Agentic evaluation
brightonSEO · 8 and 9 October 2026 · The Brighton Centre
Shipping the output is the easy bit.
The checking is the job.
Anyone can get an answer out of a model. Getting work you would put in front of a client, at volume, without reading every line yourself, is a different discipline. It is called marketing engineering, and the talk is about the part that makes it safe.
01 — The new discipline
Capacity used to be the constraint.
Now it is the one thing you have too much of.
Marketing planning has always been rationing. How many pages, how many variants, how many markets, how many people. Every plan was an argument with capacity.
That argument is over. An agent will write the thousandth variant, read every competitor page, and rebuild the report every morning, and the marginal cost of one more is close to nothing. Supply is effectively unlimited.
When supply stops being scarce, the scarce thing moves. It moves to knowing what is worth making, and to proving that what you made is right, at a volume no one can read line by line.
That is not marketing as it was taught. You design the system, you instrument it, you test it, and you trust the output because the system earns that trust. That is engineering, applied to marketing.
What used to be scarce
Everything you could count.
What is scarce now
Everything you cannot.
Marketing engineering is the name for the work in the second column. The talk is about the part of it that decides whether any of this is usable: how you check the output of a system that never gets tired and never stops producing.
02 — The job title
mar·ket·ing en·gi·neer
/ MAR-ki-ting en-juh-NEER /
noun
1
a marketer who builds the system that makes the work, rather than making each piece by hand: the marketing engineer designs, tests and proves the process, then lets it run.
2
a practitioner for whom capacity is no longer the constraint, and whose effort therefore goes into deciding what is worth making and proving that it is right.
3
informal the person in the room who asks how do we know? before does it look good?
ORIGIN 2020s: marketing + engineer, by analogy with software engineer. Emerged as generative AI made output cheap and proof expensive.
Nobody trained for this. The people doing it came out of SEO, out of CRM, out of production, and taught themselves to build the system and then check it. The title is new. The job has been forming for a while.
If that sounds like where you are heading, the talk is the practical half of it: what the checking actually looks like when the thing you are checking never gets tired and never stops producing.
After the talk
Three ways to take this further.
Pick whichever one fits. All three take under a minute.
01
For leadership teams
AI & Agencies:
Understanding the future.
A complimentary half day with your SLT on where AI moves the value in your agency, what stops being yours, and what you have to own to stay hard to replace.
Half a day · Free of charge
See the session →02
For everyone else
Free AI mentorship
for marketers.
A monthly session and a check-in, open to anyone in marketing who wants to get good at this. Agentic evaluation is one of the things we will work through, along with whatever you are stuck on.
Monthly · Free · Starts after the conference
Join the programme →03
For the slow burn
Follow the work
on LinkedIn.
Builds, evals, what worked and what did not. Everything we learn ends up here first, usually before it is tidy.
Jeremy McDonald · We Are Pioneers
Follow on LinkedIn →Free AI mentorship
Get good at this, with company.
A monthly session plus a check-in, free, for people in marketing who want to actually build with AI rather than read about it. We work through real problems: evals, agentic workflows, what to automate and what to leave alone. Put your name down and we will be in touch before the first one.
What you get
01
A monthly session.
One topic, worked through properly, with the build shown rather than described.
02
A check-in.
Bring what you are stuck on. Most of the value is in other people's problems.
03
The eval material.
The checking patterns from the talk, written up, so you can run them yourself.
You are on the list
Thanks. We will be in touch.
You will get the date of the first session and the eval write-up from the talk.