An AI agent workforce for freight forwarders is a roster of software agents that each do one back-office job: the agent reads shipment emails and documents, does the keying and cross-checking, and handles the action allowed for that job. The agents cover the jobs that eat an ops team’s week. A quote agent prices an LTL rate request and replies in a median of 12 seconds at a US trucking agency, a job that takes a desk about 12 minutes. An order entry agent builds the TMS lot, charge lines included, that takes about 30. A finance audit agent checks the charges on a job against the vendor invoice and flags the gaps. At one US import freight forwarder running these agents in production, that keying added up to about 17.5 hours a week across 25+ lots and 20 filings. Each action has its own review setting. Customs filings still require a person.
What is an AI agent, in plain English?
An AI agent is software that carries a task from the inbound document to a finished piece of work, instead of answering questions about it. A chatbot waits for a prompt and returns text. An agent is pointed at a job: it reads the email and the attachments, works out what the job needs, pulls the fields, checks them against records it has seen before, and produces the thing the job was for, a drafted filing or a TMS record. Then it runs on its own, goes for review, or waits for a person, depending on the setting for that action.
The distinction matters because the first wave of AI tools for freight stopped at extraction. They read a PDF and handed back structured data, and someone keyed that data into the system of record by hand. The keying was the job. Software that stops short of it saves some reading time and leaves all of the typing.
Most companies are still on the wrong side of that line. In McKinsey’s State of AI research, reported in March 2026, 23 percent of organizations were scaling an AI agent in at least one business function and another 39 percent were experimenting, but within any single function no more than 10 percent had reached scale. The gap between trying agents and getting value from them is, in freight at least, exactly the gap between extracted data and finished work.
What does an AI agent workforce do at a freight forwarder?
An AI agent workforce staffs the forwarder’s repeatable back-office jobs with one agent per job, the same way the team itself is organized around desks. That is how TIO is built for freight forwarders and 3PLs: a named agent for the ISF filing, one for order entry, one for finance checks, each reading the same inbox your team already reads.
Every agent follows the same shape:
- 01Your inboxBooking confirmations, worksheets, house bills, invoices
- 02The agentsOne agent per job reads the email and assembles the draft
- 03Your teamflagged recordsA person reviews what the settings send for review and corrects anything wrong
- 04TMS and portalThe lot lands in your TMS under your action settings. A customs filing waits for a person
Work arrives by email, because in freight everything arrives by email: the booking confirmation, the worksheet, the house bill, the invoice. The agent picks up the message and its attachments, does the reading and assembly a person would have done, and prepares the finished work. Each action an agent can take has its own setting: it runs on its own, it sends the work for review, or it waits for a person’s reply. Anything outside your rules goes to a person. A customs filing is prepared as a draft, and a person on your team reviews and files it.
What does the ISF Agent do?
The ISF Agent turns an origin worksheet into a drafted ISF-10, ready for a person to review and submit. By hand, this filing takes about 15 minutes: the ten importer security filing elements plus the two bill-of-lading identifiers sit scattered across a worksheet and its attachments, and a person keys each one into the filing portal, resolves the HTS code against prior filings, and confirms the seller and ship-to addresses. At the forwarder we measure, that happens about 20 times a week. If ISF-10 is new to you, our plain-English guide covers the ten elements.
The agent reads the worksheet and the supporting documents, resolves the parties and the HTS codes the way the team has filed them before, flags anything it is not sure about, and drafts the filing in the portal. The person who files reviews the draft, fixes what needs fixing, and submits.
The stakes are why the person stays. ISF-10 is due 24 hours before cargo is laden aboard the vessel, and the filing goes out under a filer code that belongs to a person. No agent should carry that signature, and this one does not. A filer on your team reviews every draft and makes every submission.
What does the Order Entry Agent do?
The Order Entry Agent takes the documents for a booked shipment and creates the complete lot in your TMS, charge lines included, and sends a review email that shows every field. Order entry is the highest-volume keying job in the back office: an ocean import lot runs twenty-odd fields sourced from four different documents, call it 30 minutes by hand, and the forwarder we measure builds more than 25 lots a week. We wrote a full deep dive on how the order entry agent works.
The agent reads the bill of lading, house bill, delivery order, and packing list, pulls the fields, and resolves the parties against lots the team has built before, so a consignee spelled three ways across three documents lands as the one record you actually use. It creates the lot with the charge lines and sends a review email that shows every field. A party it cannot match, or two documents that disagree, go to a person.
For a 3PL, the same agent does the same job under a different name: a delivery order arrives by email and a shipping entry has to exist in the system. The read and the review email are the same.
What does the Finance Agent do?
The Finance Agent reconciles the charge lines on a job against the vendor invoice and flags the gaps before they get paid. Reconciliation is the job that slips, because it is nobody’s morning priority. A rate that came in above the quote, a charge on the invoice that never made it onto the job, a duplicate line: each one is real money out, and by month end nobody has time to hunt for them line by line.
The agent runs the standing check a person rarely gets to. It reads the job’s charges and the vendor’s invoice, matches line against line, and books the lines that match, for vendors you have approved. Every mismatch goes to a person, who decides what it means: dispute it, correct the job, or approve it as it stands. The agent does not pay.
- BY HAND
- ~15 min per filing
- AGENT
- Seconds, you submit
- BY HAND
- ~30 min per lot
- AGENT
- Seconds, then you review
- BY HAND
- Month end, or never
- AGENT
- Every job, you review
How much time does an AI agent workforce actually save?
At the US import forwarder running these agents in production, the measured keying was about 17.5 hours a week, close to two days, and the agents now do it in the background. The math is plain. More than 25 lots a week at about 30 minutes each is roughly 12.5 hours. About 20 ISF filings at 15 minutes each is another 5. Those are first-party figures from a live operation, consented for use here, not an industry estimate.
Side by side, the same jobs look like this:
| The job | By hand | With the agent | Weekly volume |
|---|---|---|---|
| ISF filing | ~15 min per filing | Drafted in seconds; a person reviews and submits | ~20 filings |
| Lot / order entry | ~30 min per lot | Created in seconds; a review email shows every field | 25+ lots |
| Charge reconciliation | Month end, or never | Standing check; a person reviews every flag | Every job |
The saved hours only matter if they turn into something. Nobody hires an ops team to retype worksheets. The forwarder that gets close to two days a week back spends it on the work customers notice: answering the client before the client has to chase, catching the exception while the container is still at origin, quoting faster, taking on the next account without a new hire.
How do you put agents to work without risking a bad filing?
Each action an agent can take has its own setting, and you start with one job, not ten. The pattern that works:
- Pick the job that hurts most. For most import forwarders that is order entry or the ISF.
- Connect the agent to the inbox where that work already arrives. There is no migration and nothing new for the team to learn.
- Let the agent read a batch of your recent jobs first, so we can write your parties, your codes, and your conventions into rules for your operation instead of a generic template.
- Run it alongside your current process. The agent does its work. Your team reviews what the settings send for review and corrects it.
- Keep correcting. When your team corrects an agent, the correction becomes a rule for your operation. It is a rule, not model training.
- Add the next agent once the first one has earned it.
The settings are not a compliance checkbox. They are the design. Each action can be set to run, go for review, or wait for a reply, and you change one action at a time as the agent earns trust. A customs filing always stays with a person.
Do AI agents replace your TMS or your team?
Neither, and a tool that claims otherwise should worry you. The agents sit between the inbox and the TMS you already run. Your TMS stays the system of record. The agent writes into it through an adapter built for your TMS, so there is no rip-and-replace and no parallel system drifting out of sync.
The team stays the team, too. The agents remove the 15 and 30 minute blocks of retyping that sit between a document arriving and a record existing. Everything that requires a forwarder stays: the judgment call on a flagged HTS code, the relationship with the client, the exception no playbook covered. The agents do the work. Your team makes every call.
TIO is this workforce, built for freight forwarders and 3PLs. The ISF Agent, the Order Entry Agent, and the Finance Agent are described above. TIO runs in production today at a US import forwarder and a US trucking agency. The full roster is growing. If two days a week of keying sounds familiar, book a demo and bring a few of your own shipment emails. We will run an agent on them live, and you can watch the work land, on your documents.
Frequently asked questions
What is an AI agent workforce for freight forwarders?
It is a roster of AI agents that each handle one back-office job function: an ISF agent drafts the 10+2 filing, an order entry agent creates the TMS lot with charge lines, and a finance audit agent reconciles job charges against vendor invoices. Each action has its own setting: it can run, be sent for review, or wait for a person's reply. Customs filings remain human-controlled.
Do AI agents file ISF or write to the TMS on their own?
You choose, per action. Customs is fixed: the agent drafts the filing and a team member reviews and files it, because CBP requires a human filer of record. For the rest, each action has its own setting: it runs on its own, it sends the work for review, or it waits for a person's reply. Anything outside your rules goes to a person. You can change a setting whenever you want.
How much time do AI agents save a freight forwarder?
Measured at a US import freight forwarder running the agents in production: lot creation went from about 30 minutes by hand to seconds, across 25+ lots a week, and filing preparation went from about 15 minutes to seconds, across about 20 filings a week. That is roughly 17.5 hours of keying a week, close to two days, handed to the agents, with the team still reviewing each lot and filing the agents prepare.
Do AI agents work with the TMS a freight forwarder already runs?
Yes, through an adapter built for your TMS. Adapters run in production for two TMS platforms. Another TMS needs its own adapter first. There is no migration and the TMS stays the system of record. The same pattern covers 3PLs, where the job is a shipping entry created from an emailed delivery order rather than an import lot. Start with one agent, expand when the first one has earned it, and nothing about the existing stack changes.
