This project wouldn’t exist without the people and ideas below.
The idea that knowledge workers should train like athletes — with defined skills, structured feedback, and actual reps — comes from Nate’s video on deliberate practice for knowledge workers. He made the case that AI finally makes the feedback infrastructure cheap enough to build practice environments for cognitive skills. That framing is what turned a fifteen-year-old research interest into something I actually built.
Nate traces the idea back through a lineage worth following: David Perell’s essay “Learn Like an Athlete” argued that knowledge workers should structure their skill development with the same rigor athletes bring to training. Tyler Cowen then sharpened the challenge into a question that sticks: what is it you do to train that is comparable to a pianist practicing scales?
Whether the “train like an athlete” framing fully applies to therapeutic skill — which is deeply interpersonal, emotionally embodied, and unfolds over long timeframes — is an open question this project is exploring.
Jonathan Edwards built AI·Cred using Nate’s prompt engineering — an adaptive assessment where you describe how you’d actually use AI, and it keeps probing deeper. It pushes on your reasoning, tests your instincts, finds the edges of what you know and don’t know. Then it gives you a score and a personalized learning plan that builds on exactly the gaps it found.
The FIS Lab borrows that same logic but applies it to therapeutic skills instead of AI fluency. Where AI·Cred has you respond to AI scenarios and probes how you think about them, this tool has you respond to client scenarios and gives you layered feedback on how you handled them. Both adapt to you. Both map your patterns rather than just handing you a grade. Seeing AI·Cred work is what made this feel buildable.
The FIS framework comes from Anderson, Ogles, Patterson, Lambert, & Vermeersch (2009), who showed that therapist effectiveness can be predicted by measurable interpersonal skills — not credentials, not years of experience, not theoretical orientation. That finding is robust and well-replicated.
What’s less established is whether these skills can be meaningfully trained. The best available evidence — from the Calgary Counselling Centre — showed a small positive effect (d = 0.035 per year) from a comprehensive, multi-year program combining outcome monitoring with deliberate practice.
This tool exists in the gap between what we can measure and what we can change. The measurement science is strong. The training science is early. We’re honest about that.
For a fuller account of the evidence and its limitations, see the research page.
Built by Makoto Suwamoto · AI·Cred profile
What this is and isn’t: A practice and self-reflection tool that uses AI to give you directional feedback on your interpersonal responses. Not a validated clinical assessment, a training program with proven outcomes, a credential, or a replacement for supervision.