Srini Vijay — “Tesla Found the Cheapest Training Ground in Robotics”. Facts on this page cite the primary sources below, not the video.
Honest fence (read first): This Optimus Bot Monday flagship is about Tesla's software agent, not the walking robot. On the record: On Sat Oct 10 at 3:34 AM ET, Musk posted on X: "Digital Optimus can play about halfway through the Diablo campaign so far just by looking at the screen like a human. Skill at Counter-Strike and other fast games is good too. League is also in training. Goal is to generalize across all games." A live Tesla job listing (Palo Alto, Req. 285962) describes a "Digital Optimus Gaming Track" and says the work will be integrated into "Full Self-Driving (FSD) and physical Optimus." What's missing: Tesla hasn't released gameplay video, a benchmark, the test setup or a write-up. Musk didn't say which Diablo game, which difficulty, or how much help the agent got.
Editorial Thesis
We last covered Digital Optimus in August, when Tesla described it on its Q2 call as software that reads a screen and works a keyboard and mouse. What's new is a concrete-sounding progress claim and a job ad that explains why Tesla is using games. Games are a cheap, safe and endless place to practise "see the screen, take an action" in real time. The open question is whether those skills carry over to driving or to a robot's hands. Tesla says that's the goal. It hasn't shown it.
Honest Verdict
Believe that Musk made the claim and that Tesla is hiring for this exact work. Both are primary sources. Treat "about halfway through the Diablo campaign" as an unmeasured progress report until Tesla shows it. Treat "generalize across all games" and the jump to FSD and the physical robot as stated goals. A game agent that plays well is real AI progress, but it isn't proof that a humanoid robot can do useful work.
What Musk claimed (Oct 10)
- Digital Optimus can play "about halfway through the Diablo campaign so far just by looking at the screen like a human."
- "Skill at Counter-Strike and other fast games is good too."
- "League is also in training."
- "Goal is to generalize across all games."
- He was replying to a post about the job listing. His post had about 3 million views by Monday morning.
What Tesla's job listing says
- Title: Machine Learning Engineer, Realtime Games, Digital Optimus, Palo Alto, full-time, listed pay $124,000–$420,000 a year plus cash and stock awards.
- The work: "real-time agents that operate computers at superhuman levels. Your models will learn to browse the web, use professional apps, and play AAA games."
- It also asks for "rigorous benchmarks… to evaluate agent performance in games and measure transfer across different games, embodiments, and general computer use."
- And to "integrate agent systems into Full Self-Driving (FSD) and physical Optimus."
- A job ad tells you what a team is trying to do. It doesn't tell you what the team has done.
Why games, in plain English
A game gives an AI a screen to watch and controls to press, with instant feedback and no one gets hurt. That's close to what Tesla says Digital Optimus is for: reading a screen and using a keyboard and mouse. It's further from a robot hand driving a screw, where the AI has to handle weight, grip and touch. Srini Vijay's video, embedded here, asks the right question: does one "brain" really move between different bodies, or does each body need its own work?
Claim vs proof
| Claim | Proof so far | Status |
|---|---|---|
| Plays ~halfway through Diablo from pixels | Musk post; no footage or metric | Unverified claim |
| Counter-Strike skill is "good" | Musk post; no score or match data | Unverified claim |
| League "in training" | Musk post | Work in progress |
| Skills will transfer to FSD and the robot | Tesla job listing (goal) | Stated goal |
| Tesla is building this team | Live Tesla job listing, Req. 285962 | Confirmed |
What to watch
- Wed Oct 21, 5:30 PM ET: Tesla's Q3 call. Any Digital Optimus demo, benchmark or product date would be new proof.
- Gameplay video with the setup shown: which game, how many tries, and whether a human helps.
- Any sign of the same model running a real computer task, or a physical Optimus task, that it wasn't trained on.
Video note
Srini Vijay N10QwTwLINU, "Tesla Found the Cheapest Training Ground in Robotics" (uploaded Oct 14, 2026, about 6 minutes). He says he spent 15+ years moving one autonomy system across a desert robot, an ATV and a lunar rover. The video gives three questions to ask before believing in a "generalist robot brain." It's commentary from a small channel, and Tesla isn't involved.
Bottom Line
Believe that Musk made the claim and that Tesla is hiring for this exact work. Both are primary sources. Treat "about halfway through the Diablo campaign" as an unmeasured progress report until Tesla shows it. Treat "generalize across all games" and the jump to FSD and the physical robot as stated goals. A game agent that plays well is real AI progress, but it isn't proof that a humanoid robot can do useful work.Sources
- Elon Musk on X — Digital Optimus 'about halfway through the Diablo campaign… just by looking at the screen' (Oct 10, 2026, 3:34 AM ET)
- Tesla Careers — Machine Learning Engineer, Realtime Games, Digital Optimus (Palo Alto, Req. 285962)
- AI Shift News — Tesla's Digital Optimus plan is a computer-use bet (Aug 10, 2026; Q2 call context)
- Srini Vijay — Tesla Found the Cheapest Training Ground in Robotics (Oct 14, 2026; embed)