A good example of why most marketing teams are missing much of the value agents can create.
John Horton and the team at Expected Parrot (YC F25) have made research surveys version-controlled.
Which means every question, logic change and alternative version has a visible history. You can compare what changed, test different approaches without overwriting the original, catch errors and reproduce the research later. I love nerdy stuff like this when its actually useful.
Coding agents handle the technical plumbing, so marketers get the benefits without learning Git.
Now, why should a founder, head of marketing or product marketer care?
Because an agent is only as useful as the system you give it to work with.
Give it a page and ask, “How can we improve this?” and you’ll get slop opinions.
Give it structured buyer evidence, a controlled research method, versioned inputs and clear evaluation criteria, and it can help you investigate what your ICPs actually understand, trust and care about.
That’s also where synthetic research becomes really useful: it doesn’t mean inventing a few generic AI personas and asking whether they “like” your homepage. It means creating simulated buyers grounded in real ICP and voice-of-customer research, then using them to test specific questions before a human makes the final call.
In our teardown and messaging work, we use that principle + EP to:
→ Ground synthetic respondents in real customer interviews, sales calls, objections and buying triggers.
→ Define what the page or message needs the right buyer to understand.
→ Test what those buyers notice, remember, trust, question and would do next.
→ Compare messaging directions without losing the original research or rationale.
→ Turn repeated patterns into prioritised fixes, then review them against the source evidence before recommending anything.
I would never replace marketers with agents, but agents can help if we give them a research system that helps us make better decisions for the people we're trying to reach.
