{"slug": "i-used-grok-bot-like-spacexai-told-us-to-it-put-us-on-a-72000-mo-run-rate", "title": "I Used Grok Bot Like SpaceXAI Told Us To. It Put Us on a $72,000/Mo Run Rate", "summary": "A two-person startup founder reported that using Grok Bot, an AI workforce tool from SpaceXAI, generated approximately $1,000 in on-demand usage within 24 hours, translating to a $72,000-per-month run rate. The founder, who was still building the organizational chart, halted usage due to the high compute costs, highlighting that the AI workforce is currently affordable only to an elite few.", "body_md": "The problem is not credits or surprise billing. The problem is that the AI workforce of the future is currently affordable only to an elite few.\n\nSpaceXAI told us what Grok Bot @bot was supposed to become: a team of always-on AI employees that could use real software, learn how we work, coordinate with one another, and finish real jobs.\n\nIts launch announcement encouraged users to create multiple Bots and give each one a role. Inside SpaceXAI, people were reportedly running Bots for sales, operations, finance, marketing, support, and engineering. Some employees had a chief-of-staff Bot coordinating the others.\n\nI took that vision seriously. And was very eager!\n\nWe are a two-person founder team @atlas_v_erified. The idea that two humans could build and operate a company with a capable team of AI employees is not a gimmick to us. It is central to how we believe companies will work in the future.\n\nSo I followed the playbook.\n\nI created roughly 20 Bots. I built a chief of staff, with an incredible org chart which was our v1 test. I began assigning responsibilities, creating personalities, organizing workflows, and figuring out how these AI employees should work together.\n\nAnd I was barely getting started.\n\nWhy we moved to Grok Bot\n\nWe were already experimenting with this way of working through Perplexity Computer.\n\nWe migrated because Perplexity Computer was WAY too expensive (even more expensive than this) and overall a bit too clunky. We wanted the capabilities of an autonomous computer-using agent, but we needed economics that could actually work for a small founding team.\n\nGrok Bot appeared to be the answer.\n\nIt was backed by SpaceXAI, available through my Cursor Ultra plan, and explicitly designed around multiple persistent AI teammates. Give SpaceXAI's compute infrastructure, I hoped this might finally be the platform that made an AI workforce practical rather than merely impressive.\n\nInstead, we ran directly into the same fundamental barrier: compute cost.\n\nApproximately $1,000 in 24 hours\n\nWithin approximately 24 hours, I had generated around $1,000 in on-demand usage. At points during this usage, the invoices were generating roughly every hour or so.\n\nSustained around this run rate, $100 per hour is $2,400 per day, and $72,000 over a 30-day month. Calling it a $72,000-per-month run rate is not an exaggeration. It is a conservative description of the rate I was observing while actively building and using the system, and still ramping up!\n\nTo be clear, I had not already received a $72,000 bill. A run rate describes what the observed spending rate becomes if sustained.\n\nBy August 26, the dashboard showed $2,217.30 in on-demand usage for the billing cycle.\n\nThe underlying charges included:\n\nThe billing system began issuing repeated mid-month invoices of approximately $100 each. Seven of those invoices landed on August 23 alone.\n\nThis was happening while I was still building the organizational chart.\n\nI was still creating the assistants, deciding what they should do, teaching workflows, and determining whether the product was capable of meaningful work.\n\nI was just getting started, and so excited too! Best AI tool I have used for a while.\n\nBut, if the setup phase was already capable of producing a roughly $72,000 monthly run rate, what would the cost look like after the Bots were fully deployed and operating 24/7 as advertised?\n\nI stopped before finding out. Super bummed.\n\nI want to understand whether the output can justify the cost\n\nIt would be easy to conclude that Grok Bot needs more budgets, spending limits, and automatic shutdowns. Those controls might be useful, but they do not solve the real problem, and they may undermine the entire purpose of the product.\n\nYou would never hire an employee and shut them down halfway through the week because they reached a payroll limit. You compare their output to their cost. If they create more value than they cost, you continue investing in them. If they do not, you make a change.\n\nGrok Bot should ultimately work the same way.\n\nThe point of an intelligent AGI AI workforce is to offload work, 24/7. The Bots should continue operating, learn how the business works, monitor what needs attention, and move things forward without requiring constant supervision.\n\nIf I have to assign tiny budgets, monitor every Bot throughout the day, and repeatedly restart them when they shut down, I am managing the AI workforce more closely than I would manage human employees. We could limit the Bots to conventional working hours (9-5), but that compromises one of the products most powerful promises: an AI team that continues working, coordinating, and creating value around the clock.\n\nThe output may still justify the cost\n\nWhen I saw Grok Bot working, I became aspirational about what it could mean. I was not thinking about saving a few hours, I was thinking that two founders might be able to 100x their output. If 20 AI employees could reliably produce the work of a high-performing organization, even $72k per month might ultimately be worth paying.\n\nBut, I was still setting everything up when the spending accelerated. I do not know whether this is the product's expected cost, something inefficient in our configuration, unusual behavior in particular conversations, or simply an early-beta problem. I had to pause before we could find out whether the output would eventually justify the expense.\n\nI am not asking for credits or a refund. I am sharing this because I hope someone from SpaceXAI sees this and takes it seriously as we think this is going to be a recurring issue on what is otherwise an incredible product!\n\nI can provide our conversation IDs, bot configurations, roles, and workflows so the team can help me understand exactly what happened (fingers crossed they can help us). @lifeofjer on Cursor.\n\nI am still incredibly excited about the product and the vision! I want to keep building this AI organization and offloading work. Still bullish and hopeful. But, def putting a pause on it for now.\n\n*I think this is important in today's age: I hand wrote this entire copy, but used GTP 5.6 Sol as my writing assistant and editor, consider it a mix of human and AI generated content. Image above was generated using AI but includes a real screenshot of my cursor billing.", "url": "https://wpnews.pro/news/i-used-grok-bot-like-spacexai-told-us-to-it-put-us-on-a-72000-mo-run-rate", "canonical_source": "https://twitter.com/lifeofjer/status/2092659107383898237", "published_at": "2026-08-26 17:04:14+00:00", "updated_at": "2026-08-26 17:17:13.442338+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-infrastructure"], "entities": ["SpaceXAI", "Grok Bot", "Perplexity Computer", "Cursor Ultra", "atlas_v_erified"], "alternates": {"html": "https://wpnews.pro/news/i-used-grok-bot-like-spacexai-told-us-to-it-put-us-on-a-72000-mo-run-rate", "markdown": "https://wpnews.pro/news/i-used-grok-bot-like-spacexai-told-us-to-it-put-us-on-a-72000-mo-run-rate.md", "text": "https://wpnews.pro/news/i-used-grok-bot-like-spacexai-told-us-to-it-put-us-on-a-72000-mo-run-rate.txt", "jsonld": "https://wpnews.pro/news/i-used-grok-bot-like-spacexai-told-us-to-it-put-us-on-a-72000-mo-run-rate.jsonld"}}