Let Your AI Struggle
Why your AI gets better when you stop bailing it out
I let myself struggle with problems for longer than I need to.
I’ll spend hours with no breaks, 100% focused on the issue.
My wife will want a quick chat, and 90% of my thoughts stay stuck on why a bug is only happening in production when I should be able to reproduce it locally.
Work meetings come up. Instead of giving them my full attention, I’ll have a terminal running debug code in one window and the meeting in the other.
I’ll skip the daily walks I promised myself, because if I just give it ten more minutes I know I’ll figure it out.
I don’t actually enjoy these struggle blocks. But I can’t stop either.
Like any habit, good or bad, it’s driven by the payoff. In my case, it’s the eureka moment when the solution finally clicks after hours of struggle.
Why is that moment so special?
Because the frustration, uncertainty, and doubt you felt at the start get replaced with understanding and confidence.
You’re now a pro at this problem. You know exactly what went wrong and how to fix it next time. You’re on top of the world.
And now it’s AI’s time to struggle.
Each day AI gains more capabilities. Claude Desktop can access files on your machine. Manus can connect to your Google account and use Calendar, Docs, and Gmail on your behalf. OpenClaw can install its own software to finish the task you gave it.
Compare that to a year ago, when consumer AI was a glorified chatbot. It’s advanced incredibly fast.
But what does today’s AI have in common with the AI from a year ago?
The majority of users still act like helicopter parents when it comes to AI. They don’t trust it to do anything on its own. They hover, adding sandboxes, guardrails, and “human in the loop” processes — anything to keep the model on a leash.
Their heart is in the right place. Their strategy sucks.
Our school system has twisted what growth looks like. We think of school as the place kids grow, but schooling puts a heavier emphasis on perfection. You make honor roll by getting A’s all year. No mistakes. The “smart kids” are the ones who always got A’s — not the ones who had to work for them.
As you get older, you realize real growth comes from struggle. Nobody gets good at anything without first being bad at it.
We’re missing out on what AI can become by trying to prevent it from failing. It’s time to stop hovering.
When I was a boy, I worked on a horse farm. Every day I’d do chores — feeding, watering, cleaning stalls. I hadn’t grown up around horses, so I had a lot to learn.
The owner was always out there with me. But his help wasn’t what you’d expect. When I struggled and failed, he’d sit back, watch, and tell me what to do differently. He never did my job for me. He only corrected.
I’d be furious sometimes, not understanding why he wouldn’t just take over and finish the job. I was a kid. He was the adult. But he never would. No matter how many times I failed, he kept waiting for me to finish.
I didn’t appreciate it at the time, but he taught me one of the greatest lessons of my life. Persistence. Self-sufficiency. Nobody was going to fix my problems for me.
AI is going to fail. It’s going to fail at simple tasks. The differentiator between the people getting wild results and everyone else is that they keep pushing it anyway.
Day in and day out, I try to ask my AI agents to do everything for me.
Most of my time is spent in Claude Code. It thinks its job is to write the code and hand the rest off to me. It’s reserved about what it can and should do.
Things like installing software, running commands on servers, visually verifying work in the browser.
I used to do all of that myself. I didn’t trust AI either.
How could I trust something that constantly made mistakes to run a command against a production server with real customers on it?
Why would I ask Claude to do its own visual QA, when it kept missing obvious UI problems?
I didn’t realize it at the time, but those were the thoughts holding back my productivity.
Then I tried OpenClaw — an agent that runs on your local machine and is reachable through messaging apps like Telegram. My first interaction with it opened my eyes.
I was using it to redesign my personal site. I gave it full access to the admin section in the browser, we agreed on a direction, and it went to town.
The wild part? All of it happening inside a standard messaging app. No custom UI.
Full collaboration. OpenClaw would push design updates, send me a preview link, I’d approve, we’d ship.
It wasn’t until later that I realized OpenClaw was updating the production version of my site without me even noticing.
At some point it had decided that doing everything through a browser was too slow. So it grabbed an API key and started making the changes remotely.
That was the epiphany. Most agents hit a wall and immediately tap you on the shoulder. OpenClaw barely mentioned it. It was too busy getting the job done.
Agents are more capable than you think. You just need to let them struggle.
From that point on, whenever an agent gave up on a task and asked me to take over, my answer became: “No, you do it.”
And that has made all the difference in the world.
It’s no silver bullet. Asking the agent to do everything doesn’t automatically unlock a 10x boost.
Up front, you might even get a negative 2x return.
You’re going to run into the following:
The agent will do it wrong and you’ll have to do extra work to explain what right means.
The agent won’t have the tools it needs — maybe more software, maybe a connector to outside data.
The agent won’t have the permissions to finish the task.
Worst case, you hit all three at once and it would genuinely be faster to do it yourself.
This is the crucial moment. This is where you become the old farmer. Sit back. Wait. Do the bare minimum.
Resist the urge to jump in and do it yourself.
It may take multiple rounds of back-and-forth to get it right.
But the payoff on the other side is an agent that does the work autonomously. And once that happens for one task, it stacks. Each delegated task makes the next one easier.
It only comes after letting the agent struggle, and fighting the urge to take over.
---
The farmer probably had no idea what he was teaching me. He just stood there, day after day, watching a kid fumble through chores he could have finished in five minutes himself.
But he was teaching me to be someone who could do hard things on his own. And the only way to do that was to refuse to do them for me.
I still have my struggle blocks. I still skip the walks. I still leave my wife mid-sentence to chase a bug down. The eureka moments still feel as good as they always did.
The shift is that they’re not all mine anymore.
When my agent hits a wall and asks me to take over, I think about the farmer. I sit back. I tell it what to do differently. I wait.
And the agent comes back a pro.