A letter from Brent Vosters · revised September 2026

Where to start with AI.

I thought I was reasonably fluent in AI. Then a friend, Alex Schwartz, generously offered me a seat at his four-hour AI boot camp, and I was lucky enough to sit next to Matty Squarzoni, a serious builder. It did not take long to see that I was not savvy at all. I was the typical end user, treating ChatGPT and Claude as a glorified search box.

I am curious by nature, so that afternoon sent me down a series of rabbit holes, and I have been tinkering ever since. What I should have done first was take the fundamental courses I lay out below. It would have saved me real time.

I get a lot of questions from friends about how to get more involved with AI and make sure they are keeping up with this new world. Many of them have kids in high school or college asking the same thing. There are plenty of resources out there, and I do not pretend mine is the best list. I just thought it might be a helpful place to get started if you are feeling lost or frustrated. I learn something new every day from talking to people, so treat this as a living page. It will keep changing as I come across more. None of it is investment advice.

Feeling stuck? Start with the two free schools below and get your feet wet. You can build from there.

Where I'm working these days

I built a second brain. It turned out to be a relationship index.

Everything I capture in a day, meeting notes, mail, messages, the podcasts I listen to, goes into one place at home where a model can read it. Three things happen to it there. It becomes searchable. It lands in Notion, where I can browse it. And it feeds a graph of the people in my life, which is the part that matters. The system tells me who I owe a reply, whose paths should cross, and what I should know before I see someone. I talk to it from my phone. It only ever tells me things I already knew and had forgotten.

In

  • Meeting notes
  • Email
  • Calendar
  • Texts
  • Photos and screenshots
  • Podcasts
  • Voice notes
  • Links and bookmarks

The brain

  • Searchable
  • Browsable, in Notion
  • Connected, a graph of people
  • A Mac mini at home

Out

  • The 7:30 morning brief
  • A briefing before every meeting
  • Who I owe a reply
  • Whose paths should cross
  • Ask it anything, from my phone

Built over a summer of evenings with Claude Code, no engineering background, starting from a single question: who have I not written back to?

Why bother

Three kinds of people are forming. Be the third.

The divide is not between people who know about AI and people who do not. It is between three postures, and only one of them ages well.

I

The purist

Rejected the tools on principle, and a quarter of the world with them.

II

The automaton

Outsourced every decision. Cannot tell any more where the tool ends and they begin.

III

The third kind

Runs the whole operation with the tools and keeps judgment and taste for themselves. This page is for them.

After Dan Koe, We are in the middle of the digital renaissance, August 2026. His frame, my words. His letter is worth the twenty minutes.

Before you start

Four things worth knowing on day one.

A model is not a harness.

The model is the engine: Claude, GPT, Gemini. The harness is the car built around it: the chat app, a coding agent like Claude Code or Codex, a workflow someone wired up. Most of what feels like progress in 2026 is harness.

Models have strengths.

Some are far better at images, some at code, some at long research, some at plain writing. Pick by task, and try two on the same job before you settle.

Context is the whole game.

What you put in front of the model, the brief, the documents, an example of good, decides what comes back. Vague in, vague out.

Everything is a first draft.

Treat the first answer as a draft you edit, not a verdict you accept. The people who get the most out of these tools are the ones who push back.

Start here

Two free schools. These alone will get you up to speed.

Both labs now run proper courses, free, current, and maintained by the people who build the tools. Take them in the order below, one from each column a week, and in three weeks you will be ahead of most of the room. Everything after this section is optional.

Anthropic Academy

Starts with posture and judgment, then moves to the tools. Four courses.

  1. AI Fluency: Framework & FoundationsThe four habits that separate fluent users from casual ones. Everyone starts here.
  2. Claude 101Everyday work: writing, research, analysis, files.
  3. Introduction to Claude CoworkWorking alongside the model on your real files and projects. No code.
  4. Claude Code 101, then Claude Code in ActionOnly when you want to build. This is the door I walked through.
Full catalog, 22 courses, including AI Fluency for students →
OpenAI Academy

Starts with the tools. Three short courses, each about an hour.

  1. AI FoundationsFor people new to AI. Clear instructions, useful context, reviewing what comes back.
  2. Applied AI FoundationsTurn one recurring task of yours into a repeatable workflow, with review points.
  3. Agents and WorkflowsDirecting agents through structured work. Take it after the first two.
Full catalog →

Then, if you want more

An hour, a weekend, or a month.

One hour Watch two talks. Karpathy at Sequoia, then 3Blue1Brown on compression April and June 2026 · 30 + 32 min

The first is how the best people work with these systems now. The second is why prediction and intelligence turn out to be the same thing. An hour, and you are current.

 One weekend Read one essay, then work. The Adolescence of Technology, then two days of real work Dario Amodei · January 2026 · then a paid model

Saturday the essay, twenty thousand words from the person running one of the labs. Sunday your actual work, run through a paid model. The essay gives you the stakes. The day gives you the habit.

 One month Build one small tool. Something you will use every day Claude Code, Codex or Cursor · No code background needed

A meeting-notes filer. A reading tracker. Describe it, let the tool write it, read what it wrote. That loop is how everything I run got built.

The library

Fill-ins, in four rooms.

One-off talks, essays and guides that go deeper than the schools do, grouped by what you are after. None of them are required.

Understand

A working picture of how they are trained changes how you talk to them.

Use

The gap between a casual user and a fluent one is larger than the gap between models.

Build

Once the tools could write and run code from a plain description, building became a way of taking notes. Start at the top, go down the list as you grow.

Follow

Four sources, deliberately few, all still current in September 2026. None of them shout.

If you are twenty
  1. Learn to write clearly.The models reward a clear instruction. So does every person you will ever work for.
  2. Build one thing in public.However small. A working tool with your name on it says more than any line on a resume.
  3. Pick a domain. Go deep.Medicine, law, real estate, music, markets. The AI half is now the easier half to learn.

Anthropic runs a free AI Fluency for students course. If you already code, Karpathy's nanochat is the month-long project: train your own small ChatGPT for about a hundred dollars.

Ask me

If you read something here and want to talk it through, write to me. I read everything.

Emailbrent@vosters.ai→ Homevosters.ai→

Brent Vosters is a partner at Benesch Investment Group, a wealth management firm in Coconut Grove, Miami. He lives in Coral Gables. More at vosters.ai.