Eric McPherson

If the answer involves masking 2-year-old children, it must be a really stupid question.

@HaimBender @NousResearch @OpenRouter No -- I didn’t make the Codesota report -- just sharing their public spend ranking. I only have the same 30d OpenRouter token data everyone else sees. No extra usage numbers on my end.

@EricMcPh @NousResearch @OpenRouter did u make this? do uhave something for openclaw hermez usage? except the 30d token usage on openrouter? bro 70m$ a month is nothing lol

Hermes Agent is now #1 on the Global @OpenRouter token rankings. While our journey together has just begun, we'd like to take this opportunity to thank our contributors, supporters, and users for all they have done to get us this far.

@HaimBender @NousResearch @OpenRouter "Apps ranked by monthly cost." --Codesota codesota.com/agentic/openro…

@EricMcPh @NousResearch @OpenRouter link to thing?

Ron Baron in new interview on the upcoming @SpaceX IPO: "I think the company over the next 10 or 15 years is going to be worth $10 trillion, $20 trillion, $30 trillion, and I could be very low." Baron Capital has invested a total of $1.7 billion in SpaceX over the years, which is now worth over $15 billion.

@RoundtableSpace You overlook mentioning that using the 1M token model “free” through OpenRouter means routing your sensitive data through Beijing. “Free” can sometimes be very costly.

HERMES AGENT JUST GOT A MASSIVE FREE UPGRADE AND ALMOST NOBODY IS TALKING ABOUT IT YET. 1M token context window, runs completely free on OpenRouter and can debug itself while you use it.

@pmarca Not if you launch them into space.

Datacenters : water destruction :: Pollution : paper straws

@cooverttech @DavidSacks Unnecessary if you build them in space, as SpaceX aims to do.

@DavidSacks Unless you're trying to turn Ohio into a desert, don't forget to add 10% to cover the 8 Billion Gallons per year required to cool it assuming evaporative. For a datacenter that large you'll either need a 1 square mile lake or plan to use rock quarries with a closed loop.

Back-of-envelope numbers for 1 gigawatt data center: All-in Capex: ~$50 bn Enterprise revenue generated: ~$25-30 bn/year Electricity cost: $1-2 bn/year ~2 year payback. The boom is real.

@DavidSacks @grok What percent of the AI-focused data centers market could SpaceX reasonably capture in 5 years? In 10 years?

I’m running both OpenClaw and Hermes... and that has not been my experience. OpenClaw has issues, but Hermes is not some stability refuge - I’ve seen plenty of serious problems there too, especially around context integrity, compaction, fallback behavior, and workspace isolation. perhaps switching may make sense for some people, but OpenClaw is the more capable and battle-tested system in my real-world use

@kilocode Yeah, no. Openclaw is the last thing you should choose if you value stability. It's the main reason to switch and after you get to learn team & community you will never look back.

Hermes vs. OpenClaw: Both are open source. Both connect to your chat apps. Both run tools and remember things. The feature checklists make them look identical. The design philosophies don't. Hermes packages a gateway around a learning agent. OpenClaw packages an agent around a messaging gateway. That one sentence determines almost everything else: how you configure them, how you talk to them, what breaks first. @olearycrew ran both for two months. He gives the full breakdown of where each one wins, where each one bites, and why he ended up running them both. blog.kilo.ai/p/hermes-vs-op…

Curious how you’d apply this to issue reports rather than PRs. We run a substantial OpenClaw deployment and sometimes notice bugs early. We don’t submit PRs, but we do use Opus to help turn our findings into clear issue reports. What makes an AI-assisted bug report most useful to maintainers, and least “slop”? A sample preferred format or prompt would be very helpful.

Did teach codex to look for social signals when reviewing PRs.

@garrytan Thank you for all you’re doing, Garry. Some of us genuinely appreciate it. Please keep it up! Particularly interested in GBrain.

The thesis is simple: the future belongs to individuals who build compounding AI systems, not to individuals who use corporate-owned centralized AI tools. I'm trying to build these in open source so you can have them for free. That's what GBrain is.

@dallasmavs @Chime Does Mike Schmitz seem like the kind of guy who would get duped into a secretive process to trade away for peanuts the best player in the NBA since Michael Jordan?

We have hired Mike Schmitz as our General Manager. @Chime // #MFFL

Atomic veterans describe what nukes feel like

Grok just got a massive productivity upgrade xAI officially launched Connectors on Grok today  Instead of copy-pasting between tabs, you can now link your apps directly to Grok for seamless, end-to-end workflows Just sign in to Grok web, click the + button, select Connectors, and click + Add connector to get started You can now connect: → Microsoft 365, SharePoint & OneDrive (read/edit files & docs) → Google Workspace & Outlook (draft emails, update calendars) → GitHub (search code, summarize PRs) → Notion & Linear (edit wikis, manage tasks) → Bring Your Own MCP (plug in custom internal tools) Live now on the web, and launching on Grok iOS and Android apps very soon

@sama @gitlawb Sam Altman casually greenlighting GPT-5.5 as the default for a fork of leaked Claude Code feels like something his lawyers should’ve gotten to first

@gitlawb cool thanks!

If @sama replied we will make GPT 5.5 as default model for OpenClaude (25.7K stars, 8.3K forks, 116K downloads) and Codex as default provider.

@TomSuazo @Molson_Hart The source is @NOAA as displayed on Google after performing the search "dallas average temperature by month". 😉

@EricMcPh @Molson_Hart Not sure where you pulled those temperatures but that is not indicative of the temperatures. Usually between April and Halloween the high is up to 115 and the low is 80. For 7-8 months straight.

Let me explain Dallas(s) 1. The summers are unbearable and this cannot be understood without moving there. You can't just visit Texas for 3 weeks in peak summer and understand it, because that's not why it's bad. It's bad because it's late October and it's 90+ and you're questioning your sanity as to whether not it will ever end. 2. Dallas' culture, superficially seems good, but once you dig down... 3. Dallas living is about the airport. You save money in Dallas and then fly out all the time because being in Dallas is rough. Airline travel in the US has declined a lot, so this way of life works less well. 4. The driving. It's dangerous. You need a tank and you're going to be sitting in it all the time. Dallas sort of seems like a city but it's not really. It's more like an area. 5. You can trade 3% state income tax for better weather, more trees, and fewer problems. I think Dallas is a solid place to live in a bigger house with A/C and grind for money aggressively for a short period of time and then move out of. Other than that, unless you are tied down there (job, family), I can't recommend it.

@bindureddy Betting against or underestimating @elonmusk and his companies has never worked out very well for anyone.

The AI lab that drops a better model consistently every month will win the AGI race - Google and Grok have to iterate quicker - OpenAI has to become more efficient - Anthropic has to be more consistent - Kimi and GLM have to become faster OpenAI in the closed and China in the open source arena are the current favorites

We’re huge fans of OpenClaw and everything Peter has built. It’s transformed how our team works. You guys are our heroes. That said, we’ve seen regressions with every release since v.2026.4.24 (v.2026.4.29 was especially rough). We’re getting non-responsiveness in group chats, huge unexpected token burn, crazy looping + duplicate answers, and responses that pull context from prompts we asked several turns earlier. We finally rolled back to v.2026.4.23 to get back to normal. For context, we’re running OpenClaw on several pristine dedicated 2026 Mac Minis M4 units along with Lossless Claw + QMD active + Token Juice. We run only API — Opus 4.6 as primary, and as fallbacks we use GPT 5.5 along with Kimi 2.6 and DeepSeek v4 via Fireworks. Gemini 2.5 Flash-Lite for LCM summarizing. We love the project and hope @steipete and @vincent_koc and the A+ team of maintainers can get this straightened out soon. We’re rooting hard for you and remain grateful.

Group chats now feel more natural: the agent can think, use tools, then deliberately send or edit the visible message instead of having the chat loop post the final assistant text. docs.openclaw.ai/channels/groups

OpenClaw 2026.4.29 🦞 💬 Group chats feel much better now 📌 Follow-up commitments from context 🔐 Safer exec, pairing, and owner controls 🟩 NVIDIA provider + model catalogs ⚡ Faster startup + plugin/channel fixes Group chat finally feels agent-native. github.com/openclaw/openc…

Peter, we’re huge fans of you and everything you’ve built with OpenClaw 🦞 … it’s been transformative for our team. That said, the last week of releases have been the most unstable we’ve seen since January. Our 16-person dev team relies on 6 OpenClaw agents for core work, and we’ve been at a complete standstill since Monday. Any eyes on this would be massively appreciated! Keep crushing it. We’re rooting for you.

@keenon Because software is hard? We're on it. Use an older version or install dependencies manually or use hackable install.

You reply to everybody but now you’re afk huh? I updated to 4.29 by clicking “update” button in web ui and broke everything . Why? @steipete why would you do this?