For years, Iโve contemplated what my first tattoo would be.
While cruising at 40K feet, Gemini just about confirmed it.
{ prompt: nano banana - apply tattoo of the meal icon on my arm. below the icon, tattoo the phrase roughly translated in Chinese so it looks wise. }
Dude, I truly hope you can recover the data.
The shortfalls you called out w/Railway tokens/scopes for CLI and MCP -- this is true across many platforms (god-level tokens).
https://t.co/slLPXFxPrA has granular access controls that can be applied to any remote MCP server.
When in the hell did Google Meet get so good? Compared to Zoom, Google Meet calls used feel grainy and choppy. About 25% of my external invites are on Google Meet -- and those come from mostly SMB/mid-market sized companies.
Any enterprises you know that use Google Meet?
Vendor lock-in at the control plane is a whole different level of risk than at the model layer. And open-source isn't the answer, most teams don't have the bandwidth to self-host and maintain that. The real power is a neutral platform where you can swap and experiment across models as the landscape shifts without rebuilding everything underneath. That's what @BarndoorAI has been heads down on.
My favorite thing that has come out of OpenClaw is connecting (and reconnecting) with people to talk about it -- from all directions!
If you're using OpenClaw and want to learn how to easily and safely connect your apps/data, have questions about it more generally, or even better, have ideas/tips that you want to share as you've been embarking on making OpenClaw work better for you, please join us for a very informal and fun virtual office hours - today, Wed April 1 @ 2pm ET!
Get link to Zoom by visiting https://t.co/VckDCsgvUI
Just launched something on ProductHunt. While itโs not just for @openclaw, it does help you molt away the #BroAuth of janky integrations to Gmail, Jira, Salesforce, Notion, Slack, and more.
Think of it as your own personal smart MCP server with permission controls, usage analytics, and all of those platform tools available exactly when your agent needs them.
MCP nerds and your LLMd will dig Tool IQ. Dynamic tool selection, code execution sandbox, ginsu-knife skills (they get sharper the more you use them, legend has it).
And the new CLI fundamentalists can put down their pitchforks, because Venn also has a REST API to quell their anti-JSON RPC tendencies. ๐ชง
https://t.co/QkQXG9Dkp6
Those are some root level scopes there. I mean, imagine for a moment that Claude was convinced to call:
`POST https://t.co/lpvpy5Ibb7{calendarId}/clear`
No post body, a 204 response, and in a split second.. an empty primary calendar.
When we were going through Google app+security review for https://t.co/gYDB4JDdC3 (a safer way to connect your data to AI agents), we intentionally left out the massively destructive scopes. Not because we didn't think we could get approved, but because there's no valid reason to give anyone, or any AI, that much destructive capability.
Great post. Warning for non-dev folks -- be careful. This requires creating an OAuth app in Google Cloud console. This is not a normal or safe way by industry standards. https://t.co/gYDB4JDdC3 is free, security reviewed by Google and safely connects your Google apps to AI agents.
Works with Claude, Claude Code, ChatGPT, Cursor.. and even OpenClaw. Because it's just MCP.. but smarter and leaner. No context window bloat. Behind your one MCP endpoint are hundreds of tools that only get surfaced when your agent/LLM needs them.
Very cool tool, but it's definitely built for tech/dev folks. You can get the entire suite of Google Workspace tools (Gmail, Cal, Docs, Drive, Sheets) in a single secure MCP server endpoint, with a tiny context window footprint.. OAuth, and verified by Google - https://t.co/FTOILIWiHY
Yep, great share. BTW, this part also resonated strongly with me:
> โSometimes we over complicate things. Sometimes I've seen a solution over engineered.โ Instead, builders should ask themselves whether a given tool actually needs to be agentic.
When a non-deterministic part of your agentโs task should clearly be deterministic logic-based code, then replace it with such. Otherwise, youโre just burning tokens on something that doesnโt require a model at all.
It's all fun and games until your @openclaw token spend exceeds all of your subscription costs.. including Claude Max, HBO Max, Netflix, and Spotify.
https://t.co/XK9eo6qT4H
Getting @OpenClaw running on your machine is easy-peasy. It's a single npm install command, pop in an LLM API key, and bingo.
It took < 10 mins to install/config.. but only 5 minutes to realize I had to uninstall it from my work laptop immediately.
If you havenโt tried it yet because youโre terrified of getting pwnd, then good on you. But when you finally succumb to FOMO..
https://t.co/E9HqnWt7sm
@amit I'd wake up my kids for school every morning with same playlist.. for years.
Me: "Hey Google, play my Spotify playlist 'wake up' on kids speakers."
Selena Gomez: "โซ It's been said and done.. โซ"
Kids: <groan> "๐ฅฑHey Google, Stop!๐"
https://t.co/KVxHmUY7Bs
Developers are all-in on agentic AI & MCP. But for the enterprise, the fast-moving spec & ecosystem creates real risks re: security + governance.
I dig into this innovator's dilemma in my new post.
https://t.co/9ytZMYEEDH
#MCP#AgenticAI#EnterpriseAI
Excited to share that earlier this month I joined https://t.co/e4ZBmJzVSX. Today, we are officially announcing $13.6M in seed funding -- and our mission to help enterprises turn the promise of agentic AI into real results.
https://t.co/QVU3gGmfo2
#ai#agentic#mcp#security