I built a fast, browser based Markdown editor that automatically syncs to your GitHub.
- Obsidian inspired editing
- Highly performant
- Free and open source
Check it out: https://t.co/ykv0Db5Afl
Seems there’s two camps for when your coding agent’s context fills up.
a. “Don’t rely on compaction, just start a new session.”
b. “Compaction is good enough, no need to start a new session.”
What’s your default and why?
Working at big tech (Amazon) and then going home and doing AI side projects feels so strange.
At Amazon: “Yeah let me use this crappier internal AI coding tool and produce one PR that takes two days to get reviewed”
At home: “Ok I’ll spin up 10 codex/claude code instances and push 20 PRs in 5 hours.”
Humans really are the bottleneck now.
For software engineering to progress we can’t be bottlenecked by the human PR review process.
Code generation is no longer the bottleneck, code reviews are.
We need to change our mindset from:
Code gen -> Review -> Merge
To:
Code gen -> Merge -> Feedback/Steering
Human touch points shouldn’t block agent work.
On April 4, Anthropic blocked Claude subscription usage for third party harnesses like OpenClaw. I got frustrated and built my own with @drewmaely
If you're leaving OpenClaw, check it out: https://t.co/rbLC0UGLQU
https://t.co/IEeYvkZa9k
The problem with OpenClaw is it’s too “flexible”. i.e. not opinionated enough.
Letting users set it up “exactly” how they want has two problems:
1. Onboarding can be painful
2. The agent workspace bloats fast due to lack of structure
There needs to be a more opinionated personal agent harness.
Building one with MaelyClaw. Stay tuned.
Claude Managed Agents won’t win.
People won’t want mission-critical agents inside one vendor’s model, billing, auth and policy stack.
It’ll be multi-model, tool agnostic, and hopefully open source.
Anthropic blocked Claude subscription usage for OpenClaw. People moved OpenClaw onto another provider. OpenClaw got worse.
@drewmaely and I built a solution for this. Open sourcing tomorrow: MaelyClaw.