ex (@0xexcurrent) header
ex (@0xexcurrent) profile picture

ai maxxing open dm

Joined November 2018
Photo by ex (@0xexcurrent)Photo by ex (@0xexcurrent)Photo by ex (@0xexcurrent)Photo by ex (@0xexcurrent)Photo by ex (@0xexcurrent)Photo by ex (@0xexcurrent)Photo by ex (@0xexcurrent)Photo by ex (@0xexcurrent)Photo by ex (@0xexcurrent)Photo by ex (@0xexcurrent)
No f*cking way. Grok Bot engineer Lauren Tan: A doctor who can describe exactly where a hospital workflow breaks has a head start building software for it. They already know which mistake would make the product useless. That’s the point Tan makes to Matt Pocock about domain expertise. Her own example is painfully familiar: debugging Cursor meant sitting in Chrome DevTools, reading performance traces and carrying the findings back to her agent by hand. Run the app → Reproduce the lag → Capture a trace → Check the fix She knew how to investigate a slow screen. The breakthrough was giving the agent a way to run that same investigation. Think about the last task you delegated: which part did you quietly take back because only you knew how to check it? Save the interview for that question. The guide below applies it to a weekly AI brief: give your reviewer the exact source passage and enough context to challenge the writer’s claim ↓
10
367
Holy sh*t. Sam Altman asking Elon Musk what to play next: Elon says Overwatch. Sam already plays Overwatch. And suddenly they sound like two friends trying to figure out what to boot up that night. It’s a 2016 Y Combinator interview, and it hits differently during GTA VI week. Musk brings up Hearthstone, Altman brings up The Last of Us, and the conversation ends up on why the original Deus Ex stuck with Musk. Overwatch → Hearthstone → The Last of Us → What makes a game’s story stick That little detail about what Sam already played changes the conversation. Once they get into what each of them liked, the recommendations get a lot more personal. Bookmark the clip for the most unexpected 90 seconds of that interview. Then read the Grok Bot guide below and turn the details you keep repeating into a brief your bot team can work from ↓
12
551
Holy f*ck. OpenAI founding researcher Andrej Karpathy: He asked Deep Research for a table of U.S. AI labs. It came back with funding figures, headcounts, and one very obvious hole. It left out xAI. In the walkthrough, Karpathy shows how AI can search, read, and assemble a report while you do something else. Then he opens the results and spots what slipped through. Question → Search → Draft → Check the sources → Check what’s missing That last step changes how you review AI work. Every row can look right and the table can still give you the wrong picture. Bookmark this and jump to 42:04 in the full walkthrough for the Deep Research demo. Then use the Grok Bot guide below to give your research bot a reviewer that checks the draft against the original brief and opens the sources itself ↓
18
546
This is f*cking wild. NVIDIA co-founder Jensen Huang: He floated a 1,000× increase in the energy needed for computing. Then he said his estimate could be off by orders of magnitude. The reason he gave at Stanford is more interesting than the number. Software used to sit idle until you opened it. He sees more of it generating new work continuously in the background. More background work → More model calls → More compute → More power You can see a small version of that problem in an agent workflow. A bot can spend tokens searching, drafting, and retrying all morning and still leave you with zero claims you can publish. Bookmark the two-minute clip for Huang’s explanation of what happens when AI keeps working between prompts. Then read the Grok Bot guide below and measure what your setup actually costs per accepted result ↓
15
530
This is f*cking wild. Former Meta chief AI scientist Yann LeCun: “The AI industry is completely LLM-pilled.” At Davos, he said that was a big reason he left Meta. His point about agents is more practical. To plan, a system needs to predict what its actions will do. Here’s where that gets real with Grok Bot. Your bot tries to save a draft. The app times out. Did the save fail, or did it go through and the confirmation never come back? Save draft → Check the destination → Retry if it’s missing → Send for review That one check keeps your reviewer from having to figure out which of two drafts is current. Bookmark the clip before you automate a job that writes to another app. Then use the guide below to build your first Grok Bot team around one job, with clear rules for when to retry and when to stop ↓
20
579
This is f*cking wild. Grok Bot engineer Lauren Tan: A bug report landed in Slack with a cropped screenshot and three question marks. Her agent could run the app. It still had to guess which feature the user meant. In this 39-minute talk, Lauren explains what she added: a feature map that tells agents where each feature lives, how to reach it, and what it’s supposed to do. Identify the feature → Open it → Reproduce the issue → Capture evidence Steal that idea for your weekly AI brief. A docs page updated on Friday can describe a feature that shipped months ago. Have your reviewer open the original announcement and check the release date before that item makes the cut. Bookmark the talk for the next time your bot says it checked its work. Then use the guide below to set up one recurring job in Grok Bot and give your reviewer a clear rule for what passes ↓
1
18
423
Holy f*ck. NVIDIA CEO Jensen Huang: “Most software is going to be done agentically anyhow.” In this 98-second clip, Jensen explains why systems thinking matters. Chip design already went through this shift. Once tools took over tasks like gate synthesis, designers could focus more on how the whole chip works. His questions apply to AI agents too. What goes in? What comes out? What constraints does the system have to work within? Define the job → Set the inputs → Run the bot → Check the output → Approve Try it with a weekly AI brief. If you change the window from 30 days to seven, give the reviewer the updated brief. Otherwise, a polished draft can pass review against last week’s rules. Save this clip before you automate a recurring job. Then use the guide below to build your first Grok Bot team around one deliverable and give every bot the same definition of done ↓
1
16
549
This is f*cking wild. Anthropic CEO Dario Amodei: “Google and OpenAI are fighting it out in consumer.” At Davos, he used cats riding donkeys to explain why consumer video was low on Anthropic’s list. If business customers ever needed it, he said Anthropic could license a model. Apply that same filter to your own AI setup. What would you still ask your bot to do next Monday, once the novelty wears off? Pick a recurring job → Run it with one bot → Check the result → Measure cleanup time → Add a role if it helps Try it with your weekly research brief. Run the same assignment with one bot and with a team. Count how many minutes you spend getting each draft ready to send. Worth 87 seconds before you spend a weekend building an agent team. Bookmark the clip, then use the guide below to run that test with Grok Bot and see if the extra roles earn their keep ↓
8
540
This is f*cking wild. OpenAI CEO Sam Altman: A $1B company with 10 people. Altman thinks AI could eventually bring that number down to one. In 2024, he told Alexis Ohanian that his group chat of tech CEOs was already betting on the year the first solo unicorn would show up. There’s a smaller version of that bet you can run this week. Can one bot take a recurring job off your plate and hand back work you’d actually use? Pick a job → Run it with one bot → Check the output → Add a specialist if needed Try it with your weekly research brief. If checking the sources eats up the time you saved, you’ve found the first role worth adding. Save the clip before you build your first agent team. Then use the guide below to run that test with Grok Bot and see whether a second bot earns its spot ↓
16
569
This is f*cking wild. xAI founder Elon Musk: In November 2025, he put Grok 5’s odds of reaching AGI at about 10%. In this eight-minute clip with Ron Baron, he describes what he was betting on: a model with six trillion parameters, built to handle text, images, audio and video and create tools when it needs them. Just before that prediction, he describes Grok 4 Heavy as a study group. Several agents work in parallel, then compare their answers before returning a conclusion. For your own Grok Bot team, give that comparison step a clear job. Assign the brief → Gather sources → Draft → Check each claim → Approve Give the reviewer the exact draft and its source log. If a source doesn’t support a claim, send that claim back to the researcher with the reason. Bookmark the clip. Then use the guide below to build that review step into one weekly AI brief ↓
10
317
Holy f*ck. SpaceXAI product team: A PM bot briefs a design bot in a group chat. The design bot sends back mockups, and an engineering bot joins to take the prototype forward. In this workshop, Kevin Niparko and Roshan Sadanani show how they use Grok Bot to move a voice feature from an idea to a spec and mockups, then hand it off to engineering. 27:39 — Ask PM Pete to draft the voice feature spec in Notion. 29:54 — Put Pete and Pixel in the same chat so Pete can pass along the design context. 34:25 — Bring Emily in to build from the spec and mockups. Research → Draft the spec → Hand off context → Create mockups → Start the build The move I’d copy is giving the next bot the actual document, its current version and a clear job to do with it. Bookmark the workshop for your next product experiment. Start at 27:39 to see the handoff take shape. Then use the guide below to try it on one weekly AI brief, with a source log and a reviewer who checks the draft against the evidence ↓
11
229
Holy f*ck. Anthropic CEO Dario Amodei: AI can make people more productive and still eliminate the work that gets someone their first job. In this 42-second clip, he singles out entry-level work in law, finance and consulting—and warns that substantial job losses could follow. His concern is the speed and breadth of the change. Several career paths could be disrupted while people are still deciding where to start. One practical response: Define the job → Gather evidence → Draft → Review → Deliver Before buying another course promising an “AI-proof career,” explore these six starting points: Claude plugins for legal, finance, sales and HR github.com/anthropics/knowle… Skills for working with Word, Excel, PowerPoint and PDF github.com/anthropics/skills A research agent that produces reports with citations github.com/assafelovic/gpt-r… A framework for coordinating agents with distinct roles github.com/crewAIInc/crewAI Anthropic’s Python SDK for building agents github.com/anthropics/claude… 18 lessons from Microsoft on building AI agents github.com/microsoft/ai-agen… Pick one task. Learn to give it clear requirements, inspect the evidence and judge the finished result. Bookmark the video and this list. Then read the guide below for a Grok Bot workflow built around one weekly brief, with clear owners, traceable sources and a reviewer who checks the work ↓
1
17
744
No f*cking way. SpaceXAI engineer Amrita Venkatraman: Your bots can keep working after you close your laptop. Give them a recurring job, clear responsibilities and a result to return. In this 22-minute clip, she walks through Grok Bot as a team of persistent coworkers, with reusable workflows and ways to coordinate their work. Assign a Role → Teach the Workflow → Coordinate → Check → Repeat A skill saves how to do the job. A routine decides when to run it. Get the process working before putting it on a timer. For your first team, make every handoff clear: who owns the next step, which files they need and what counts as finished. Worth watching before paying for another agent engineering course. Bookmark it. Start with one recurring job and one bot delivering a result you can inspect. Then read the guide below: give Chief, Researcher and Reviewer one weekly brief to finish—with explicit handoffs and sources you can actually check ↓
1
17
729
No f*cking way. NVIDIA CEO Jensen Huang: Asked whether he’d rather be 20 today or relive his own twenties, he chose the past. “There’s some joy in ignorance.” In this 2:39 clip, he argues that constantly seeing everything wrong with the world can drain your optimism before you’ve had a chance to build anything. My practical takeaway for AI: One Job → First Build → Check → Improve → Repeat Make your first project small enough to finish: a weekly brief with linked sources and a draft you can review. The first run gives you specific problems to solve instead of an endless list of things to study. Watch this before buying another course just to feel ready to start. Bookmark the video. Choose one recurring task and get the first version working. Then read the guide below: build your first Grok Bot team around that task—with clear responsibilities, explicit handoffs and a finished weekly brief you can verify ↓
2
12
708
Holy f*ck. Ex-Google Distinguished Engineer Kelsey Hightower: Microsoft CEO Satya Nadella emailed him personally. Then he opened an offer with a number he hadn’t known was possible. “Not very often in your career does a zero get added to the equation.” In this 2-minute clip from The Pragmatic Engineer, he tells the story—and explains how Microsoft came back even higher after his counteroffer. Interview → CEO Email → Offer → Counteroffer → Higher Offer My takeaway for learning AI: turn what you study into something you can demonstrate. One working project gives you concrete decisions, mistakes and results to explain. Worth watching before your next salary conversation. Bookmark the video. Then pick one small workflow you can build, run and verify from beginning to end. The guide below gives you a starting point: your first Grok Bot team, built around one weekly brief—with clear responsibilities, explicit handoffs and evidence behind every finding ↓
1
8
354
Holy f*ck. Ex-Google product lead Jacob Bank: 40 marketing agents. A reported AI bill of $500/month. His estimate for comparable contractors: $50,000/month. In this 16-minute interview, the Relay.app founder explains how he built that system one small workflow at a time. One Job → One Agent → Check Results → Improve → Expand One agent turns new YouTube videos into LinkedIn posts. Another monitors competitor pricing. Each has a specific job—and he changes or retires the ones that stop delivering value. His advice is the part worth stealing: “Do not start with a 40-agent org chart. Start with one.” Worth watching before you pay for another AI agent course. Bookmark it. Pick one recurring task you keep doing by hand. Then read the guide below: build your first Grok Bot team around one weekly brief—with clear owners, explicit handoffs and evidence behind every finding ↓
9
506
Elon Musk, founder of xAI: An AI team gets interesting when you stop moving work between every chat yourself. One clear brief goes in. One checked result comes back. In this 47-minute launch recording, Musk and the team discuss Grok’s reasoning, tool use and parallel problem-solving. Your first workflow: Brief → Chief → Researcher → Reviewer → Checked Result Start with a weekly AI brief: five verified findings, one short draft and a source log. Run it once, fix the gaps, then save the working method as a skill. Worth watching before you spend $1,000 on another AI course. Bookmark the video. Choose one recurring job to build your team around. Then read “Build Your First Grok Bot Team Around One Finished Job”—with copyable role prompts, handoff cards and a practical plan for your first week ↓
12
471
Microsoft CEO Satya Nadella: “Use all, but be independent of all.” Your AI workflow needs a quality bar that survives a change of model. In this 3:46 All-In clip, he gives a practical test for AI dependence: remove a model and check whether the system can still meet your standards. Define the Outcome → Set the Test → Change the Model → Compare Results Apply that discipline to your Grok Bot team: same brief, same sources, same acceptance criteria. Track accepted work, total cost and the time you spend fixing mistakes. Worth watching before you spend $500 on another AI strategy course. Bookmark the video. Write down what “good” means for one real task. Then read “Before You Add Another Grok Bot, Run These 7 Tests”—and make the team prove its value before you expand it ↓
Your Grok Bot can finish the task twice—and call both runs a success. I put together 7 tests for duplicate work, interrupted runs, stale evidence, and weak reviews. Includes a job card and a one-bot vs. team scorecard. Full breakdown below ↓
6
136
Elon Musk, xAI founder, on Grok’s reasoning capabilities: An AI that tackles expert-level questions still needs a workflow you can trust with an ordinary task. A saved file. A dropped connection. A retry. That’s where your next test starts. In this 47-minute launch recording, Musk and the team discuss Grok’s capabilities and benchmarks such as Humanity’s Last Exam. Question → Reason → Use Tools → Compare → Answer Try this with your Grok Bot: interrupt a test job after it saves a draft. When it resumes, check whether it finds the existing file before creating another one. Worth watching before you spend $500 on another AI course. Bookmark the video. Pick one workflow and test its recovery. Then read “Before You Add Another Grok Bot, Run These 7 Tests”—to catch duplicate work, conflicting edits and weak reviews before expanding your team ↓
Your Grok Bot can finish the task twice—and call both runs a success. I put together 7 tests for duplicate work, interrupted runs, stale evidence, and weak reviews. Includes a job card and a one-bot vs. team scorecard. Full breakdown below ↓
1
11
469
Cursor’s Amrita Venkatraman: Your AI team can keep working after you close your laptop. The real test is what you can trust when you open it again. In this 56-minute workshop, she builds a Chief of Staff setup with inbox, calendar and task specialists, then explains reusable skills, routines and approval rules. Brief → Chief of Staff → Specialists → Handoffs → Results Try this afterward: send the same job twice. Inspect the files. You want one accepted result and a clear record of the duplicate. Worth watching before you spend $1,000 on another agent engineering course. Bookmark the video. Start with one workflow you can reliably verify. Then read “Before You Add Another Grok Bot, Run These 7 Tests”—and use the scorecard to decide whether another bot earns its overhead ↓
Your Grok Bot can finish the task twice—and call both runs a success. I put together 7 tests for duplicate work, interrupted runs, stale evidence, and weak reviews. Includes a job card and a one-bot vs. team scorecard. Full breakdown below ↓
1
11
344