Building
Two things I build and use every day. A personal AI chief of staff that knows my work, remembers it, and tells me what is slipping. And a course development platform at the University of Miami where a professor's own materials become a finished course. Both are early, and both are built one real use at a time.
A personal AI chief of staff
Most people use AI the way they use a search box: open a chat, ask, close it, and start from zero next time. Since spring 2026 I have run something different. A folder on my computer holds who I am, how I work, what I am working on, and who owes me what. My AI assistant reads that folder at the start of every session, so it opens already knowing me. I call it my chief of staff, because that is the job it does.
Four things separate it from a chat window. It persists: I stopped re-explaining myself. It routes: ask about a project and it reads that project's file before it answers. It compounds: every workflow I write down makes the next run faster and better. And it chases: it tells me that the thing I am waiting on has been sitting with someone else for eighteen days. That last part is what earns the name, and it is the part most people skip.
None of this depends on one vendor. The whole system is plain text files in a folder. Any AI assistant that can read files will work with it, and if I switch assistants tomorrow, I keep a well organized folder of my own work.
How to build your own
About 45 minutes for a working version. About six weeks of light, regular use before it changes how you work.
1
Pick one folder and keep it.
A synced drive is usually right, so it follows you between machines. Decide up front whether it holds work only, or work and personal. It will hold whatever you give it.
2
Let the assistant interview you, then write the first file.
Who you are, what fills your week, the tools you use, what you dread starting from a blank page, what you repeat, where things fall through the cracks, and how you want to be spoken to. Answer honestly; aspirational answers produce a generic assistant. The assistant turns the answers into one file it reads first, every time.
3
Give every ongoing project its own file.
Current state, who holds the next move, the date you will chase them, the next physical action. This is the memory. Update it at the end of each session, or have the assistant do it.
4
Write down anything you do twice.
A review, a slide deck, a weekly report: when you do it the second time, have the assistant capture the steps as a short recipe. Next time you say the name of the recipe instead of explaining it.
5
Make it chase.
Ask it, at the start of each day, for the list of things waiting on other people and how long they have waited. Then act on it. This is the step that turns an assistant into a chief of staff.
Three tips from running mine. Keep the first file short; a page is enough, and it grows with use. Keep everything readable by you, so the system stays yours. And start with one painful workflow and add the second when the first one works: the system is sharper in month six than in week one, and it gets there through use.
The guide has the starter prompt to paste on day one, the folder structure, the file that makes it chase, and the ways this usually fails.
I teach the full build, and the step from one assistant to a team of specialized agents, in a three-hour hands-on workshop. Next session September 28, 2026 →
A course development platform
A course development platform where a professor's own materials become a complete, interactive course, guided by their own Claude. It is an initiative I lead at the University of Miami, still early, built one real course at a time.
What the platform does
A professor's biggest obstacle in building a self-paced course is rarely the material. It is turning a folder of slides and notes into a finished course without deciding where each case or timeline belongs. My platform removes that step. A faculty member connects their own Claude, points it at their materials, and the AI proposes and builds it: modules, activities, a knowledge check, a certificate. The professor reads, corrects, and approves. I designed and built this myself, and I improve it course by course.
Use cases
Self-paced, multi-week certificates for working professionals, built from a faculty member's own course materials.
Turn a course's materials into an interactive, week-by-week review site: a companion learners use alongside the live class.
The same engine builds a private, branded program around one organization's own content and people.
How it works
A faculty member adds the platform as a connector inside their own Claude.
Claude reads the folder and proposes a course map for the professor to approve.
The AI builds every module: teaching, evidence, application, a knowledge check. The professor reads, corrects, and approves it.
Where it stands: an internal pilot at the University of Miami Business School, fall 2026, built and run by me with AI agents.
I am a professor at the University of Miami, and this is a university initiative, not an outside company. The software is a University of Miami innovation under review by the university's technology-transfer office: this page describes what I am building and learning, not an offer to buy or license anything.
Get in touch
If you lead a program, teach a course, or this interests you, I would like to hear from you.