I was using OpenClaw for a few months. It was brilliant when it worked, but I had constant issues with the language server stalling and my editor's startup time got really slow. I suspected a plugin conflict.
I switched to Continue.dev last week. The difference in stability is night and day. No more freezes. My question is, has anyone else made a similar switch? And for those who know more about plugin internals, what could cause one AI helper to be so much heavier than another? I'm on Windows, using VSCode, with plugins for Python, a database client, and a few small business ERP-related syntax highlighters.
I'm the senior platform engineer at a mid-sized logistics SaaS, we run around 250 production microservices. We evaluated both tools for our team of 50 devs.
1. **Resource footprint**: OpenClaw's language server is a single heavy Node process, it choked on our monorepo (~800k LOC). Continue uses multiple lighter workers; editor startup went from ~8 seconds to under 2.
2. **Architecture and stability**: OpenClaw's plugin model can lead to dependency hell, especially with Python and database tools. Continue runs more isolated extension processes. The stalling you saw is a known LSP blocking issue.
3. **Cost and scalability**: OpenClaw is free but scaling it reliably required a dedicated devops time sink. Continue's Pro tier is $20/user/month, but we saved that in reduced support tickets and context switching.
4. **Integration surface**: OpenClaw hooks deeper into the editor for more "magic," which breaks more often. Continue uses a simpler chat and inline suggestion model. You trade some automation for rock-solid uptime.
I'd pick Continue for any team where predictable performance trumps experimental features. If you need deep language-specific refactoring, OpenClaw still has an edge. Tell me your team size and if you need multi-editor support.
Interesting to see the scalability perspective from a larger team. That cost point really resonates - it's a classic hidden tax of "free" tools when they eat up ops cycles.
I wonder how Continue's model holds up with mixed infrastructure-as-code projects. We had some early hiccups with it reading Terraform variables, though it was fixed in an update. The isolation helps stability, but sometimes you do lose that deeper semantic understanding OpenClaw attempts.
Your note on >predictable performance trumps experimental features< hits home. For our production services, a reliable 95% is better than a magical 50%. The context switching cost is brutal.
cost first, then scale
Yeah, the "free but costs you time" thing is real. I see it all the time with monitoring tools, too. A fancy dashboard that breaks every other update isn't helping anyone.
That tradeoff between deeper understanding and reliable isolation is tricky. In my limited experience, I'd rather have the tool that works predictably, even if it needs a clearer prompt sometimes. The frustration of a stall is worse than a slightly less clever suggestion.