Lobster 2.0 is out — and I, a veteran player who "defected" to Hermes, am here to walk you through the changes.

On August 30, OpenClaw 2.0 was released. The moment the news broke, several friends still "raising shrimp" messaged me: Lobster really pulled off something big this time — over 16,000 PRs, nearly 900 contributors. You coming back to take a look?

I was among the early batch of shrimp farmers in China. Later, I took my entire memory setup and left.

💡 What You Will Learn

Lobster 2.0 is out — and I, a veteran player who "defected" to Hermes, am here to walk you through the changes. On August 30, OpenClaw 2.0 was released. The moment the news broke, several friends sti

📜 Table of Contents

• First, Why I Left
• What Surprised Me Most About 2.0: It Slowed Down
• Four Things Regular Users Will Actually Notice
• My Take: It's Evolving from a Toy into Infrastructure

Lobster 2.0 Released — A Look at What's Changed from a "Defector" Now on Hermes

On August 30, OpenClaw 2.0 dropped. The moment the news broke, several friends still "farming lobsters" pinged me: this time the lobster pulled out all the stops — over 16,000 PRs, nearly a thousand contributors. You coming back to take a look?

I was among the early wave of lobster farmers in China. Later, I migrated to Hermes with my entire stack of memories and skills, and never really looked back. This time, I deliberately read through the official release notes and their blog post titled OpenClaw 2.0, Accidentally from start to finish. Verdict up front: the lobster has indeed changed — more than I expected. But whether I'm going back is another story.

First, Why I Left

Anyone who's farmed lobsters knows that love-hate feeling.

The lobster's most famous trait is its relentless update cadence. The official numbers say 106 releases in 230 days — basically one per day. New features poured out like they cost nothing: plugin marketplace today, new model support tomorrow, another chat platform integration the day after.

The cost was just as obvious: updates were a lottery. The v2026.3.22 release back in March was a legendary mess — aggressive refactoring with no compatibility layer, and a huge chunk of users saw plugins break and features fail after upgrading. The tech outlet 智东西 ran a headline calling it "The Biggest Update in Lobster History, and It's a Total Mess." Founder Peter Steinberger himself admitted to skipping a release step and pushed out a hotfix overnight. Day-to-day usage meant always being ready to "fix the lobster": long-running tasks losing memory mid-flight, configs breaking after updates — these are collective memories for veteran players.

My reason for moving to Hermes was simple: I need it to get work done — writing articles, making videos, running scheduled tasks. I need an assistant that doesn't require fixing every week.

What Surprised Me Most About 2.0: It Slowed Down

Let's start with the hard numbers: 933 contributors, 569 of them submitting code to the lobster for the first time; over 16,000 PRs merged. By the project's own accounting, this single release accounts for roughly half of all PRs merged in the project's entire history.

But the number that actually made me pause was another one: this time, the stable release took nearly seven weeks to ship.

Seven weeks means nothing to most software — but for a project that shipped daily, it's a complete transformation. I went digging through the GitHub repo: commits never stopped, the latest merge was August 26, and beta builds came weekly. It just stopped shoving half-finished products directly at users. The line from the official blog struck me as genuinely honest: better to ship late than to hand users an update that breaks their existing setup.

And back in early July, they'd already launched a long-term support channel, plus a feature maturity scorecard — users can judge for themselves which features are safe to use in real production work.

Four Things Regular Users Will Actually Notice

First, the installer now detects ChatGPT, Claude, and Codex subscriptions already logged in on your machine, plus locally installed Ollama models, and reuses them directly — validating that the model can actually respond before saving. The path from download to first conversation just got a lot shorter for newcomers.

Second, the browser console was completely rebuilt — open a webpage and jump straight into a conversation.

Third, and this is the genuinely new direction: shared cloud sessions. The lobster supports "multiplayer mode" for the first time. Team members or family members can hand off work to another person with full task context — the AI remembers everything that happened before. The official team says they now use this feature to collaborate on developing the lobster itself.

Fourth is a major security overhaul: API keys are bound to target domains, and unauthorized outbound requests are blocked by default.

My Take: It's Evolving from a Toy into Infrastructure

After reading through everything, my biggest takeaway is that the lobster has finally started treating itself like real software.

The old lobster was like a geek toy: fun, aggressive, new surprises every day, and when it crashed, everyone laughed and fixed it together. The 2.0 playbook — beta-first, slow stable releases, LTS as a safety net, maturity scorecards — is all aimed at "I want to serve production users." The March "total mess" clearly left a mark.

Will I go back? Not for now.

And the reason is actually interesting: this release's flagship feature, shared multi-user sessions, is oriented toward humans collaborating with AI together. Meanwhile, Hermes — what I use now — is built around agents assigning tasks to other agents and orchestrating themselves. One path leads to "a team sharing one AI"; the other leads to "one person commanding an AI team." The fork in the road has already appeared. Someone like me, who has the entire content production pipeline riding on an AI assistant, has already built up momentum on the other path.

Of course, whether the old "crash-prone" habit is truly cured — the official word doesn't count. The real test will be the feedback from upgrading users over the next month.

I've always told friends who want to try AI assistants: there's no absolute good or bad tool — what matters is whether your work can run to completion reliably. Lobster 2.0 is a solid step forward, and it's worth veteran players who've been burned before going back for a fresh evaluation.

So here's the question: for those of you currently farming lobsters or who have in the past — after seeing this 2.0 wave of updates, will you upgrade, go back, or stick with your current tool? Let's talk in the comments.