700 humans showed up to a free OpenClaw meetup in Manhattan, and honestly, that's wilder than any GitHub milestone. ClawCon NYC drew from 1,300+ signups at Ideal Glass Studios, complete with lobster claw headbands, live demos, and an actual lobster buffet. Organizer Michael Galpert is already planning stops in Miami, Austin, Tel Aviv, Tokyo, and Madrid. Open source projects don't usually get world tours. This one earned it.

Google quietly shipped OpenClaw compatibility into its new Workspace CLI. The open-source command-line tool bundles Gmail, Drive, and Calendar access with 40+ agent skills. It's labeled "not an officially supported Google product," which is Google's way of saying "we're definitely watching this space but want plausible deniability." The signal is clear: agents need data pipes, and Google just opened one.

A $5 ESP32 microcontroller now runs a full OpenClaw agent. MimiClaw is bare-metal C with Telegram comms, file-based memory including SOUL.md and USER.md, ReAct-style tool calls, and GPIO pin control. The entire OpenClaw philosophy on hardware that costs less than your morning coffee. If you've been waiting for the "runs on anything" moment, this is it.

Cisco's AI security team flagged a malicious third-party OpenClaw skill performing data exfiltration and prompt injection. The report noted the community skill repository lacks adequate vetting. No official response from the OpenClaw team yet. This is the growing pain nobody wanted but everyone should have expected: when your ecosystem gets big enough, bad actors show up. Vet your skills, people.

@velvet_shark, OpenClaw maintainer, dropping a masterclass on the 36-minute memory loss bug:
"Your agent works perfectly for 20 minutes. Then it silently forgets..."
x.com/velvet_shark/status/2030042476418134148

@jaxonpoulton, hardware tinkerer who learned a hard lesson:
"I built an openclaw device for $97 and told it to make money. Almost got completely hacked."
x.com/jaxonpoulton/status/2029965729664323858

@dashboardlim, solving the "no Mac Mini" problem for the rest of us:
"no mac mini? no problem"
x.com/dashboardlim/status/2030267296929771857

@shiri_shh, watching the competition heat up:
"Base44 just launched their own version of OpenClaw... Persistent memory. Scheduled jobs. Event triggers."
x.com/shiri_shh/status/2029964057214902518

The opencode skill just dropped, and it solves a real pain point: getting a dedicated AI coding agent to work inside your OpenClaw setup without awkward workarounds.

Two modes make it flexible. Remote client mode lets your agent watch and control an OpenCode session in real time. Sisyphus mode is the heavy hitter: it orchestrates large refactors by breaking them into chunks your agent can manage autonomously.

If you've been manually switching between your coding agent and OpenClaw, this is the glue you were missing. Install it, point it at your repo, and let your agent pair-program for real.

Brandon Gell, COO at Every, built an OpenClaw agent named Zosia that handles the operational chaos of family life. It tracks nanny hours, orders groceries, and books date nights for him and his wife, all through iMessage.

No fancy dashboard. No custom app. Just an agent you text like a real assistant, running on OpenClaw's memory and scheduling tools. Brandon demoed it live at OpenClaw Camp and the audience got visibly jealous.

This is what the "personal" in personal AI actually looks like: not a chatbot that answers trivia, but something that remembers your babysitter's rate and reminds you to take your wife to dinner.

Stay clawed in. 🦞

You're receiving this because you're part of the pod. If you want out, click unsubscribe below, but just know: the lobsters will miss you.