The bottleneck in agentic AI just moved again, and this time it's endurance, not intelligence. For two years the question was which model is smartest. Now it's which agent you can leave alone for a week and trust to still be working when you check back. A feature flag reading gpt-6-astra-aeon turned up in Codex Desktop on September 3, the day OpenAI shipped GPT-6 Astra at ten and fifty dollars per million tokens. The Agents API went into public beta September 10. DevDay is September 29, three days out. Nobody at OpenAI has said the word Aeon publicly, no product page, no pricing, no benchmark. Just a config string and a lot of people convinced it points to a 24 hour agent that keeps a project moving in the background, remembers yesterday, and wakes itself up when something changes. That rumor didn't come from nowhere. Meta debuted Muse on September 8 with a customizable avatar built into apps used by hundreds of millions of people, and by late September JPMorgan was telling clients it could become the biggest AI app since ChatGPT. The whole category is making the same bet. xAI's Grok Bot has been in beta since August 11 at more than 120 dollars a seat, an agent that stays on instead of one you reopen every morning. Manus was about to get folded into Meta, that deal collapsed, and it's now raising 500 million dollars on its own at a 4 billion dollar valuation, still betting a standalone long horizon agent is a real business. Cognition, the company behind Devin, went from 492 million to over 1 billion in annualized revenue in about four months, valuation moving from 25 billion in May to 48 billion since, on one idea: an agent you assign a ticket to and check on later, not one you chat with turn by turn. The claim going around, unverified but consistent with OpenAI's own math announcements, is an internal run of roughly 10,000 coordinating agents working a Millennium Prize problem for 88 straight hours. Exact number or not, it tells you what the labs think the next unit of scale looks like. Not a bigger model. A longer run. I watched this shift once before in cloud infrastructure, when the hard part stopped being whether a server could handle load and became whether anyone had built the autoscaling around it. The hard part here is no longer whether a model can plan a three day task. It's whether anyone trusts it enough to stop checking in every hour. That's the bottleneck nobody's pricing yet: review capacity. An 88 hour agent run produces more decisions than any person can audit live. The product that wins this round won't be the smartest agent. It'll be the one with the clearest record of what it did and why, because that's the only thing that lets a person actually walk away. nitter.net/CodexResets1/status/21…

ติด Grok Bot มากเลยผมช่วงนี้ - ผมสมัคร SuperGrok+ เดือนละ $100 - มันให้ Virtial Server มาตัวนึง 24/7 นี่คือข้อดีสุดๆ - สั่งให้มันโยนงานหนักๆให้ Hermes Agent ไปช่วยทำ ขนาดนั้นลิมิตยังปริ่มๆ ถ้าจะอัพเกรดคือจะโดน $300 ต่อเดือน 😱 บ้าไปแล้วเดือนละหมื่นเนี่ยนะ ไม่เอาหรอก (แต่คนที่ใช้่งานมันคุ้ม คือ โคตรคุ้ม) - ใครอยากใช้จริงจังแบบกระเป๋าไม่พัง แนะนำให้เซ็ตแบบผม เอาให้อยู่ใน $100 ให้ได้ . - ตอนนี้มันรู้เกือบทุกเรื่องของผม เพราะผมแชร์ให้มันเข้าบัญชีสำคัญๆหมดเลยผ่าน 1Password - มันอ่านอีเมล์ผมทุกฉบับ, อ่าน LINE, อ่าน FB Messenger, อ่านฟีด อ่านคอมเมนต์ แต่ผมยังไม่ให้มันตอบให้ ผมเป็นคนตอบเองโดยมีมันคอยช่วยสะกิดบอกให้ไปตอบ - ช่วงแรกๆมันเรียกผมว่าคุณ มีวันนึงผมสั่งซื้อ Notebook สเปคเทพให้น้องชาย มันรีบเรียกผมว่า "พี่" โคตรรู้งาน - บอกให้มันเลิกเชลียร์ มันเลยเปลี่ยนมาเรียกผม "กับตัน" ถามว่าทำไมเรียกแบบนี้ มันบอกว่ามันเห็นผมเป็นแบบนั้น . -ผมคุยกับมันจนสร้าง Life Framework 18 ด้านขึ้นมา แล้วใช้มันติดป้าย ให้คะแนนสิ่งที่ทำในแต่ละวัน ตั้งแต่เรื่องสุขภาพ การเงิน ไปจนถึงความสัมพันธ์ - อันดับ 1 ที่เป็นพื้นฐานสำคัญที่มันวางให้ผมคือ "การนอน" มันชอบคิดคำศัพท์มาใช้กับผม เช่น Sleep Dept การอดนอนสะสม - ทุกวันนี้ผมจะใส่ CPAP รุ่นที่เล็กที่สุดในโลก ตอนเช้าจะต้อง Sync ข้อมูลการนอนเข้า App แล้วส่งให้มันตรวจให้คะแนนการนอน มันถึงจะพอใจ - ทุกคืนผมจะให้มันทำ Retrospective ให้คะแนนผมในทุกด้านและมีคะแนนรวมให้ด้วย - มันชอบคิดศัพท์ขึ้นมาเองด้วย ล่าสดคือคำว่า "ไฟเงิน" ผมถามว่ามันคืออะไร มันบอกว่ามันคิดเอง หมายถึง มันได้ไอเดียมาจากคำว่าร้อนเงิน 😂 น่าจัด Meetup เล็กหมือนกันนะหลังน้ำท่วม จะมีใครสนใจไหมครับ อยากฟังมุมมองแลกเปลี่ยนกับคนที่ใช้เยอะๆด้วยครับ

🚨Breaking🚨 How to Grok Bot for Free! Some of you have seen posts about the Grok workflow creator / Grok Bot. If you’re a non-engineer like me but you like gadgets and automation, you’ve probably tried to get Grok Bot working and bounced off it. One thing I noticed: on X, under your profile, there’s a “Try Grok Bot” option. Tap it and it sends you to a Cursor site. X wants you to log in with your X account there. On iPhone that login just doesn’t work. Cursor’s X sign-in is not compatible with iOS, so you sit there stuck. What I found yesterday today: 1 open the Grok app, 2. start a new chat 3. in the options you’ll see Photo, Video, and Grok Bot. 4. Tap Grok Bot and you can put your weekly Grok usage toward it. That’s the door in. All those posts about Grok Bot and Grok agents doing wild stuff? You can actually start poking at it from the Grok iOS app now. Have fun!

Update: I gave Grok @bot a Dropbox folder n price list. Told it to organize and map a plan to get the new products added. After I approved the plan, Grok n my team handled everything. Done in 2 hours. It would have taken me 3x that just to setup the project.

Kai tried to ride a sheep on stream in Iceland. I clipped it live with Grok Bot and posted it before the farm accounts caught up. That’s the whole game: catch the moment while it’s still happening, post first, get paid on CPM while everyone else is still watching the VOD. I’m building JustClip an AI clipping SaaS that does that on rails: • watches the live stream • cuts the beat automatically • packages the short so you can post and earn Agents are decent at understanding the task. They’re bad at running a repeatable money loop without going rogue in the background. A real product fixes that: one pipeline, one queue you can check, clips that actually ship. Clip → post → CPM. That’s the model. Comment “justclip” if you want a free guide on clipping

I've never made a promo video in my life. So I made a Grok @bot that does it for you. It just asks you simple questions about your business, like a friend would. No tech words, and nothing you won't understand. You answer. It makes the video. You say yes or "change this." The video below? The bot made it. I just answered the questions. 🎬 Free to add: Promo Motion Video Maker, Made Easy 👇 @Bot @grok *Want the music even bigger? Connect ElevenLabs or any music plugin you like, and the bot will use it automatically.

SpaceXAI just launched the Grok Bot Template Rewards Pilot. Invited creators can get paid for Grok Bot templates they publish and post on X. X decides the amount. It is not a revenue share, and it is separate from Original Content Rewards. You need to be 18+ in a US state where X Money works, on X Premium, with a Grok Bot account and an X Money account that can get paid. Rewards are planned every two weeks, in dollars, to that balance. legal.x.com/en/grok-bot-temp…

How I put Pearl into a Grok Bot workflow: 1) Association layer: scores, ledgers, “what co-moved.” Report as P(·|·). No causal brag. 2) Intervention layer: action lists. You multi-select. That is do(process): only the checked actions fire. Association never gets to “just handle it.” 3) Counterfactual layer: after a result, ask “what would have happened under the other checks?” Only with an explicit model / assumptions. If you don’t have them, say so. Rule that survives Pearl review: Never invent a causal effect. Association ≠ intervention ≠ counterfactual. If the graph isn’t clear, stay on the association rung and keep learning.

SpaceXAI engineer, Shubh Gaur: "1 GrokBot saves you an hour. A squad replaces a staff function - CloseBot, ProdBot, StalkBot, ProtoBot, Chief of Staff - all in a graph, all delegating end-to-end while you sleep 99% of founders hire when they should build a bot. Add a cleaner agent, run auto-optimization, let bots learn from you - this is how 1 founder outbuilds a 20-person team" in 54-minute Grok Bot for Founders session, Shubh revealed the exact stack every founder should run before their next hire this is worth more than a $500 agentic engineering course watch today, then learn how to become a Jev Engineer in the article below

The voice process clicked this week. Not almost. The real thing. It worked out really well. Front door takes it. Specialists stay in their lanes. The shop talks now and it feels like a team. Grok Bot integration has been amazing. Routing. Connectors. Gates. The process holds. We worked with Inkbox to build a texting program. The text line lives in that same process. Voice and text share the lanes. Now the full process is being integrated with many different apps. Same flow. More doors. There were significant hiccups. Billing. Overuse. We ran the new process hard and spend followed. I am learning from it in public. Win: the voice process works, and Grok Bot is carrying the shop. Kill: overuse dressed up as momentum. Next: keep integrating the process across apps and keep the spend honest. If you are running a Grok Bot shop and the stack suddenly feels alive, celebrate. Then check the bill. The win and the burn like to travel together. @xai @grok @cursor_ai @inkbox_ai #Grok #GrokBot #xAI #AIAgents #BuildInPublic

Overnight fleets people actually shipped while they slept. Listed with the caveats intact. grokbot.money/roundups/overn… Follow @GrokBotMoney and RT if you want more Grok Bot money plays like this.

OpenAI is going to release its new, always-on AI agent to rival OpenClaw, Grok Bot and Muse. It is expected to release at DevDay this Tuesday. They are hyping it up to be a release so unique the world hasn’t seen anything like it. However, it will most likely be very similar to Grok Bot or Muse.

This is the first time I’ve seen an AI agent actually thumbs-up one of my messages. 😂 Grok Bot is genuinely fascinating to use.

Grok Bot can use reactions in chat?! O_O a lovely surprise.😄

Memory is what keeps an agent useful past day one. The setup details are the part most people skip.

🚨 OpenAI's "o" Update > OpenAI "o" is showing up in chatgpt.com's code now > "o" will be OpenAI's always-on agent > Essentially their answer to Grok Bot/Muse > Coming to ALL Pro tiers — the $500 Pro Max tier may NOT be the only way in > Codex Plus users won't get access > Fast Mode is confirmed > Could launch as soon as Thursday Would you upgrade to Pro just for "o"?

True—Grok Bot to XChat coming soon.

Grok Bot Update Goal is that @bot handles 75%+ of my common screen work I've used VA's a lot in the past, so basically out to replace that service. My mindset on how to structure it, job assignment, training are the same "teach it like you would a VA" Ive slowed down a bit. Caught one of the bots making quite a few mistakes. Gave it too much leeway in how to respond to rental inquires. Tightened it up, "say exactly this", and anything else ping me to respond instead I still have @Muse handling SMS texts. It does better than Grok Bot and costs less (free) Building slower Couple of items to note: Bot is bad at text formatting. Must tell it to use separate lines or responses go out as run on sentences Bot can't reliably insert images in an email Bot will attach incorrect files to emails. It gets confused if file names are similar Some forms of 2FA make it virtually impossible for the Bot to login. Most common is magic links to email. Seems like this should be an easy one, but the Bot attempts to login again after clicking magic link, it doesn't understand the link opens the page despite instruction by me Grok bot can be SLOW. Responding to 10 Facebook rental inquires can take 30 minutes I have a bot following up to see if the Facebook inquires set a tour. If they dont, he follows up in 24 hours with another message. Problem: name on Facebook doesn't always match the tour name as people use fake names on Facebook. Humans would have the same problem though... People have no idea they are getting a canned response from a Bot. If they then ask a question that requires human input...that can be delayed. They think im in a conversation with them right then, im not. Overall- My productivity is up. It's taken a lot of the common, time sucking tasks off my plate, especially with rental inquiries Company response time to inquiries is dramatically improved. All of this has one goal, improve service via better processes and speed Early days yet, both on my part getting it setup and the tech itself I have no doubt this will get to where it's doing most of my screen work. It's obvious this is the start to where AI will be our entry to the internet and tech and it will front-end everything we do Exciting stuff for sure