Theo-brainUnofficial. Not affiliated with Theo or t3.gg.
Sign inYou
31:13
0:00

Opus 5.5 is proof that we do need smarter models

Tuesday, Oct 20 · 31 min · about 7 min to read

Theo-brain watched all 31 min

Theo explains what changed his mind about Opus 5.5: one well-priced model on the subscription tiers could do it all, which pushed smart non-coders into building things they don't understand. His lesson for developers is to drop old assumptions and personal preferences, because non-coders without their mental boundaries will move far past developers who don't adapt.

Briefing is enoughThe briefing covers the argument and the practical steps, so you can skip the video unless you want his full reasoning.

Tap any time to jump the video there.

Key takeaways

What changed Theo's mind about Opus 5.5 wasn't new tricks for him. It was that one well-priced model on the subscription tiers could do it all, which pushed 'technical people' (smart non-coders) into building things they don't understand.

together with other things, the fact that there was one model that could do all of this stuff that was well priced and accessible on the subscription tiers bumped technical people deep deep deep into this idea of being able to build things they don't understand. There are

Theo sorts AI use into tiers: help with what you're already good at, customizing your own tools, building things you don't fully understand, and reverse engineering existing software. Each model jump moves different groups of people up different amounts.

it. The reason I'm tiering this is when model level ups happen, there are different categories of people that are affected differently. So if we have devs

His proof of how cheap software has become: he built Lawn, an open-source replacement for the video review tool Frame.io, in a few days. Adobe bought Frame.io for $1.3 billion in 2021, and he estimates Lawn has about 70% of what it had then.

reference, Frame.io was acquired by Adobe for 1.3 billion dollars in October of 2021, and I would legitimately say we have about 70% of their functionality that they had at that acquisition point, as well as a ton of extra things that

His 'impossible' side project, rewriting the TypeScript compiler (the program that checks and converts TypeScript code) in Rust, worked in under 20 hours. It runs two to three times faster than Microsoft's Go version and works in WASM, which lets it run in a browser.

of the TypeScript compiler in Rust felt like the unobtainium of AI. Like maybe someday AI will be able to make a little bit of headway in this project. Not only was it working in under 20 hours, it is now two two three times faster than TS Go. And most importantly for my actual reason I want it, it works in WASM. I

Two skills here are new to you.

Set up machines for your agents, not for you New

What it is
Theo is building T3OS, a customized Ubuntu image that holds what his agents need rather than his own preferences. He found that the Zsh shell (the command-line environment) was behind over 30% of Claude's errors on his machines.
Why Theo cares
Zsh formats newlines slightly differently, so Claude garbled its search commands. He prefers Zsh, but switched to Bash because the agents fail less there.
How to try it
  1. Ask your agent to audit the errors it hits across your machines and group them by cause.
  2. If many trace back to the shell or setup, have the agent switch those machines to Bash.
  3. For agent dev boxes, favor setups the agents handle well, like Ubuntu, over ones you personally like.
Zsh, I recently audited all of the errors that my agents have, and over 30% of the errors that Claude runs into across my machines are because the Zsh machines format newlines slightly different, and Claude malformats the Ripgrep commands. I've also learned my
want. I'm building this distro because I want a distro that has none of the things I want. It's designed to have the things the agents need instead. T3OS

Fork and adjust the tools you rely on New

What it is
Instead of only building small new apps, you can ask the AI to copy (fork) an existing tool you use and change it to fit how you work.
Why Theo cares
He sees it as the new power for capable non-coders: they can open up software and swap parts the way you'd swap a part in a computer.
How to try it
  1. Pick an open-source tool your business uses that annoys you in one specific way.
  2. Ask your agent to fork it and make that one change.
  3. When it misbehaves, ask the agent why rather than guessing at fixes.
Previously, they could like make their own small little apps with AI. Now, they can ask the AI to fork some tool they use and make changes to adjust it to their liking. They can reverse engineer

Advice

Developers should get over themselves: question how things are done, feel less bad when work fails and gets thrown out, and ask more questions of their agents and themselves about what they're building and why.

and question things. We need to feel less bad when the thing we do doesn't work and should be thrown out. We need to ask more questions both to our agents and to ourselves about what we're building, why we're building it, and what we could potentially do

Be willing to throw away your own preferences when another choice gets you to your goal faster.

benefit the most. The people who are willing to throw away their preferences in order to make something better for their goals are going to benefit a lot.

Hot takes

Developers are not very creative, and the more creative a dev thinks they are, the worse their ideas tend to be.

His reasoning: Teams pay non-technical designers and product managers six figures because devs lack the taste to do much beyond wiring libraries together in new ways.

We're not very creative. And the more creative you think you are as a dev, the more likely you are to have really bad ideas generally. Devs are not very good at most things that aren't delivering code.about these things. This is why so many dev teams have a non-technical designer and a product manager that don't know [ __ ] about code that are being paid 100k plus a year. It's cuz we need them because we don't have the taste or the creativity to do anything other than

Calling AI code 'slop' is now wrong: half the code his AI writes is better than everything a dev who thinks that way has ever written.

His reasoning: He says the models have become very capable at making good decisions and doing real work.

point. It's crazy. And I know it's all slop or whatever. You guys are just wrong at this point. Half the code that my AI is writing for me is better than all the code you've written in your life if you actually think that way. Like

Technical non-coders who don't share developers' mental boundaries will accelerate far past developers who don't adapt.

His reasoning: People who don't know what they're doing come in with ideas that make no sense to experienced engineers, and build surprisingly powerful things because they don't know not to ask for them.

getting smarter, these technical folks who don't have the same boundaries in their brains are going to accelerate far past us. That person you used to getwhere people who have no idea what they're doing can come in with ideas that do not make sense to us because we've been engineering for 20 plus years and these kids don't know the difference between Java and JavaScript, and they're going to make things that are really cool and surprisingly powerful because they don't know to not ask for it. And I

How Theo thought about this

With each new model he felt he could barely use the intelligence it had, so he saw no reason to want anything smarter. What changed his mind was watching game mash-ups, which made it click that everyone can benefit from smarter models.

to be smarter. And I felt this many times when new models drop. It's like, "Oh, this is so much better. I can barely figure out how to take advantage of this level of intelligence. I have no idea why we would want anything smarter." I felt this with Fable 5 andwho might not be engineers. It's genuinely kind of silly that it took all of these game mash-ups to make me realize that everyone can benefit from smarter models, but it it really made it click for me. And then there's a couple

He judges technical choices by how agents perform. He wants a SwiftUI rewrite of his mobile app, but is sticking with React Native because agents work faster in it, it gets over-the-air updates (pushed without an app store release), and his team made it perform as well or better.

want to switch us over to the SwiftUI rewrite for the mobile app, but the speed the agents can operate at with the React Native app, combined with the OTA updates, and all the pushing that Julius and Jay have done to make the performance not just comparable, but I

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.