AI will take our jobs.

I heard it in almost every conversation at Commodity Week in Geneva, usually from someone whose working week is made of exactly the tasks AI is now good at.

They are right about the tasks. They are wrong about the job.

The work changes shape. The 1980s ran a version of this - spreadsheets removed the manual entry, not the people who understood the trade.

What follows is the argument, one uncomfortable question, and what it looks like on one presentation.

The tasks that go first are not the job

Walk through a document workflow and the hours are rarely spent where the job description says they are. A value is read off a PDF and typed into another document. A field is checked three times because the first two checks found nothing. A figure is chased across a booking, an invoice and a certificate of origin.

None of that is the job. It is the friction around the job.

Some of those roles do shrink. Pretending otherwise is why people stop believing the rest.

That is where AI earns its place first: reading a document, extracting a field, comparing a presentation against a letter of credit, drafting the next document from the one before it. The routine part arrives prepared, and the person starts from a reviewed draft instead of a blank page.

What remains is the part that needs a person

A discrepancy is flagged. Someone still has to decide what it means, whether the certificate is good enough, and whether the presentation can go. A draft is prepared. Someone still approves it.

AI assistance is not commercial approval. It supports the examination. It does not replace professional judgement.

The uncomfortable question

If the friction is most of someone's job, what happens to that person?

That is the question most vendor blogs skip, and it deserves a straight answer. In a trade operations team, some roles are largely made of the tasks above: reading the same fields, retyping them, checking them again. AI reduces that work, and pretending otherwise is exactly why people stop believing the rest.

What we would argue is that the outcome is a management decision, not a property of the technology. The same capability can be used to cut a team, to take on more volume with the people already there, or to move experienced people onto the work that was always under-served: exceptions, counterparties, the trades nobody had time to look at properly. Two of those are better businesses. All three are choices someone makes and should be able to defend.

What it looks like on one presentation

Take a letter of credit presentation: a bill of lading, a commercial invoice and a certificate of origin, checked against the credit before the set goes to the bank.

The checking moves. The decision stays.

Prepare, review, approve

Inside a controlled workflow, AI prepares and a person confirms. Access, permissions and the record of what happened are part of the workflow, not added afterwards. Every action leaves a trail, whether a human or an agent did the work.

That is the difference between AI as a shortcut and AI as part of the process. One is a demo. The other is something an auditor will accept.

"Scale without headcount" does not mean fewer people

Handling more volume with the same team is not the same as handling the same volume with fewer people. One is a capacity decision, the other a headcount decision.

Where we land

"The tasks AI takes first are the ones nobody was hired for. That is not the job. That is the friction around it." - Laetitia Alonso, Director of Sales, Banking & Corporates

Start with one workflow

Pick one process your team handles every week. Map the steps honestly, agree what a reviewed output looks like, and test it with the people who do the work.

Then answer the only question that matters: is this for more volume with the same team, or for a smaller team? Say which one out loud, before you start.

See how DocChecker works here.