How To Give Claude Unlimited Usage
Wednesday, Oct 28 · 39 min · about 9 min to read
Theo-brain watched all 39 min
Theo shows how he keeps Claude working on long-running projects like a Skate clone and TS Rust (TypeScript's compiler rewritten in Rust). The key is giving the agent a goal it can check by itself, then repeatedly asking what slows it down and fixing that. He says not to build this way yet.
Skim itSkim it for the screenshot-judging setup; the notes cover the rest of the method.
Tap any time to jump the video there.
Key takeaways
Agents do better the more a task resembles how they were trained: a clear goal plus a check that says whether it worked. Theo says you'd be amazed how good models are when given clear goals like that.
The more you make your task similar to the ways the model was trained, the better it will do at completing those tasks. So, how do we do this for
package TS Rust." You would be amazed at how good models are when you give them clear goals like that. That's why TS
For TS Rust the check was simple: match the official TypeScript outputs exactly. Once it worked, he set a speed goal. The first version was slower than the official compiler; he said don't stop until it's twice as fast, and it got there.
tests still pass." When I did TS Rust, the first working version was slower than TS Go, which is the official TypeScript. I told it, "Do not stop until you're two times faster," and it kept going until it was two times faster. Because it knows how to check if it worked, and it knows how to check if it's fast.
The check should not need a human. A person reviewing each attempt stalls progress. TS Rust had 7,884 commits during its loops. Theo wants code that says pass or fail, or at least another agent judging the results.
there waiting. TS Rust had like 1,800 revisions or so. 7,884 commits were filed on TS Rust during the loops. You need some way to programmatically verify with code that says pass or fail, or at the absolute least, a way that another agent can look at the results and decide if it's working or not. And then we have step
Once the goal was set, the bottleneck was computers, not the AI. Using Boat cut the wait per test run (a 'gate') from about 70 minutes to about 6. That more than doubled, close to tripled, how much code landed each day.
when I say me, I mean my agents. And it knocked the time per gate down from 70 minutes to 353 seconds, only 6 minutes. It more than doubled the amount of code I could land in a given day. It's actually closer to a triple.
Theo says running agents is like being a manager. The job is not checking every line. It is setting agents up to succeed, asking what slows them down, and improving their environment.
my subs pretty quick. When people say that running agents is like being a manager, this is what we're referring to. It's not your job to check every single piece of work your employees do. It's your job to set them up for success, ask them what's slowing them down, and keep building the best environment for them to be in. And ready
Two skills here are new to you.
Screenshot comparison judged by an AI panel New
- What it is
- A way to make a fuzzy goal like 'looks as good as the real game' checkable. After each build it is happy with, the agent screenshots set moments, compares them to real Skate footage, and a panel of AI judges says whether his version looks good enough.
- Why Theo cares
- It gave the model something to chase. He admits quirks: he thinks at least $800 in tokens went on making the skybox look like Skate, and it kept brightening the game until he reset it to use Skate 1 as the reference.
- How to try it
- Pick a reference your project should resemble, such as a competitor's screens.
- Have your agent screenshot each good build at the same moments as the reference.
- Have other AI models compare each pair and give feedback the agent works from.
I find it. What I eventually effectively did is I told the model that every time it has a build that it is happy with, it should take a bunch of screenshots of it in specific states, compare those screenshots to real Skate footage from the real games, and then have a panel of judges compare the original game with my game, and give feedback as to whether or not mine looks good enough. And now we
pretty sure this spent at least $800 in tokens trying to get the skybox to look more similar to Skate. It also kept trying to make the game brighter cuz I was comparing it to Skate 3 originally and I think Skate 3's graphics are hideous. So I reset it and told it like no, Skate 1 is the thing to reference. I
A separate check-in thread New
- What it is
- Theo ran one agent thread on the endless goal and a second one he used to ask for status updates and what was slowing the work.
- Why Theo cares
- When progress slowed, the check-in thread revealed the real blockers. Too many jobs crashed the machine, so the agent built a queue. Then it needed more computers.
- How to try it
- Keep a second chat that reads the main agent's history.
- Ask it what is currently bottlenecking the work and what the biggest problems are.
- Fix the top blocker, then ask again.
update on what's going on. Eventually I noticed that progress was slowing. So I asked the second thread, "Hey, what is currently bottlenecking the work being done? What are the biggest problems?" To
Advice
Run these long loops on $200 Claude subscriptions you can abuse, not at pay-per-use API prices. The costs he quotes are API prices, which is not what he actually pays.
usage through Claude. This is all API costs. That's not what we're paying. We're all paying sub prices. Don't do this over API prices. Do this with $200 Claude subs that you can abuse because
Lean into tedious work. A job that would be miserable for a person, like rewriting a 100,000-file codebase, is what agents are good at grinding through.
need to lean into it. The more tedious a task I can come up with, the more likely it is that the models can take advantage of that and do crazy things. Rewriting a 100,000-plus file code base in Rust would be miserable as a human, but an agent doesn't care. So, find something
Don't have agents use Blacksmith, a service for automated build-and-test runs (CI), in their day-to-day loop unless they are just running existing checks. He says it is built for serious projects.
your insights really good. I personally would not recommend that you tell your agents to use Blacksmith in their local dev loop, unless it is just checking existing CI stuff. And here is where I
Don't build this way today. Theo says it has been hellish, even though he loves it.
crazy things. So, do I think y'all should be building this way today? Absolutely not. Like seriously don't. This has been hellish. I have found so
Hot takes
Claude Code's /goal command, which keeps an agent running until a goal is met, is a foot gun you shouldn't use. Yet he used it for the Skate clone because he was knowingly going overboard.
His reasoning: He calls it a tool that is only there to hurt you, but fun if you're trying to get hurt.
talked a lot of crap on slash goal in the past. I still don't think you should use it. It is a foot gun that doesn't work when you point it up. This is a tool that is only there to hurt you, but if you're trying to get hurt, it's actually quite fun. So, slash goal.
Spending about $38,000 worth of API-priced usage on TS Rust was a good deal.
His reasoning: It will save his team many hours of build time in real work. He calls his other projects nonsense.
about it. Um TS Rust only being 38,000 dollars is worth it. Considering how many hours this will save us in build times for our real-world work, 38 grand is actually a good deal for TS Rust. The rest of these are nonsense though. But
Loops like these will be cheap within about a year, except for the computers they run on, so the skill is worth learning now.
His reasoning: He says intelligence currently gets four to five times cheaper every month.
now. And remember, intelligence gets four to five times cheaper a month right now, which means that around this time next year, loops like this are going to be pretty cheap except for the compute that they're running on. So if you can find the opportunities to build these skills today, they will be more and more useful over time because the way that we
How Theo thought about this
He chose 'good enough' checking over perfect checking. His plan was to record his own controller play and have agents replay it. Out of laziness he only had them compare screenshots, and that worked. His rule: the check only has to line up closely enough with what you actually want.
It gave the model something to pursue and work on and as long as it has something it can check and decide if it's good enough or not and then keep going if it's not, you can get the model running for a while. And as long as the thing it is checking is aligned enough with what your goals are, you're going to end up with something at least somewhat resembling what you asked for.
He followed each bottleneck one layer deeper. Too many jobs crashed the machine, so the agent built a queue. It said it needed more compute, so he brought spare computers online, then moved to Blacksmith. Builds still took 20–40 minutes, which inspired Lust, a faster Rust compiler.
would oom and die. So, I asked it how to fix that. It created the queue. I asked it how to speed things up, and it told me it needed more compute. So, I brought all the computers I had around online, gave it SSH access, and let it split the work across those. Helped. I moved over to Blacksmith cuz I wanted the compute back. Helped even more. And then,
faster. But, each of these jobs was still taking way too long. We're talking 20 to 40 minutes for my builds, just to get the code in a state where I can check if it worked or not. That inspired me to do my other insane YOLO project. And that's why Lust exists, because I didn't want TS Rust to take so goddamn long to compile. But, I promised you
Sign in to mark what you've watched and keep notes.
Have a question about this one? Ask about this video →
Theo Daily is unofficial and not affiliated with Theo or t3.gg. Written by Claude Opus 5.5 from each video's transcript.
