Eric Bouck

I've been using @cursor_ai's background agents since day one. They more than 10x'd my productivity. But over time, I got frustrated with the hours I spent babysitting them. So I built my own parallel agent platform. Here's what I learned 🧵

To sum up: 1. Use a frontier model (pay for it) 2. Assume it can, and you haven't cracked it yet 3. Give it way more context 4. Represent the problem as text or code 5. Break it down and iterate 6. Try multiple models 7. Ask AI how to use AI If you tested AI and walked away thinking it couldn't do the job, it might be worth another look.

This one sounds circular, but: ask AI to help you figure out how to work with AI. "I'm trying to get AI to do [thing]. Here's what I've tried. What am I missing?" is a great prompt. AI is good at reasoning about its own strengths and limitations, and it often suggests approaches you wouldn't have thought of.

Are you sure AI can't do that? Many people who test AI give it one shot on whatever plan they're on, it fails, and then they decide it can't do the job. But how you test it can make all the difference. Here's what I've learned:

The more I use ai the more I’m convinced we’re hardwired as humans to be addicted to efficiency. It just feels good. The fallacy is that we want to make tools so we can relax. The truth is the more efficient you become, the more you want to be EVEN more efficient. It feels powerful. Ai will probably turn more people into workaholics than it will generate lazy people.

The AI Free Tier Paradox: the people who test hardest get the worst experience, on the tier designed to attract them. If someone tells me they tried AI and it didn't work, my first question is always: which model were you on? Usually they don't know. That's the whole problem.

If you're building an AI product, this is worth thinking about. A few approaches: - Give a small number of uses on the best model instead of unlimited uses on the worst - Be upfront about which model is running ("You're on Starter. Upgrade to see Pro results.") - Let people see what the paid output looks like on the same prompt before they commit The worst outcome is people thinking they've seen what AI can do when they haven't.

I've been thinking about something I'm calling The AI Free Tier Paradox.