a company-wide assistant that is always reading and sorting.

During the pilot, put the hour meter beside the work. For every run, show the task, clock time charged, records changed, checks performed, exceptions raised, retries and final status. If a run fails, keep the time and the failure together. If a person waits 20 minutes for an explanation of a bad record, add that human time too. The invoice should not be the first place anyone learns that an hour was expensive.

Then compare the new lane with the old one: clean records completed, corrections, exception-handling minutes, late customer replies and the amount of work a person had to repeat. Faster runtime is nice. A smaller end-of-day pile is the reason to keep paying.

Ivy Chen sees the appeal of no seat limits for a dispatch office where several people may touch the same queue. She would still name who can launch work, who gets the 80% warning and what happens at zero. A shared pool with no owner can turn a quiet Wednesday afternoon into an admin argument about which team burned the month’s allowance.

Priya Rao would make the usage page translate hours into accepted work. Ten hours should appear beside orders entered correctly, contracts processed without rework, exceptions handed to the right person and human correction minutes. Runtime by itself cannot tell a buyer whether the machine finished useful work or practiced being busy.

Those views pull in different directions. Ivy wants a calm buying and rollout rule before the clock starts. Priya wants evidence after it runs. A serious agent product needs both, because predictable access without useful output wastes money, while useful output with an opaque meter is hard to trust or budget.

Ask Trimble for the exact billing event: trigger to finish, active model time, tool time or wall-clock time. Ask whether approvals, authentication, retries, failures, custom-skill testing and parallel work count. Ask whether the 10 hours reset monthly, roll over or expire, and whether an administrator can cap overages before they happen.

Ask to see one real usage statement from a task close to yours. It should be possible to trace an hour from the bill back to the work without reading a technical log. If a contract-intake run used six minutes, the team should see which document it handled, what record it created, whether it retried and how the exception ended.

Finally, price the human edge of the system. Trimble says Arc escalates work that falls outside its scope. Good. Count how often that happens, how much context arrives with the escalation and whether the person can finish without opening the original email and rebuilding the story.

Agent-hour pricing may be fairer than charging for every employee who needs occasional access. It may also make inefficient software look like productive labor. The difference is visible once the clock, the outcome and the cleanup sit on the same screen.