What does agentic AI actually do in a forwarding office?

An agent is software that does a job for you, from one end to the other.

Think of a new starter on the pricing desk. You give them an enquiry. They read it. They hunt for the rates. They write the quote and bring it to you. You check it and send it. An agent works the same way.

The word agentic just means the software acts. It does not sit and wait to be asked for every step. Work lands, and it starts.

That matters because of what a forwarding office is actually full of. Not big calls. Reading, fetching, typing, chasing and passing work along.

An agent takes the reading, the fetching, the typing and the chasing. It does not take the judgement. That split is the whole idea, and the rest of this page is what it looks like in practice.

Five words come up again and again. Here they are without the sales gloss.

Five words cover nearly all of it. It reads. It fetches. It drafts. It chases. It hands over.

The wording in the table comes from one AI-native system, FreighAI, used here as the example. Any vendor should be able to give you five lines like these about their own product.

The same five words show up all day, on other jobs:

• Answers “where is my cargo?” on the customer’s own thread, from carrier status, for approval.
• When the customer pushes back on price, a counter-offer is drafted and waits for approval.
• Quotes expire on their own and can be revived if you want.
• Reads a carrier bill and matches every airway bill back to its job, checking line by line before it is booked.
• FreighAI talks to carriers, portals, your TMS and your accounting system itself. Nobody hops between websites to put an answer together.

Notice what none of these is. None of them decides anything. They fetch and they type. That is the work your people were doing between the calls that matter.

The judgement. And the send button.

Read that twice. Your pricer still owns the price. She just stops hunting for it.

These stay with your people, on every system worth buying:

• What margin to take on this lane, for this customer, this week.
• Whether to fight for the job at all, or let it go.
• Which customer gets a phone call instead of an email.
• What to promise when the date is tight.
• When something looks wrong and needs a human eye.

A good system also shows its working. You should be able to see where each number came from. A rate card, a carrier reply, a past quote. If you cannot see that, you cannot approve it honestly.

One sentence holds it. A chatbot answers questions. An agent does the work and brings you the result to approve.

Most people meet AI as a chat box. You type, it types back. That is useful, and it is not the same thing.

Both have their place. A chat box is a good helper for one question. It cannot run your Tuesday.

This is worth knowing before you buy. Plenty of older systems have added a chat box and called it AI. Ask to be shown the work happening, not the chat.

Here is one enquiry, from the minute it lands. Ten cartons, Dubai to Hamburg, by air. The times are a shape to hold, not a promise.

• 0110:00. The email arrives. It is a forwarded thread with a photo of a packing list. The weight is missing.
• 0210:00. The agent reads it. It works out that this is an enquiry, and what it needs to price it.
• 0310:01. It drafts the email asking for the weight. Your pricer reads the draft and approves. It goes out in her name.
• 0410:20. The weight comes back. The agent pulls your rate card and checks the carrier.
• 0510:21. You hold no rate for one leg. It emails every agent you use on that lane at once, with the route filled in.
• 0610:40. Two agents reply. The third does not. It chases the third by itself. Nobody had to remember.
• 0711:05. It drafts the quote. The price follows what this customer paid on this lane before, inside your margin rules.
• 0811:06. Your pricer checks the margin, changes one line, and approves. The quote goes out.
• 09Day 3. The customer pushes back on price. The counter-offer is already drafted and waiting.
• 10Day 5. The customer says yes. The quote becomes a job on its own, filled in.
• 11Day 6. Operations picks it up with the email and everything it found attached. Nobody re-types a thing.
• 12Day 12. The customer asks where the cargo is. The answer is drafted from carrier status, on the same thread.
• 13Day 30. The carrier bill arrives. Every airway bill is matched back to its job, line by line.

Now count what a person did in that story. Four approvals and one margin call. Nobody opened a portal. Nobody chased anybody. Nobody re-typed the quote into a job.

Count the waiting too. The work took minutes. The hours were waiting on other people. That is the honest picture of most quoting, and it is why the fetching is where AI pays.

The worry we hear most: what if the AI misbehaves, and do I have any control? Here is the honest answer, part by part.

You now know what an agent does. The next questions are where it pays in your office, and how to tell a real one from a chat box bolted onto an old system.