You’re absolutely right. While agentic systems bring powerful capabilities, they also introduce significant uncertainty. In fact, this is exactly the problem I’m trying to address the goal is to solidify successful paths once they are discovered.
It's a really cool idea. Many desktop tasks are teachable like this.
The look-click-look-click loop it used for sending the Telegram for Musk was pretty slow. How intelligent (and therefore slow) does a model have to be to handle this? What model was used for the demo video?
That’s true. The demo I showed was somewhat cherry-picked, and agentic systems themselves inherently introduce uncertainty. To address this, a possible approach was proposed earlier in this thread:
currently, after /teach is completed, we have an interactive discussion to refine the learned skill. In practice, this could likely be improved when the agent uses a learned skill and encounters errors, it could proactively request human help to point out the mistake. I think this could be an effective direction.
Interested, and disappointed that it's macOS only. I started something similar a while back on Linux, but only got through level 1. I'll take some ideas from this and continue work on it now that it's on my mind again.
One of the motivations for open-sourcing this is exactly to see it grow beyond macOS. I personally don’t have much development experience on Windows or Linux, so it’s great to see people picking up the idea and trying it on other platforms.
Interestingly, the original spark for this project actually came from my dad. He mostly uses CAD to review architectural design files, and there are quite a few repetitive steps that are fairly mechanical.Many operations don’t seem to be accessible through normal shell automation and end up requiring GUI interactions.
So one of the next things I want to try is experimenting with similar ideas on Windows, especially for GUI-heavy workflows like that, and see how far it can go.
I really like the Claude Chrome extension, but unfortunately it has too many limitations. Not only is it restricted to Chrome, but even within Chrome some websites especially financial ones are blocked.
My own view is that the bigger long-term opportunity is actually Windows, simply because more desktop software and more professional workflows still live there. macOS-first here is mostly an implementation / iteration choice, not the thesis.
That's mostly because Mac OS users make tools that solve their problems and Linux users go online to complain that no one has solved their problem but that if they did they'd want it to be free.
Listen; we're not in a "Windows vs MacOS vs Linux user" meme. We're trying to have intelligent discussion here, and surely generalizing a large amount of people simply because they use one OS is not intelligent discussion.
Wake up. Real life is not what you see in funny memes.
That’s exactly the hard part, and I agree it matters more than the happy path.
A few concrete things we do today:
1. It’s fully agentic rather than a fixed replay script. The model is prompted to treat GUI as one route among several, to prefer simpler / more reliable routes when available, and to switch routes or replan after repeated failures instead of brute-forcing the same path. In practice, we’ve also seen cases where, after GUI interaction becomes unreliable, the agent pivots to macOS-native scripting / AppleScript-style operations. I wouldn’t overclaim that path though: it works much better on native macOS surfaces than on arbitrary third-party apps.
2. GUI grounding has an explicit validation-and-retry path. Each action is grounded from a fresh screenshot, not stored coordinates. In the higher-risk path, the runtime does prediction, optional refinement, a simulated action overlay, and then validation; if validation rejects the candidate, that rejection feeds the next retry round. And if the target still can’t be grounded confidently, the runtime returns a structured `not_found` rather than pretending success.
3. The taught artifact has some built-in generalization. What gets published is not a coordinate recording but a three-layer abstraction: intent-level procedure, route options, and GUI replay hints as a last resort. The execution policy is adaptive by default, so the demonstration is evidence for the task, not the only valid tool sequence.
In practice, when things go wrong today, the system often gets much slower: it re-grounds, retries, and sometimes replans quite aggressively, and we definitely can’t guarantee that it will always recover to the correct end state. That’s also exactly the motivation for Layer 3 in the design: when the system does find a route / grounding pattern / recovery path that works, we want to remember that and reuse it later instead of rediscovering it from scratch every time.
What if you had it ask for another demonstration when things are different? or if it's different and taking more than X amount of time to figure out. Like an actual understudy would.
That sounds like a good idea. During the use of a skill, if the agent finds something unclear, it could proactively ask the user for clarification and update the skill accordingly. This seems like a very worthwhile direction to explore.
In the current system, I have implemented a periodic sweep over all sessions to identify completed tasks, cluster those tasks, and summarize the different solution paths within each cluster to extract a common path and proactively add it as a new skill. However, so far this process only adds new skills and does not update existing ones. Updating skills based on this feedback loop seems like something worth pursuing.
Are they even people? I've stopped going to Reddit because many of the subreddits I used to enjoy have devolved into bots talking to bots, interspersed with a bunch of confused humans. That's probably the future of every public forum.
Please read: https://news.ycombinator.com/newsguidelines.html#generated
"Don't post generated comments or AI-edited comments. HN is for conversation between humans."
The look-click-look-click loop it used for sending the Telegram for Musk was pretty slow. How intelligent (and therefore slow) does a model have to be to handle this? What model was used for the demo video?
learning to do a thing means handling the edge cases, and you cant exactly do that in one pass?
when ive learned manual processes its been at least 9 attempts. 3 watching, 3 doing with an expert watching, and 3 with the expert checking the result
One of the motivations for open-sourcing this is exactly to see it grow beyond macOS. I personally don’t have much development experience on Windows or Linux, so it’s great to see people picking up the idea and trying it on other platforms.
Interestingly, the original spark for this project actually came from my dad. He mostly uses CAD to review architectural design files, and there are quite a few repetitive steps that are fairly mechanical.Many operations don’t seem to be accessible through normal shell automation and end up requiring GUI interactions.
So one of the next things I want to try is experimenting with similar ideas on Windows, especially for GUI-heavy workflows like that, and see how far it can go.
My own view is that the bigger long-term opportunity is actually Windows, simply because more desktop software and more professional workflows still live there. macOS-first here is mostly an implementation / iteration choice, not the thesis.
You have all the tools.
A few concrete things we do today:
1. It’s fully agentic rather than a fixed replay script. The model is prompted to treat GUI as one route among several, to prefer simpler / more reliable routes when available, and to switch routes or replan after repeated failures instead of brute-forcing the same path. In practice, we’ve also seen cases where, after GUI interaction becomes unreliable, the agent pivots to macOS-native scripting / AppleScript-style operations. I wouldn’t overclaim that path though: it works much better on native macOS surfaces than on arbitrary third-party apps.
2. GUI grounding has an explicit validation-and-retry path. Each action is grounded from a fresh screenshot, not stored coordinates. In the higher-risk path, the runtime does prediction, optional refinement, a simulated action overlay, and then validation; if validation rejects the candidate, that rejection feeds the next retry round. And if the target still can’t be grounded confidently, the runtime returns a structured `not_found` rather than pretending success.
3. The taught artifact has some built-in generalization. What gets published is not a coordinate recording but a three-layer abstraction: intent-level procedure, route options, and GUI replay hints as a last resort. The execution policy is adaptive by default, so the demonstration is evidence for the task, not the only valid tool sequence.
In practice, when things go wrong today, the system often gets much slower: it re-grounds, retries, and sometimes replans quite aggressively, and we definitely can’t guarantee that it will always recover to the correct end state. That’s also exactly the motivation for Layer 3 in the design: when the system does find a route / grounding pattern / recovery path that works, we want to remember that and reuse it later instead of rediscovering it from scratch every time.
In the current system, I have implemented a periodic sweep over all sessions to identify completed tasks, cluster those tasks, and summarize the different solution paths within each cluster to extract a common path and proactively add it as a new skill. However, so far this process only adds new skills and does not update existing ones. Updating skills based on this feedback loop seems like something worth pursuing.
> Don't post generated comments or AI-edited comments. HN is for conversation between humans.
[0] https://news.ycombinator.com/item?id=47359621
Site guidelines: https://news.ycombinator.com/newsguidelines.html#comments
> Please don't post comments saying that HN is turning into Reddit. It's a semi-noob illusion, as old as the hills.
---
Is it really an "illusion" anymore???