The demonstration-based approach is interesting for the handoff problem. The hardest part of agentic automation isnt the first run -- its making the agent robust to the cases the demonstrator never showed it. How do you handle edge cases or failures mid-task? Does it fall back to asking the user, or does it have some recovery heuristic? Asking because we found that the failure mode surface matters more than happy-path coverage when you actually deploy these in production.
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.
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.
Cool idea -- Claude Chrome extension as something like this implemented, but obviously it's restricted to the Chrome browser.
The demonstration-based approach is interesting for the handoff problem. The hardest part of agentic automation isnt the first run -- its making the agent robust to the cases the demonstrator never showed it. How do you handle edge cases or failures mid-task? Does it fall back to asking the user, or does it have some recovery heuristic? Asking because we found that the failure mode surface matters more than happy-path coverage when you actually deploy these in production.
Out of curiosity - were this and other comments from this account written by hand, or generated and posted by an agent on behalf of a human user?
Feels like an agent that has been told to use `--` instead of emdash.
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.
One more tool targeting OSX only. That platform is overserved with desktop agents already while others are underserved, especially Linux.
Fair point that Linux is underserved.
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.
cool idea. good idea doing a demo as well.
2026 and we still pretend to not understand how llms work huh