In April 2025, Jeremy MacDonald was demoing automation and API workflows on a main stage as CEO of an SEO agency. Then, at another conference, someone showed a single image that changed his career: Amazon’s website in 1995. His takeaway was that AI wasn’t bad, it was early, and he left the agency world to start We Are Pioneers, which trains teams, leads AI transformation and builds bespoke tools. (This is part two of my series on his BrightonSEO talk. Part one was what a moon rocket teaches AI teams.)
The adoption curve
Two weeks later, speakers at another event showed the adoption curve for any technology, and Jeremy mapped Amazon onto it:
AI is on the same path. Most people are at stage one, asking how to do their work cheaper and faster. Stage two is doing it better. Stage three is doing work that wasn’t possible before. The problem, he said, is that many people stop at stage one. They used to run a technical audit by hand, now AI runs it, and they call it done.
That matches what I heard at the Heads of SEO roundtable: lots of agencies using AI to go faster at the same playbook, and lots of AI-generated audits that nobody reads.
What AI is actually good at
He gave three strengths:
- Following a defined process. Give it a clear one and it will understand it.
- Scale. It can do more than you could by hand.
- More data. It can use sources you wouldn’t normally touch.
The “if I had a week” exercise
Jeremy’s method for finding stage-three work uses keyword research as the example because it’s universal:
- Best version of the process: if you had a week, how would you do it? Maybe analyze Reddit sentiment, product information and white papers.
- Remove constraints: with two weeks and no limits, you’d use more sources and read more consumer feedback.
- Add data you can’t normally reach: maybe search volumes or social sentiment at a scale no team could read.
AI isn’t just for doing your job faster, he said. It’s for reimagining how you do it. If you want a head start on the automation side, here’s how to use n8n for SEO.
The catch
Once you work this way, you can produce 1,000 outputs where you used to produce five. When Jeremy managed a team of 50, he could train people and review their work by hand. At this volume he can’t, which is why the talk turns to testing. The next two posts in this series cover the how.
Try it this week
Pick one recurring task. Write the process you’d follow with unlimited time, list the data sources you’d add, and note which steps could be checked by an AI grader. That’s your first stage-three design.
Want help designing AI workflows that go past “faster”? Get in touch.
Leave a Reply