271–280 of 625 posts
Either the models are good and this sort of platform gets swept away, or they aren’t, and this sort of platform gets swept away.
The context for every single turn could in theory be nearly 1MB. Since this context is being stored in the repo and constantly changing, after a thousand turns, won't it make just doing a "git checkout" start to be really heavy?
For example, codex-cli stores every single context for a given session in a jsonl file (in .codex). I've easily got that file to hit 4 GB in size, just working for a few days; amusingly, codex-cli would then take many GB of RAM at startup. I ended up writing a script that trims the jsonl history automatically periodically. The latest codex-cli has an optional sqlite store for context state.
My guess is that by "context", Checkpoints doesn't actually mean the contents of the context window, but just distilled reasoning traces, which are more manageable... but still can be pretty large.
> Checkpoints are a new primitive that automatically captures agent context as first-class, versioned data in Git. When you commit code generated by an agent, Checkpoints capture the full session alongside the commit: the transcript, prompts, files touched, token usage, tool calls and more. This thread is extremely negative - if you can't see the value in this, I don't know what to tell you.
Sure... you `git add` the context text generated by AI and `git commit` it, could be useful. Is that worth 60 million?
Earlier quoted context omitted.
Sure... you `git add` the context text generated by AI and `git commit` it, could be useful. Is that worth 60 million?
It’s good to know that a few decades later the same generic Dropbox-weekend take can be made.
How is LangChain doing? How about OpenAI's Swarm or their Agent SDK or whatever they called it? AWS' agent-orchestrator? The crap ton of Agent Frameworks that came out 8-12 months ago? Anyone using any of these things today? Some poor souls built stuff on it, and the smart ones moved away, and some are stuck figuring out how to do complex sub-agent orchestration and handoffs when all you need apparently is a bunch of markdown files.
My experience is that Cursor's reliance on VS Code's clunky panel-based UI and jack-of-all-trades harness is holding it back. Likewise, Claude Code shoe-horning a GUI into a TUI and perma-binding to a single model family is not the ideal end-state.
The VC play here? The git context CLI thing is a foundational step that lays the groundwork for a full IDE/workflow tool, I guess.
> Checkpoints are a new primitive that automatically captures agent context as first-class, versioned data in Git. When you commit code generated by an agent, Checkpoints capture the full session alongside the commit: the transcript, prompts, files touched, token usage, tool calls and more. This thread is extremely negative - if you can't see the value in this, I don't know what to tell you.
[flagged]
> Checkpoints are a new primitive that automatically captures agent context as first-class, versioned data in Git. When you commit code generated by an agent, Checkpoints capture the full session alongside the commit: the transcript, prompts, files touched, token usage, tool calls and more. This thread is extremely negative - if you can't see the value in this, I don't know what to tell you.
Sure... you `git add` the context text generated by AI and `git commit` it, could be useful. Is that worth 60 million?
If it were also their last, I would be inclined to agree.
Feature plans generated by agents are often transient documents that fall away once the plan is executed. Ideally, that artifact would be preserved alongside the implementation. My experience is that Cursor's reliance on VS Code's clunky panel-based UI and jack-of-all-trades harness is holding it back. Likewise, Claude Code shoe-horning a GUI into a TUI and perma-binding to a single model family is not the ideal end-…
Earlier quoted context omitted.
Sure... you `git add` the context text generated by AI and `git commit` it, could be useful. Is that worth 60 million?
I love this one so much! The arbitrary decision to cherry-pick critique a particular product to this degree, when it’s something that could be said about 99% of the stuff SV churns out, including in all likelihood anything you’ve ever worked on.
```markdown # Run NNNN
## First Impressions [What state is the project in? What did the last agent leave?]
## Plan [What will you work on this iteration? Why?]
## Work Log [Fill this in as you work]
## Discoveries [What did you learn? What surprised you? What should the next agent know?]
## Summary [Fill this in before committing] ```
This is surprisingly effective and lets agents easily continue in progress work and understand past decisions.