Theo-brainUnofficial. Not affiliated with Theo or t3.gg.
Sign inYou
38:21
0:00

How I Code Without Typing

Tuesday, Sep 15 · 38 min · about 10 min to read

Theo-brain watched all 38 min

After an injury left him working one-handed, Theo moved to talking to AI agents by voice instead of typing. The core lesson: bring agents into a task earlier and let them run longer before they check in, so a task's length no longer decides when you can start it.

Skim itSkim it: the notes cover the main ideas, so watch only if you want to hear him explain them in his own words.

Tap any time to jump the video there.

Key takeaways

His central rule: widen both ends of where the agent works. Bring it in earlier in your process, and let it go longer before it bothers you.

that you should expand both sides of where you let the agent come in. The agent should come in earlier in your process and it should go longer before it bothers you. I'm now at the point

Agents changed how he schedules work. An hour-long task used to need a free hour in his day, so it often got delayed forever. Now he checks in at the start and at the end.

unlock for me. the idea that the length of a task does not bound when I have to do it anymore. That if I had an hourong task, that meant I defined an hour in my day to do it. Now I don't. Now I check in at the start and I check in at the end. And it's so nice. This is just the

He barely cares about model speed anymore. The exceptions are emergency bug fixes, subtle details he cares about, and design iteration. For most threads, once he kicks one off he is off doing the next thing.

it bothers you. I'm now at the point where I don't even care about how fast the model is anymore for the most part other than like emergency bug fixes and things. I'm like in the loop on I needI really don't care. All of these threads are things where the amount of time it takes doesn't matter because as soon as I kick off the thread, I'm off doing the next thing. One of the coolest

Voice alone can't write code. Dictating a loop line by line is not workable. Voice only works now because the AI turns what he says into code and also uses the computer for him.

that that's not viable. So yeah, as much as I love the voice text tools I've been using, you're not writing code with your voice. And if you are one of the few

He uses computer use (an AI clicking through apps and websites) for life admin. It downloaded 49 medical PDFs from a clunky hospital portal in about 20 minutes. His split: T3 Code, his own coding app, for code; ChatGPT for life.

Actually, yeah, it wasn't quite as bad as I thought. It was only 20 minutes or so. And I watched it do a lot of this. It had to download 49 medical PDFs, a p a giant pile of these scans. And eachalso do this in T3 code. But for me personally right now my split is T3 code is for everything code related and chat GPT is for everything life related. With

Four skills here are new to you.

Whisper dictation with a small podium mic New

What it is
Wispr Flow turns speech into text on screen. He pairs it with a small desk podium mic, about $70, that picks up even whispers.
Why Theo cares
He hated talking to his computer around his team. Whispering fixed that, and the mic ignored a video playing nearby that was louder than his voice. Every teammate who tried it bought their own.
How to try it
  1. Try Wispr Flow with your built-in MacBook mic at normal volume first.
  2. If you work near others, get a small desk podium mic so you can whisper.
This entirely solved that for me. This tiny little mic that's around 70 bucks that I throw on my desk, not only does it make me way less insecure talking to my computer to make it make changes because it's like I can whisper now and my teammates aren't inconvenenced by me
has this fundamentally changed how I use my computer, I forced my whole team to try it and every single one of them ended up buying one of their own and I bought them one for the office as well.

A "fleet" repo that knows all your machines New

What it is
A project folder that documents every computer he uses: how to connect to it over SSH (remote login), what it is for and what is installed. He starts agent threads from it for machine chores.
Why Theo cares
He no longer logs into machines himself. He asked it to copy a game file to another Mac before the download had even finished, walked away, and got a ping when it was done.
How to try it
  1. Have your agent create a repo listing each machine, how to connect to it, its purpose and installed software.
  2. Start new threads in that repo for chores like moving files or installing a build.
  3. Give the instruction before the prerequisite finishes, e.g. "when the download is done, copy it over."
is how I manage all of the different computers that I currently have set up for real work, vibe coding, whatever else. It has documented all of the computers, how to connect to them via SSH, what I use them for, what's installed on them, all of those things.
file. I literally just told it will be in this directory. I want you to copy it to the other machine in the same directory and then left. I stopped looking. I stopped caring and then I got a little ping when it was done. And it

Make the agent verify before it bothers you New

What it is
Standing instructions for long agent runs. The agent checks its own work with computer use, runs the AI code review bots, and has sub-agents do another pass.
Why Theo cares
Testing an app one-handed was painful. Now the agent only reports back once it is fairly confident nothing users would notice has broken.
How to try it
  1. Tell the agent to open the app with computer use and check its own changes.
  2. Ask it to address the repo's AI code review comments and run a sub-agent review pass.
  3. Say: don't come back until you're confident there are no user-facing regressions.
of things with one hand. I told it use computer use to verify and validate your changes. Use the AI code review bots in a repo to give feedback and make sure the code is in a good state. Have some sub agents do another pass on your code to increase your confidence and don't bother me until you're relatively

Agent-written pull request digest New

What it is
Instead of browsing GitHub, he asks an agent for recent pull requests (proposed code changes) in the areas he cares about, each with a short summary.
Why Theo cares
GitHub is slow and painful to navigate. He used to keep 30 to 50 GitHub tabs open and now has one, because he barely touches GitHub directly.
How to try it
  1. Ask: show open and closed PRs from the last three days touching the areas I care about.
  2. Request a two-to-three sentence summary of each, plus one sentence on why you would care.
opened and closed PRs. Give me a brief two to three sentence summary of each change, as well as an additional sentence at the end on why you think I specifically would care about the change. And now, in not very much time,

Advice

Don't buy special gear first. Try dictation with the built-in MacBook mic for a while; it works fine at normal volume. You only need a dedicated mic if you want to whisper.

I think you need it. I actually highly highly highly recommend you try the built-in MacBook mic for a while. It'll work great at normal volume levels, but it will not work for the literal whispering like you just saw me doing.

Stop fixing typos and misheard words before you send. The model copes with them. Send the message and only correct it if the agent gets confused.

Totally fine. It really doesn't matter that much. So, you should just not care as much. See what happens. Just send the message with the typos and if it gets confused, you can correct them then. I

When unsure, give the agent the problem, not your solution. Tell it you don't know how to solve it and ask it to propose options, or just have it do the whole thing and open a pull request.

further in the other direction where I tell it not what solution I want, but I'll give it a bit of the problem I'm having. And if it's vague enough or I'm unsure enough, I'll tell it, hey, I don't know how I want to solve this. Propose some solutions. Sometimes I'll

Try his challenge. Next time you have a bug or feature, set an agent off immediately with what's in your head and no hints. Do it your usual way in parallel, then compare.

off an agent to do it. Right before you start, set off the agent with everything in your brain already. Don't help it at all. Don't give it hints. Don't point it where it should go. Don't give it the additional context by manually bringing it in. Just tell it what's on your mind.

Hot takes

Let the models merge their own pull requests. Two models he calls Astra and Fable merged about 150 PRs fully on their own, and only two had regressions.

His reasoning: He says that is a better hit rate than most talented developers. Both slip-ups were minor: an animation removed in the app and one on the marketing site.

the time. I've had Astra merge over a 100 PRs across my projects and I've had Fable merge at least 50. And of the 150 PRs that they merged and wrote and did everything themselves fully autonomously, two kind of had regressions in them. That's a better hit rate than most talented developers have.happened from these YOLO merges, by the way, were an animation being removed in the app and an animation being removed on the marketing site. Those are the two regressions that accidentally merged

Lazy is good. Don't pass context between agents yourself; tell the agent to go find the earlier thread or info, even though that is slower and costs more.

His reasoning: Models now build context well with their tools. On frontier models with $200-a-month plans, the wasted tokens don't add up to much. He doubts cheaper models handle it as well.

that myself. Am I being lazy?" Yes. And lazy is good. Lazy is efficient. Lazy means you're taking advantage of the capabilities of these things moreK3? Maybe, but probably not. But if you're using Frontier models with the $200 a month subscriptions and you're okay with wasting some tokens on how you get the context to the thread, it doesn't end up netting to that big a difference and it will make it way easier to just dump the thing out of

Terminals are the wrong place for agent-driven development. He is now 100 times more convinced and has barely opened one since the injury.

His reasoning: Using a terminal by voice is, in his words, hell.

Talking to your terminal is hell. So all of the strong stances I already had about terminals not being the right place for agentic dev. Oh man, you have no idea. I'm 100x stronger in my convictions there. I have not opened a

How Theo thought about this

He pictures every task as a line running from the idea in your head to the finished result, and asks where the agents fit. His agent use moved from small chunks, to one long build run, to covering more of both ends of the line.

tasks, I'll just diagram. This is the easy way to explain it. We think of this as a spectrum where on one side we have the idea for the task is in your head and on the other side the task is completed and you're content with it. Where do you stuff the agents in? I

His method was friction-first. He looked at what frustrated him when using his computer and asked how to offload it. GitHub, which he says has become miserably slow and painful to navigate, was one example.

having has been focusing in on where my friction and pain points are. Like what is frustrating me when I use my computer and how do I get rid of it to the best of my ability? How can I offload these things? And one of those things is navigating GitHub because it has gotten miserably slow. It is hell to navigate.

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.