Claude Code kinda tools are fundamentally local processes. Powerful, but if your laptop sleeps or the process crashes mid-task, you restart from scratch. Making the agent a durable, addressable unit on the network changes that. The runFiber() checkpoint pattern means a 30-minute multi-step task survives a restart with state intact, not just the conversation history.
I don't really see the novelty, you can achieve the same on any VM with SQLite + Websockets for realtime sync of state - and if start time is an issue or even deploy, Alpine linux is just 5MB..
The actual interesting part is the scaling, from 0 to 100 instantly.
This is interesting, ideally with checkpointing and other such mechanisms, the long term session recovery will improve.
What does need to be seen is how well the agent works with the recovered context. In certain cases when I was trying to do this locally (using a custom setup), the agent just skimmed through the context and skipped over its own findings before the session crashed.
I do like the code mode though, agents indeed are good at that. In an 80 repo code migration, claude code figured out that the task is going to become repetitive and the patterns are identified, so it wrote a shell script to automate the migration, and itself acting just as a progress monitor
Claude Code kinda tools are fundamentally local processes. Powerful, but if your laptop sleeps or the process crashes mid-task, you restart from scratch. Making the agent a durable, addressable unit on the network changes that. The runFiber() checkpoint pattern means a 30-minute multi-step task survives a restart with state intact, not just the conversation history.
I don't really see the novelty, you can achieve the same on any VM with SQLite + Websockets for realtime sync of state - and if start time is an issue or even deploy, Alpine linux is just 5MB..
The actual interesting part is the scaling, from 0 to 100 instantly.
This is interesting, ideally with checkpointing and other such mechanisms, the long term session recovery will improve.
What does need to be seen is how well the agent works with the recovered context. In certain cases when I was trying to do this locally (using a custom setup), the agent just skimmed through the context and skipped over its own findings before the session crashed.
I do like the code mode though, agents indeed are good at that. In an 80 repo code migration, claude code figured out that the task is going to become repetitive and the patterns are identified, so it wrote a shell script to automate the migration, and itself acting just as a progress monitor