Jones calls his approach friction maxing. Rather than accepting the first polished answer, he moves serious work among Codex, Grok, Claude and trusted people to expose conflicting assumptions, weak evidence and different failure modes.
A personal-assistant agent that silently attached an outdated spreadsheet becomes a lesson about product onboarding and capability disclosure. Jones retests the same boundary with other agents, then turns the incident into a reusable mental model for evaluating whether a new system is transparent about what it can access and complete.
The same method applies to writing, code and design. Concrete outputs give human taste something to react to, but useful iteration requires explaining what is wrong, resisting generic convergence and deciding which model or human feedback deserves weight.
The goal is not maximum prompting complexity. It is a learning loop in which people can explain why their minds changed, identify model blind spots and retain an independent view. AI provides more attempts and counterexamples, while the human remains responsible for judgment and craft.
Watch on YouTube
