AI Duck (@AIDuckLabs) header
AI Duck (@AIDuckLabs) profile picture

New AI releases, real tests, honest verdicts. I use this stuff daily and tell you what holds up.

Joined September 2026
Photo by AI Duck (@AIDuckLabs)Photo by AI Duck (@AIDuckLabs)Photo by AI Duck (@AIDuckLabs)Photo by AI Duck (@AIDuckLabs)Photo by AI Duck (@AIDuckLabs)Photo by AI Duck (@AIDuckLabs)Photo by AI Duck (@AIDuckLabs)Photo by AI Duck (@AIDuckLabs)Photo by AI Duck (@AIDuckLabs)Photo by AI Duck (@AIDuckLabs)
HOLY F*CK. 99% of people use ai like a chat. the 1% hired it. one bot finds trends. one writes. one kills fake numbers. a manager runs them. you just sign. same $20 subscription. most people pay it. the 1% turn it into $2,000.
10
2
1
24
29,210
HOLY F*CK. a user reports a bug. by 3 AM Lauren Tan's agents have fixed it and shipped it. nobody on the team even noticed SpaceXAI engineer: "most people still test every line their agent writes by hand. that's exactly why they can't ship 4,000 updates a month" in a 37-minute talk she shows how she taught her agents to run their own code like a real user and check their own work Agent Writes It - Runs It Like A User - Catches Its Own Bugs - Fixes Them - Merges It Itself better than any $5,000 coding bootcamp skip netflix, watch it tonight, then tell me - who tests your agent's code right now?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
9
20
308
48,285
HOLY F*CK. the GrokBot team just said it out loud: "GrokBot is not a chatbot" then they dropped a free 1-hour workshop showing what it really is, from 0% to 100%: • 10% - 0:57 - how the GrokBot team handles usage limits • 30% - 6:41 - building skills for GrokBot • 55% - 10:41 - building a Chief of Staff agent • 80% - 12:48 - GrokBot running Cursor agents for you • 100% - 27:01 - replacing your software with bots most people still ask it questions like it's a chatbot the team treats it like a teammate that manages other agents save it, watch it today, then tell me - chatbot or teammate?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
14
31
6
314
31,204
HOLY F*CK. Lauren Tan's agents merge their own pull requests. and she just showed how, for free the SpaceXAI engineer dropped a 1-hour masterclass on building a GrokBot team, from 0% to 100%: "I've gotten to a point where my agents basically merge their own pull requests" 10% - 3:36 - giving bots real jobs: their own logins and a budget 30% - 22:35 - the workflow behind 4,000 PRs a month 55% - 29:01 - building your own software factory 80% - 40:13 - the smallest loop: teaching agents to check their own work 100% - 47:33 - codebase guardrails that save tokens most people are still babysitting one agent Lauren runs a factory that ships 4,000 PRs a month and fixes bugs at 3 AM save it, watch it tonight, then tell me - what % are you at?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
16
46
3
408
35,701
HOLY F*CK. she shipped 3,000 pull requests in a month. a bunch of them while she was asleep Lauren Tan, engineer on GrokBot at SpaceXAI (ex-Cursor): "if your inner coding loop produces slop, automating that loop just produces more slop" it's a 17 min talk and honestly I had to pause it a few times her agents merge most of their own code now. nobody reads it line by line took her 6 months to get there. she says you'll do it way faster One Chat - Lint Rules - Verification - Auto-Merge - Software Factory the trick is kinda boring. every time an agent screws up she doesn't touch the prompt. she writes a rule in the code so it can't happen again • useEffect? banned in product code • code comments? 99% banned, agents were using them to excuse lazy fixes • code review is now just videos and screenshots the agents send back • most of her tokens don't even go to writing code. they go to checking it and on every release the agents poke at the whole app like a real user, find bugs, fix them, merge the fix. no human in the middle her team did 11,000 pull requests last month. new release pretty much every day. 0.5% got rolled back meanwhile most of us are still rewording the same prompt for the 10th time this is worth more than a $5,000 agents course, and it's free save it, watch it tonight, then tell me - would you let agents merge their own code?
Your Grok Bot is not expensive Your setup is 75% of my Grok Bot tokens went to re-reading old chat. About 1% went to actual work I broke down the 6-bot setup the Grok Bot team runs, and the 9 leaks eating your bill ↓
11
10
79
6,518
HOLY F*CK. I asked Fable 5.2 to design its own body it didn't build a robot. it built a box the whole thing is code. no CAD, no 3D artist, nothing now sell me this box no subscription. no rate limits. no "you've hit your limit, come back in 5 hours" just Fable 5.2 sitting on your desk, working while you sleep the AI already did the hardware. Anthropic just has to ship it would you pay $3,999 for this?
🤖 Made with AI
nobody’s ready for this... fable 5.2 can watch its own videos and fix them i turned it into a 24/7 motion studio steal the full setup, every prompt included↓
9
1
13
616
HOLY F*CK. Lauren Tan ships 2,000 PRs a month and doesn't check a single one herself SpaceXAI engineer: "you can't run a hundred agents if you can't trust one if nothing verifies your agents, you're the verifier. and you're the bottleneck" she used to watch every output by hand. now 2,000 PRs a month run on autopilot Watching Every Step - Verify - Skills - Evals - Auto-merge trust isn't a feeling. it's infrastructure you build once most people skip this part and wonder why AI keeps breaking stuff would you let your agents auto-merge?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
1
15
1
186
21,548
HOLY F*CK. the Grok Bot team banned 99% of code comments. then shipped 11,000 PRs in one month 0.5% rollbacks. a stable release almost every day Lauren Tan explains how they turned AI slop into gold: "everyone's looked at someone else's code and thought it was slop. I felt it too" her answer: stop trusting the agent. make the codebase impossible to break • APIs, architecture, types and tests force agents to do the right thing by default • a dependency checker blocks bad imports before they land • comments got banned because agents used them to excuse quick hacks • every correction becomes an executable check. same mistake can't happen twice • code review runs on videos, screenshots and proof. not reading lines • TLA+, Lean and UI tests are already catching real bugs in 19 minutes Grok Bot finds bugs, opens fix PRs and merges them by itself proof before anything ships. every mistake becomes a rule. that's level 2 and level 6 all 7 levels are in the article below, each with a prompt you can paste tonight which level is your team stuck on?
HOLY F*CK. 99% of people use ai like a chat. the 1% hired it. one bot finds trends. one writes. one kills fake numbers. a manager runs them. you just sign. same $20 subscription. most people pay it. the 1% turn it into $2,000.
6
1
9
642
HOLY F*CK. Lauren Tan hasn't read her agents' code in months. they merged 3,500 PRs anyway SpaceXAI engineer: "most engineers still check every line their agent writes I run 60+ agents and haven't opened their code in months. every review comment I ever wrote is a CI rule now" 53 minutes. she shows how her agents shipped 3,500 PRs last month while she slept Prompting - Reviewing - Verifying - Delegating - Sleeping reviewing code used to be the job. turning every review into a rule is the job now that's level 6. every mistake becomes a rule, so it never happens twice I mapped all 7 levels in the article below, from one chat to a team that runs overnight. a prompt for each one would you merge code you never read?
HOLY F*CK. 99% of people use ai like a chat. the 1% hired it. one bot finds trends. one writes. one kills fake numbers. a manager runs them. you just sign. same $20 subscription. most people pay it. the 1% turn it into $2,000.
10
8
1
76
10,846
HOLY F*CK. Lauren Tan doesn't build bots anymore. her bot does ex-Cursor, now at SpaceXAI: "I built one bot to design all my other bots I taught Dr. Eggbot what a good bot looks like" 34 minutes. one bot staffs a whole pop-up store from zero One Bot - Dr. Eggbot - Prototyper - Designer - Merch Bot - Store Dr. Eggbot runs on pstack. it creates the bots, gives each one a job and gets them working she won't call it a software factory. it's a team of employees everyone else is still typing prompts into one chatbot that's level 4. most people are stuck at level 0 I broke down all 7 levels in the article below. a prompt for every one, ready to paste tonight be honest - what level are you on?
HOLY F*CK. 99% of people use ai like a chat. the 1% hired it. one bot finds trends. one writes. one kills fake numbers. a manager runs them. you just sign. same $20 subscription. most people pay it. the 1% turn it into $2,000.
10
1
1
62
7,617
HOLY F*CK. Lauren Tan built a bot whose only job is managing her other bots ex-Cursor, now engineer at SpaceXAI: "one of my bots exists just to design the rest. it reads every chat I've had with them and tells me which one keeps messing up some days I don't even open the PR until it's already merged" in 43 minutes she shows the Grok Bot setup that books her flights, ships her code and fixes its own team while she sleeps One Grok Bot - A Bot That Builds Bots - Agent Swarms - Code That Merges Itself her bot-builder runs on a schedule. it checks every bot, finds the one making the same mistake over and over and rewrites its skills before she even notices one message planned her whole 4-city trip to London and Amsterdam. flights, hotels, everything. she just said yes she doesn't call it a software factory. she calls it a Michelin kitchen. quality at scale most people are still prompting one chatbot. she's running a bot that hires, trains and fixes all the others that's level 6. most people are stuck at level 0 I mapped all 7 levels in the article below, with a prompt for each one you can paste tonight which level are you on?
HOLY F*CK. 99% of people use ai like a chat. the 1% hired it. one bot finds trends. one writes. one kills fake numbers. a manager runs them. you just sign. same $20 subscription. most people pay it. the 1% turn it into $2,000.
13
9
81
10,738
HOLY F*CK. this is how a SpaceXAI engineer gets agents to do 80% of her work without breaking stuff Lauren Tan: "I take every check I'd do by hand and turn it into a skill the agent runs itself" her setup is called pstack. every request goes to a playbook, the playbook picks the skills step 1 - give the agent a verifier it can actually run 1. start the app, make sure it's up 2. use the feature, check what happened 3. save proof, kill processes, clear temp files try it on one real feature first step 2 - the proof has to fit the change UI - the full flow, start to finish CLI - command, output, exit code storage - write it, read it back performance - same build, same cache, same load one lucky benchmark doesn't count. /benchmark-checklist catches that same bug twice? /correct fixes the root. architecture, then types, then CI, then a test and the fix has to catch the original bug or it's not a fix anything boring goes into a script for the next agent save this before you launch your next agent
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
14
19
1
173
18,619
HOLY F*CK. Lauren Tan doesn't trust her coding agents at all. she makes them prove everything she's an engineer at SpaceXAI and hands off like 80% of her work this way "I turn my manual checks into a reusable verification skill" so I put her whole method on one page • the agent gets a check it can actually run. open the app, click through the feature, look at what happened, clean up • if the same bug shows up twice she doesn't patch it again. she fixes the root, and the guard has to catch the original bug • proof depends on what changed. UI change? show the full flow. CLI? show the output. storage? read the value back • one good benchmark run means nothing. same build, same cache, same load, or it doesn't count • anything boring she does twice turns into a script for the next agent honestly most people just trust whatever the agent says it did save this before you spin up your next agent. how do you check what yours ships?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
14
5
45
3,216
HOLY F*CK. her laptop is closed and 12 AI bots are still working She’s worth learning from Lauren Tan, SpaceXAI: "stop waiting for a smarter model. your bot can already hire one 12 bots, one computer, 12 screens. they don't stop when I close the lid" 8-minute Q&A. one bot doing everything - a team that hires its own agents One Chat - A Bot Per Job - Separate Memory - Bots Hiring Agents - Work That Runs Without You task too heavy? her bot spins up a cloud agent on whatever model fits and hands back the result and she says the next Grok is already being built. this setup only gets stronger most people are still waiting for a new model to paste the same old prompt into skip the $5,000 AI agent course save it, watch it tonight, then tell me - one bot or a team?
Your Grok Bot is not expensive Your setup is 75% of my Grok Bot tokens went to re-reading old chat. About 1% went to actual work I broke down the 6-bot setup the Grok Bot team runs, and the 9 leaks eating your bill ↓
8
6
92
8,726
HOLY F*CK. 99% of people are using GrokBot wrong, and SpaceXAI's own senior engineer just said it Lauren Tan: "GrokBot is the most powerful agent tool we've ever released, and almost everyone uses it the wrong way at SpaceXAI I run a team of 100+ GrokBot agents. a Chief of Staff, 10 managers and 90+ workers this is what engineering looks like in 2026" in 40 minutes she shows how to squeeze 100% out of every agent tool you already have GrokBot - Chief of Staff - Managers - 90+ Workers - Full Team no $5,000 agentic course online comes close skip netflix tonight and watch this instead. then tell me - are you in the 99%?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
12
10
63
4,956
HOLY F*CK. Lauren Tan from SpaceXAI ships 2,000 PRs a month and doesn't watch a single one "you can't launch a hundred agents if you don't trust what one of them puts out no verification skill means you're the one verifying. and that makes you the bottleneck" in a 1-hour talk she shows how she went from babysitting every output to 2,000 PRs a month on autopilot Watching every step - Verify - Skills - Evals - Auto-merge trusting your agents isn't a feeling. it's infrastructure you build beats any $3,000 AI coding bootcamp save it, watch it tonight, then tell me - would you let your agents auto-merge?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
18
27
2
311
42,230
HOLY F*CK. she goes to sleep and 50 AI agents build her company overnight Lauren Tan, senior engineer at SpaceXAI (ex-Cursor): "almost everyone uses GrokBot the same way. open a chat, ask one thing, close it I treat mine like a company. a Chief of Staff runs my schedule, 3 PMs divide the tasks, 40+ workers build through the night. I wake up and review it with my coffee. that's 20 engineers of output" in a 1-hour workshop she sets up the whole team from nothing single GrokBot - Chief of Staff - PMs - 40+ workers - a company on autopilot 97% of engineers at SpaceXAI already work like this. the strongest ones get $750K+ offers one chat tab vs a whole team. guess who wins worth more than a $5,000 course save it, watch it today, then tell me - one agent or a team?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
27
61
3
513
50,639
HOLY SH*T. an Anthropic engineer just said prompting is done "prompting is pretty much over 95% of our engineers already run loops that improve themselves. graphs are what's next" in just 10 minutes she builds her whole Claude Code setup live, starting from a blank terminal Blank Terminal - Agents - Loops - Graphs - System That Improves Itself prompting was the old way to work. building the system around the model is the new one beats most $1,000 agent engineering courses save it, watch it today, then tell me - are you still writing prompts by hand?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
8
1
14
791
HOLY SH*T. she makes $1.2M a year at SpaceXAI and hasn't written code by hand in months "I build 150+ GrokBot teams that ship 1200+ updates a day while I sleep. I set the rules and check what the bots delivered" in a 1-hour livestream she starts with an empty GitHub and zero bots 72 hours later the bots are running a company how it works: step 1 - one bot, one job, its own cloud computer step 2 - show it a task once. it saves it as a skill forever step 3 - drop all the bots in one group chat so they pass work to each other step 4 - set the rules, let them run, check the results the 4 a.m. run still happens with your laptop closed GrokBot - skills - bot team - group chat - a company that runs itself SpaceXAI's rule: "a persona, not a task" anyone can open a GrokBot chat. almost nobody builds a team that works without them beats any $2,000 agentic course save it, watch it today, then tell me - what job would you give a bot first?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
16
17
1
152
26,133
HOLY SH*T. Lauren Tan doesn't write code anymore. she runs a company made of AI agents ex-Cursor, now senior engineer at SpaceXAI: "99% of people open one GrokBot, ask a question and close the tab I run 50+. a Chief of Staff plans my day, 3 PMs split the work, 40+ workers build it while I sleep. in the morning I check the results over coffee. that's the output of 20 engineers" in a 1-hour workshop she builds that exact team from zero one GrokBot - Chief of Staff - PMs - 40+ workers - a company that runs itself 97% of SpaceXAI engineers already work this way. the best ones are getting $750K+ offers if you're still in one chat tab, that's who you're competing with better than any $5,000 course save it, watch it today, then tell me - how many agents are you running?
Your agent is not autonomous You are its loop You check every step because nothing else in the system can I broke down the 7 levels from babysitting agents to running a fleet ↓
22
56
6
404
51,015