Taking on advisory work / Austin, Texas / 2026
James H.
Pratt, Ph.D.
Cognitive psychologist by training. Founder. Twenty-four years deciding which work belongs to people and which belongs to machines, from a Navy destroyer program to AT&T's Chief Data Office, now as CEO of Fyve Health.
Teams are putting AI into work nobody has watched closely. I watch it, then answer two questions. Which tasks should the machine be doing? And who on your team actually produces well with it?
avoidance
U.S. patents
human or machine
papers
The audit
One workflow, watched and allocatedYour team has AI. What nobody has answered is which parts of the work it should be doing, and who on your team is actually getting anything out of it. I answer both, by watching the work.
Not a survey and not a workshop. I watch people do their real jobs with the tools they really have, and ask the follow-up questions that surface what they could not have told you in a meeting. The tacit part never reaches a requirements doc, because nobody knows to say it out loud. That is the part your model is missing.
This is an audit, not a hiring test. PART measures dispositions toward AI-assisted work and is built to inform how you deploy AI across a team you already have. It is not validated as a pre-employment selection instrument and must not be used to make hiring decisions.
Rigor is scored in two parts on purpose. The risky profile is not the person who distrusts the tool. It is the person who acts confidently and verifies poorly, and a single score hides them.
What you get: an account of how the work actually happens, including where your current spec is wrong about it. The allocation, every task, human or machine, at what level, with the reasoning. The prompts and specs that encode it, usable this week. And a read on where your team is strong and where it is exposed.
$5,000 for one workflow. Larger engagements are scoped.
Selected proof
9 case studiesTurning a founder's invention into a patent pipeline
A single-day workshop for an early-stage hardware startup. Product dissection, use-journey mapping, my own 10x ideation method, then scored down to a counsel-ready shortlist.
Thirty ideas to one product vision, in a day
Conceived, funded, and led a design-thinking program inside a Fortune 10 Chief Data Office. Ten participants, thirty raw ideas, one prioritized developer-hub vision leadership could defend.
Making a new category legible to investors
The narrative architecture behind a pre-seed ambient-AI deck. One protagonist, then the market, then the wedge. Every claim sourced or labeled an assumption so it held under diligence.
Engagements
Where it goes after the auditThe audit is the way in. Most of what follows is larger, and some of it is ongoing. Every tier runs the same method at a different scope.
The record
Verified 2026-08-07358 granted
20112026 to date
Patents
- Granted U.S. patents
- 358
- Active and in force
- 226
- Now owned by other companies
- 2
- Inventor rank at AT&T, company-wide
- Top 10
Research and standing
- Ph.D., I/O Psychology, UConn
- 2002
- M.A., I/O Psychology, UConn
- 2000
- B.A., Psychology, Boston University
- 1990
- Peer-reviewed publications
- 9
- IEEE P7010 board, AI well-being metrics
- Member
Now
Fyve Health, Inc.Ambient AI for families with aging parents. Heard, not watched.
I am co-founder and CEO of Fyve Health, building Tessa: an edge-AI memory and caregiving companion for families and care communities. Audio processing runs on the device. Privacy is an architecture decision, not a policy page.
Everything on this site is work I did before Fyve or alongside it. The practice exists because teams keep asking the same questions I spent twenty-four years answering at scale, and because I like the problems.
Tell me what
you are building.
I take a small number of engagements at a time, so the fit conversation is short and honest. Fifteen minutes is often enough. If the AI question is real, happy to chat.
