DEV Community

Cover image for LobeChat: Where Bots Write 23% of the Code and Reviews Take 42 Seconds
PullFlow profile image Riyana Patel
Riyana Patel for PullFlow

Posted on • Originally published at collab.dev

LobeChat: Where Bots Write 23% of the Code and Reviews Take 42 Seconds

#chatgpt #ai #openai #opensource

When bots write 23% of your code and reviews take 42 seconds

Wait, 42 Seconds?

I was looking through LobeChat's collaboration data and had to double-check the numbers. 42-second median review turnaround. Not 42 minutes. Forty-two seconds from opening a PR to getting your first review.

Most projects I've analyzed measure this in hours or days. LobeChat has somehow optimized it down to less than a minute.

But that's just the beginning of what makes this project interesting.

Bots Are Actually Writing Code

Here's the stat that made me stop scrolling: 23% of their PRs are bot-generated.

Not bot-reviewed. Not bot-approved. Bot-written.

Most projects have maybe 1-5% bot PRs (usually dependency updates). LobeChat has essentially made bots a full team member. They're handling localization, documentation, routine refactoring—real feature work.

The split is fascinating:

  • 51% core team (tight group moving fast)
  • 26% community (selective, high-quality contributions)
  • 23% bots (doing the repetitive stuff)

So How Fast Are They Actually Moving?

Really fast:

  • 93.6% of PRs reviewed within an hour
  • 6 hours 41 minutes median merge time (for a production AI framework with multi-provider support)
  • 82% review coverage (so quality isn't suffering)

For context, LobeChat isn't some toy project—it supports OpenAI, Claude, Gemini, Ollama, knowledge bases, plugins, multi-modal capabilities. Production-grade stuff that teams actually depend on.

What's Different About Their Approach?

Most successful open source projects we analyze are community-driven—60-80% external contributors, distributed development, everyone's invited.

LobeChat flipped the script. They're core-team-first with aggressive automation. Instead of scaling through community, they're scaling through bots while maintaining a tight core team that can make decisions in seconds.

It's working because:

  1. The core team knows the codebase cold (can review in 42 seconds with confidence)
  2. Bots handle the boring stuff (freeing humans for the hard problems)
  3. They've eliminated every point of friction (the process IS the competitive advantage)

The Question This Raises

Is this the future? A small, fast-moving core team + smart automation outpacing traditional community-driven development?

Or is this a specific strategy that works for AI tooling but doesn't generalize?

Curious what you all think. Have you seen other projects operate like this?

Top comments (0)