Artemis I launched on November 16, 2022, flew around the Moon and came back with no one on board, only a mannequin, Commander Moonikin Campos. Jeremy MacDonald, founder of We Are Pioneers and a former SEO agency CEO, opened his talk in my session at BrightonSEO 2026 with it. The shocking part, he said, wasn’t the price tag. It was that the whole mission was a test.
NASA also funds a facility whose only job is to test. The Katherine Johnson Independent Verification and Validation Facility in Fairmont, West Virginia, houses NASA’s IV&V program, which does independent analysis to find defects in safety- and mission-critical software. Jeremy’s point was that a team of roughly 270 people produces nothing; it only tests what others build. The program’s scope includes Artemis, Orion and the Space Launch System.
The cheer when the rocket crashes
He followed with a clip of a team cheering as a rocket landed and then crashed. Their excitement, he said, comes from knowing that failure and finding what’s wrong equals success. The test did its job.
A glance isn’t a check
His point for marketers was uncomfortable. When most of us use AI at work, we look at the output and think “yeah, that looks right.” That isn’t a check, he said, it’s a glance. It’s the opposite of the engineering mindset the cheering team has.
It echoes something I said at the Heads of SEO roundtable the same week: AI-generated audits fall flat when nobody really checks them. A glance is how slop gets shipped.
Why this matters more now
Jeremy described a shift in how work happens. We used to spend hours producing and owning the output. Now output is quick and plentiful, but testing it has become the harder part, and businesses he works with aren’t approaching the testing side strongly enough. His line from later in the talk captures the idea: testing isn’t the tax on scale, it’s what makes scale possible.
What to take from it
- Treat AI output as a test subject, not a finished product.
- Write down what “right” means before you review: relevance, accuracy, brand fit, compliance, whatever your work needs.
- Reward the people and systems that find problems, rather than the ones that ship fastest.
- Budget for testing as a permanent part of any AI workflow, not a pre-launch step.
This is the first of a short series on Jeremy’s talk. Later this week I’ll walk through how to build the first version of that testing layer.
Want help building quality checks into your AI content workflow? Get in touch.
Leave a Reply