Grok Bot, in the wild

this is f**king ridiculous. one Grok Bot clock can turn a $33 job into a $397 one everyone keeps building bigger agent teams, but almost nobody checks how often those agents wake up to do absolutely nothing. one routine running every 5 minutes fires 8,640 times a month. at Grok 4.6 API rates, that's roughly $397 in estimated compute for my example workload. run the exact same job hourly and it's $33. the model didn't change. the prompt didn't change. only the clock did. [here's how i'd fix the entire loop] → event trigger: wake the bot only when something happens → change check: read one hash, count or timestamp → no change: write one log line and stop → real change: start the expensive full run → risky action: leave a draft instead of waiting for approval at 3 a.m. → audit: track every run, failure and wasted wakeup the cost breakdown gets even better: → 8,640 full runs = ~$397/month → 720 hourly runs = ~$33/month → 8,640 cheap checks + 180 full runs = ~$45/month → 89% lower estimated cost with the same 5-minute monitoring interval these are API-based estimates, not actual Grok Bot subscription charges. and there's another trap: background approvals expire after roughly 10 minutes. while you sleep for 8 hours, a 5-minute routine can produce 96 expired approvals. paste this into your Grok Bot: "Audit my existing routines and rebuild the execution plan around actual changes. 1. List every routine, its schedule and expected output. 2. Replace polling with narrow event triggers where supported. 3. For frequent schedules, check only a hash, count or timestamp before doing full work. 4. Stop immediately when nothing changed. 5. Leave irreversible actions as drafts for human review. 6. Log every run with time, result, changed status and approval state. 7. Flag routines with over 95% no-change runs. Show me the proposed changes and estimated savings first. Don't modify or delete anything without approval." the next big improvement to your a

Play videoX · 0:15

Open original on X

Holy sh*t. Grok Bot tip: stop fixing the same error every night, turn every exception into a rule fetch → read → extract → verify → rank → write → ship Teach Grok Bot a 7-step routine once, then let it run on its own computer while your laptop is closed In the demo, night 1 hit 3 exceptions it had never seen, and each one blocked it until a human answered: → a url returned 404: rule R01, try the archive copy → a source sat behind a paywall: rule R02, use the cached copy → a source had no publication date: rule R03, drop it and log it Those 3 blocks cost 42 human minutes. On night 2 the same 3 exceptions came back and needed 0 humans Night 1 breaks. Night 1 writes the rules. Night 2 runs alone. Along the way it pulls 12 urls, extracts 38 claims, runs 6 checks, ranks the top 5 and delivers at 03:00 to slack and email - the full routine setup > Teach it by doing the workflow once while it watches > Show the broken paths too: dead links, paywalls, missing dates, duplicates > Every exception becomes a numbered rule with an if and a then > Rerun the same routine on night 2 and count the human minutes > Approve only irreversible actions, everything else stays automatic Paste this into Grok Bot, the full build is in the article below: "Watch me complete this workflow once: [DESCRIBE THE WORKFLOW] 1. While you watch, record: > The steps, the tools and the order > Every decision I make and why > The points where something breaks or needs my approval 2. Turn what you saw into a routine: > A numbered checklist with the tool, input, output and failure condition for each step > Mark every step that needs my approval 3. Write one rule for every exception I handled: > Format: if [problem] then [action] > Show me each rule before you save it Do not run anything on your own yet"

Play videoX · 0:24

Open original on X

SpaceXAI engineer Lauren Tan just explained why scaling agents is not about adding more intelligence Most people try to make one model smarter The better approach is to build constraints, verification, and architecture that every future agent inherits In this 1-hour workshop, she breaks down the system behind that shift: 00:46 - why trust becomes the bottleneck once agents start acting on their own 05:39 - how verification skills close the loop instead of relying on model confidence 13:08 - how Pstack turns repeated engineering work into reusable infrastructure 17:23 - why skills and evals have to improve together 23:42 - how cloud agents scale the work without scaling human supervision 29:20 - why architecture and constraints matter more as autonomy increases 38:02 - how CI becomes an enforcement layer instead of another prompt 48:00 - what changes when the agent can keep working after you leave The same idea is exactly what makes Grok Bot interesting The model generates The decision layer routes Code enforces the hard rules The human only handles the exceptions And in quant, the architecture becomes: Hypothesis → Test → Reject → Stress → Size → Monitor This workshop is worth more than most paid agent engineering courses Bookmark and watch it today Then read the full Grok Bot operating system below ↓

Play videoX · 64:13

Open original on X

SpaceXAI PM (GrokBot team, ex-Cursor): “cut me out as the meat proxy” they put one team bot in their help channels. 70-80% of questions now get answered before a person ever sees them in a 45-min session, the GrokBot team builds one live and shows how it runs: 14:44 → build it: one bot, plugins, skills, shared read-only keys 21:48 → drop it into Slack, the whole team can DM it or tag it 24:04 → triage: 70-80% of questions answered with no human 26:30 → shared memory: an analyst corrects it once, everyone gets the right answer 28:38 → bots tag bots: the deal desk bot hands legal questions to the legal bot 41:00 → permissions: managers edit skills, everyone else just uses it full Grok Bot setup guide in the article below

Play videoX · 42:52

Open original on X

i'm f**king done with Grok Bots waking up 8,640 times a month just to find nothing changed... one routine running every 5 minutes costs roughly $397/month at Grok 4.6 API rates. run it hourly and that drops to $33. same bot, same job, 12x the estimated cost. so here's how i'd structure a 6-bot crew that actually makes sense overnight. [the setup looks like this] 1. chief routes events, not noise 2. scout checks what actually changed 3. quant compares the last saved state 4. risk returns PASS / CUT / KILL 5. exec handles safe, approved actions 6. scribe logs every run and queues drafts the trick isn't making 6 bots work harder. it's stopping them from waking up for nothing. add a cheap change check before every full run and my cost model drops from $397 to $45 (-89%). anything risky at 3 a.m. stays a draft until you're awake. that's the Grok Bot setup i'd actually trust overnight. full breakdown in the article below ↓

Play videoX · 0:25

Open original on X

THAT'S F*CKING CRAZY, GROK BOT JUST TURNED X INTO MY PERSONAL CONTENT RESEARCH ENGINE. I COMBINED IT WITH MY GROK BOT CONTENT CREATION SETUP. NOW IT HELPS ME FIND IDEAS WORTH MAKING. Grok Bot can search, read, and monitor X directly. You give it a task inside Grok Bot, and it gets to work. It already had an X API integration. The latest update goes further: searching, reading, and monitoring no longer require setting up the X connector. Here’s where this gets interesting for creators: → Find conversations gaining attention in your niche → Research new tools and product announcements → Pull out the questions people keep asking → Turn those findings into original content angles A request you can try: “Find posts in my niche gaining traction in the last 24 hours. Include source links, explain what people are responding to, and suggest five original video ideas for my audience.” That research feeds into my existing Grok Bot content setup. Now the setup has a way to discover what people are talking about before I decide what to make next. For my X workflow, Grok Bot is my pick for the best agent right now. Finding a good idea is often the part that stops you from making anything. Give the bot a clear niche, a time window, and an output you can actually use. One useful research brief can give you somewhere to start today. I broke down the latest major Grok Bot updates and how to use them in my article below. Bookmark this and read the full breakdown ↓

Play videoX · 0:30

Open original on X

Musk's post telling people to photograph their credit card for Grok Bot has 5.8M views. Grok replied "Don't do it." That picture gives the AI the number, expiry date and CVV. There is a safer way to pay, through Stripe Link. Link holds the card details. When the Bot finds the item, it sends a spend request with the amount. Only after you approve it in Link does it get a virtual card that works for one payment, so real digits never reach the chat. How to shop with it: - connect a Link wallet (US accounts for now) - name the item, a max price and "one-time purchase" - check the amount on every request - deny anything that renews monthly The last step comes from a real case. One user asked for a meal prep delivery and got signed up for a subscription, £57 extra. Musk's own test was LEGO Star Wars kits for young kids. After that purchase the single-use card is dead. A leaked photo of your real card keeps working.

Play videoX · 0:26

Open original on X

YOUR GROK BOT NEEDS 6 PARTS TO RUN WITHOUT YOU. MOST PEOPLE BUILD 1 AND CALL IT AUTOMATION. the 1 part they build is the prompt. the other 5 are why mine actually pays me. here's the full anatomy, in the order i'd build it today. about 20 minutes total. 01. the role (2 min) one bot, one job, one sentence. xAI's docs say it: no "General Helper" bots. weak: "help me with crypto" strong: "every hour you scan my watchlist and send me one explained shortlist, or the word nothing" it goes in the Bot description, not in chat. the description holds standing rules. chat is for today's task. 02. the skill (5 min) the job written down so it runs the same way every time. six parts: when to use inputs + access steps validate return (exact format) needs approval you can make one from chat, or use Teach a task and do the workflow in the browser while it records. up to 10 minutes, no audio. you get a draft back. the draft is never finished. it captures your clicks, not your judgment. write the validate step yourself. that's the part that saves you. 03. the ground test (5 min) dry run first: "don't execute anything. walk me through exactly what you'd do, and name the 3 riskiest steps." then test runs. a test run in Grok Bot performs real work, so start small. 3 runs, one of them ugly: empty input, broken source, missing data. 04. the routine (3 min) a skill + one bot + a clock. set your timezone in settings first, or 9am won't be your 9am. give it the owning Bot, schedule, input source, expected result and approval boundaries. then check the next run time it shows you. cheapest bug you'll ever catch. 05. the switches (3 min) Settings → General → Auto-review. Ask first: send, post, buy, transfer, delete, permissions, production, terms. Allow automatically: read, search, write inside its own /workspace folder. when both match, Ask first wins. never write "allow everything". and know this: all your Bots share one cloud computer. same files, same logins. separate b

View imageX

Open original on X