3 Topics I’m Exploring in My Life Right Now

Here are three topics I spend a ton of time thinking about right now.

1. Getting agents to do a buttload of work without breaking the bank

A big tangled job broken into chunk, check, and continue steps, leading to a robot, finish flag, and piggy bank.

I have this system I’m pretty proud of. At least I used to be.

It’s my /client-comms system, and it basically looks like this:

  1. Read all my client communication.
  2. Figure out who needs a response.
  3. Pull together the relevant context.
  4. Identify work we should do.
  5. Produce something meaningful.
  6. Draft a short, useful client update.

Sounds pretty good, right?

But then a couple Tuesdays ago, I blew my entire weekly 20x usage in one day.

And paid an extra $400 in credits.

So I broke the bank.

And failed.

And now I think a lot about how not to do that.

It’s basically the same with humans. You need to chunk things way down and have clear ways to evaluate as you go. Otherwise, agents can run amok.

2. The Holy Grail of AI Harnesses

Claude Code, Cursor, and Codex connected by hand-drawn wires to a golden grail.

At work, I think about AI more than anything else. I’m obsessed with it.

By “harnesses,” I mean the tools and setups we use to work with AI agents. These almost feel like the major social media platforms, but for building.

Which is way fucking cooler.

One of my favourite ways to learn is to ask the agents to research best practices for their own platforms, especially how to set up instruction files like AGENTS.md or CLAUDE.md.

Then create a cheat sheet for me to reference.

Then come up with a benchmark to score my existing setup.

Then apply it to our existing systems.

3. Oldschool Advertising

I just LOVE reading guys like Claude Hopkins, David Ogilvy, Albert Lasker, and Gary Halbert.

They use simple, fun language to share timeless lessons.

I’m reading Claude Hopkins’ My Life in Advertising right now. This bit stayed with me:

Try to hedge or protect yourself, and human nature likes to circumvent you. But remove all restrictions and say, “We trust you,” and human nature likes to justify that trust.

They don’t make em like they used to.

It would be great to connect with other people interested in these topics. If any of the above resonates, feel free to reach out!