I build robots. I write about AI, startups, and the future of autonomy. Boston, MA.

Boston, MA
Joined April 2021
Job boards are full of reposts from 3 months ago. This Fo prompt only finds roles posted in the last 7 days, straight from company career pages, with direct apply links. Swap in your role + city. No padding: mine found 8 real ones. wajo.ai/f/06ec1b5b0bf954a1e7…
43
Applied to 100+ jobs and lost track? Same. My Fo prompt reads your inbox and gives you: - every application + status - OA/interview deadlines this week - open calendar slots - follow-up drafts for the ones that ghosted you Try it 👇 wajo.ai/f/3dddcdbabb3aa21d9f…
Introducing Fompt Feed to answer the burning question, what do I do with these agents? Steal the billionaire’s Fo prompts, give your self-help books an execute button or copy your impressive(?) friends. - for the curious, @vkhosla’s life hacks. - for entrepreneurs and small businesses, @tferriss's delegation playbook, @JamesClear's habits. - for job seekers @AliAbdaal's strategies. - for parents, Emily Oyster and @GoodInside's hacks - your friends's funny fompts, - your mom's "i can't believe you did this" fompt. Meet the Fo feed. You’ve spent enough time watching other people have their shit together. Your turn to become a frompter ↓ (the top creator earns $10k this week)
2
75
I took 2nd at #Hackachusetts building robots in NVIDIA Isaac Sim on a laptop: RTX 4050 with 6 GB VRAM and 16 GB RAM. NVIDIA's minimum spec for Isaac Sim is 8 GB+ VRAM and 32 GB RAM. I'm below the floor. @nvidia @NVIDIARobotics @elonmusk @MrBeast, help me get a real rig? 🧵
3
1
4
299
Day to day: Isaac Sim, ROS 2 and RViz fighting over 6 GB of VRAM, 40 GB of swap so the sim isn't killed, overnight renders, and today I deleted whole projects to free disk. I build bridge-inspection robots and 36 robotics courses on Udemy. A real GPU = bigger sims, faster 🤖
63
We won 2nd place at #Hackachusetts 2026! 🥈🏆 250+ projects, 9 campuses across Massachusetts, and our team took home $1,000 after pitching at WHOOP HQ in Boston. Couldn't ask for better teammates than Megha and Thiago 🙌
1
1
125
Mayday: a drone and an underwater ROV that inspect bridges together. We rebuilt Boston's Charles River Dam Bridge in NVIDIA Isaac Sim, ran both on ArduPilot + ROS 2, and fused lidar + sonar in one 3D map. Megha built the AI: YOLOv8 defects + a report agent. #Robotics #ROS2
2
90
When one of our 20 robots breaks, Claude gets a real person on the way. Fo phones the on-call tech, moves down the list when she says no, books the repair and emails the report. Code rules decide who Fo may contact. Robots: Isaac Sim. Calls: real. @hey__fo @AnthropicAI
AI can think. AI can talk. AI can get things done. Today, we're giving Fo a way to feel and control the physical world. - a head shake and a reminder in a lamp to get you going - nudge your vacuum cleaner for a deep clean - make your cli call you for updates Introducing Fo MCP. AI has escaped the screen.
11
3
1
15
1,691
Thanks! After the last no, it escalates: Fo emails the ops lead "robot 7 needs a human decision" and the panel shows an approve button. Nothing gets booked and nobody off the list gets called. A live run hit exactly that when someone said "67 days".
21
Thank you! 🙌 The phone calls in the video are the real recordings, which was the most fun part to build.
32
Trust by design: an approved contact list, max 2 calls per incident and 1 per person, no spending, a kill switch. Enforced twice: before each request is built, and by a hook on every Fo tool call, so the model can't route around it.
43
The ETA goes back into the sim as the repair timer. The allocator routes the other 19 robots around the broken one. When the tech arrives, the robot goes back to work and Fo sends the closing email.
36
Fo makes the real call. We read the call transcript and code checks the answer: an ETA only counts if it's 1-120 min. Can't come or no time? Fo calls the next person on the list. Then it checks the calendar, books the slot and emails the report.
43
Ferbin retweeted
We just made an AI FAMILY MEMBER! Meet Fammy, our flagship virtual agent that teaches the kids, checks your outfit, shops, makes calls, and reminds Dadaji to take his BP tablet. Brain: @claudeai Hands: Fo MCP by @wajo_ai What should Fammy learn next? 👇
AI can think. AI can talk. AI can get things done. Today, we're giving Fo a way to feel and control the physical world. - a head shake and a reminder in a lamp to get you going - nudge your vacuum cleaner for a deep clean - make your cli call you for updates Introducing Fo MCP. AI has escaped the screen.
17
9
3
49
4,585
How we built it 🧵 Hyperion: 20 NVIDIA Nova Carter robots in Isaac Sim, driven live by our fleet code. Four agents on Claude watch them: surveillance, war room, scheduler and allocator.
59
Thanks! In the video: Megha couldn't come, so Fo called Thiago: "half hour." Fo saw a real 9:22 repair on the calendar, booked him at 9:52 PM and emailed the report. And when a tech said "67 days": xitter.cf/Ferbin08/status/211223…
Our technician told Fo he'd be on site in "sixty-seven days." A model would happily book that. Our code didn't: a valid ETA is 1-120 minutes, so it went to a human instead. Real call, real rule. Trust by design is code, not a prompt.
83
Our technician told Fo he'd be on site in "sixty-seven days." A model would happily book that. Our code didn't: a valid ETA is 1-120 minutes, so it went to a human instead. Real call, real rule. Trust by design is code, not a prompt.
When one of our 20 robots breaks, Claude gets a real person on the way. Fo phones the on-call tech, moves down the list when she says no, books the repair and emails the report. Code rules decide who Fo may contact. Robots: Isaac Sim. Calls: real. @hey__fo @AnthropicAI
1
2
409
Everything passed because nothing ran. Fourteen checks, none wired to the orchestration layer. Two weeks of trusting green. Trip a gate deliberately before you trust it. if you can't make it fail, it's not running.
4
311
The sensor was publishing at 3.3Hz. I had it configured for 10. Spent two weeks re-tuning the controller. Rate bugs look exactly like tuning problems. You chase parameters. The rate is invisible. The fix was one line.
1
253
Translation matched within 43mm. Rotation came out at 0.926 rad. Gazebo gave 1.857 on the same turn. <dynamics damping> is in both URDF specs. PhysX reads it as a velocity gain. At 0.01, the joint slips. Diff the numbers after every transfer. Not the video.
193
Same reward score. Stride frequency 2x apart, completely different gaits. The optimizer didn't hallucinate. You forgot to penalize regularity. It filled that gap however it wanted, and you never got to choose.
1
165