Ray Villalobos ✝️

@noahzweben @bcherny Nice, If you make it indefinite loops you win

Loops now run for up to 7 days instead of 3. Let me know what you’re /looping on!

Claude updates are like those filenames I used to give prodects. Claude Claude Update Updated Update version of Claude Final New Update Version of Claude

@adocomplete Just wait until you can't remember what you even did and don't feel it until the next day you spring chicken you. ;)

You know you're not young any more when you bend over to pick up something and completely throw out your back. I'm in an Opus amount of pain right now 😰

Remember those good ole days when using AI meant having a conversation? You'd open ChatGPT, type a prompt, wait for a response, refine, repeat. Oh yeah, that was last week. I've got this whole content pipeline thing set up. My bot Otis reads my newsletters, then updates an Obsidian Kanban board for me with top 5 content priorities. I ask it to draft content and, then I jump into Claude Cowork to write scripts. Poor Otis by bot is working overtime and my Claude Cowork connects to my local hard drive to manage all my work. The chat era is over. The dispatch era is here.

@adocomplete 😭😭😭 Not worthy. ;)

If you received a Claude for OSS email giving you 6 months of Max but had trouble redeeming, please try again. Make sure to go through the entire flow to make sure usage is adjusted. Applying for the program? Please complete your GitHub profile before applying.

There’s an old story I love about a plumber who gets called to fix a problem that’s stumped everyone. He walks in, listens for a minute, taps a single pipe, and everything starts working perfectly. The client is furious when they see the bill and says, “You’re charging me that much for one tap?” The plumber replies, “You’re not paying for the tap. You’re paying for knowing where to tap.”

I’m no longer sure what it is I do that anyone with some very basic training couldn’t also do. This is of course not entirely true. But my skills used to be infinitely valuable and hard to replace. That is just no longer the case.

@burkeholland At least you know what localhost is

So far, I created several bots. Otis, Hoss, Flo and Slim. I think I'd like Cowork better if I could name it something. Does anyone agree? Or do you hate named bots?

I think Google rediscovered that it's moat it's...wait for it...it's context. Pretty much what it's always been, but by having it's AI build around it, it becomes an almost unpenetrable platfrom. It might not win with the best model. But doing it by putting AI where billions already work. Gemini in Maps, Gmail, Docs, Sheets, Drive, Meet, Photos, Android. The moat is presence. Platform distribution beats model performance any day. Build where users already are and the world is your oyster (or whatever clawed lobsters feed on).

Companies can't afford to miss the agentic era. That's why everyone is making CLIs, APIs, skills and gasp...MCPs even. I'm pretty surprised Google and Microsoft have both done this, but I shouldn't have. This next iteration of the web means business and only those granting access will make it.

Remember when we thought AI would look like the Terminator? Turns out we got it wrong. The machines are not coming with lasers and explosions. They are coming for our Slack messages and Jira tickets. Six weeks ago, OpenClaw was just another GitHub repo with a cool demo. Today it has 250,000+ stars and the creator just joined OpenAI. The gold rush is officially on. But here is the thing: OpenClaw is powerful, but also a security nightmare waiting to happen. Now we have an entire ecosystem of "clones" trying to fix that problem. Here are some of your current options. - NemoClaw (Nvidia) - Enterprise wrapper with sandboxing and audit logging. Huang says every company needs an OpenClaw strategy. This is Nvidia's bet on making it enterprise-safe. - Perplexity Computer - Cloud-based with a literal kill switch. Every action tracked, every sensitive task needs sign-off. You lose local file access but gain "someone else to blame if things go wrong." - Manus My Computer - Meta's desktop app for Mac and Windows. Every command requires individual approval. Slower than full autonomy, but exactly what cautious users need. - NanoClaw - Built in a weekend by Gavriel Cohen. The pitch: OpenClaw's 500K lines are too large to audit. NanoClaw does the same in hundreds of lines, with mandatory container isolation (Docker deal signed). - ZeroClaw - 3.4MB Rust binary. Boots in 10ms on 0.6GHz cores. Uses less than 5MB RAM. Runs on a $10 Raspberry Pi Zero. Built by Harvard and MIT students. - Moltis - 150K lines of Rust, zero unsafe code blocks. Voice I/O, Prometheus metrics, OpenTelemetry tracing. Enterprise observability for regulated environments. Each one attacks a different weakness of the original. - Maximum power? OpenClaw is still the default. - Security? NanoClaw has you covered. - Run it like it's 1999? Prince...ugh...I mean...ZeroClaw works on a Pi Zero. - Need enterprise blessing? NemoClaw has Nvidia's backing. Fortunately, the rise of the machines is not a dystopia. It is a toolset. It will make the people who learn to use it superhuman. Links: - OpenClaw: github.com/openclaw/openc… - Nvidia NemoClaw: blogs.nvidia.com/blog/gtc-2026-… - Perplexity Computer: perplexity.com/computer - Manus My Computer: manus.im/desktop - NanoClaw: github.com/qwibitai/nanoc… - ZeroClaw: github.com/zeroclaw-labs/… - Moltis: github.com/moltis-org/mol… The only question left: which claw do you choose?

Someone at work keeps on bugging me that since OpenClaw, I sort of got soured on CLIs. Here's the scoop. AI agents using MCP servers suffer from context bloat: extensive tool definitions consume tens of thousands of tokens. Me: Just give your agent a CLI. Progressive discovery through reduces token consumption and improves reliability. I don't get the resistance. The tradeoff: CLI runs locally and is pain to setup, worse at org-wide authentication. But for individual developers, the token savings are sweet. IDK. Am I wrong? Are you team CLI or MCP. I saw this cool balanced discussion and I agree there are some limited times, but to me CLI + Skill FTW. apideck.com/blog/mcp-serve…

@ryancarson @linear 100%, it’s the biggest mess in AI right now

The company that actually builds the agent-first code factory is going to be worth hundreds of billions of dollars. No one has cracked it yet. It can't be the model labs because then you're tied to one model. I'm hoping a company like @linear will do this. I'd happily pay thousands of dollars a month for that (+ the token cost). Basically, we need SDLC 2.0 for the agent age. (Also, the right solution can't rely on gh - we need whoever does this to completely replace it as well.)

If you want to use google services on your agents like Openclaw or Claude Code, then you've got two choices. Sign up for every single API: Worskspace, Gmail, Calendar, Sheets, Docs, etc. OR give your agents access to the single sign on Workspace CLI. At first, this looks like another Command Line Interface for nerds, but behind the terminal commands lies its real prowess, connect to anything, everywhere, all at once. And then you can issue commands that are just not possible, across a variety of surfaces with natural language commands. "Look through my conference materials and cross reference the speaker's list with the map of their locations and make sure their capacity matches their signups. Highlights any potential issues, then schedule a meeting with the stakeholders at the beginning of next week, best availability." ...Done. Oh yeah, the heck with prompting, working with agents just leveled up. Check this out on LinkedIn Learning! I have made it free for you with the link below. #artificialintelligence #ailiteracy linkedin.com/learning/ai-tr…

@gdb This is the sort of idea that I think is like handlebar mustaches. I wouldn't touch them with a 10 foot pole, but it looks good on you.

building a working game in excel, with a built-in game AI powered by complex formulas:

@johnlindquist What's an IDE? ::ducking::

I'm starting to forget my IDE shortcuts... I took _a lot_ of pride in those 😢

@elonmusk @peterwildeford I don't know if hyperbolic Elon is right, but he still is funny as heck.

@peterwildeford xAI will catch up this year and then exceed them all by such a long distance in 3 years that you will need the James Webb telescope to see who is in second place

Based on the data I see, I think: - Anthropic🇺🇸/Google🇺🇸/OpenAI🇺🇸 all ~tied - Meta🇺🇸 / xAI🇺🇸 each ~7mo behind - Moonshot🇨🇳/- Deepseek🇨🇳 / zAI 🇨🇳 / Alibaba🇨🇳each ~9mo behind - Mistral🇫🇷 ~1.5 years behind - No other companies competitive

Something is begging to become the GitHub of agent orchestration, there are so many options, it's overwhelming right now and difficult to decide. What do people like. I don't have the heart to try them all.

Is my attention deficit getting worst or better with AI. Now that I have a bot, I assign it something, then I might move into my coding platform, update something there, then I go write something, then remember I need to add something to my grocery list. The AI is making me MORE of a multitasker, but is that good?