Epic tells developers: no more hand-written code on Disney game
A slow-news Monday turned sharp: Epic Games confirmed a report that its delayed Disney game is now an "AI first" project, robot startups have raised nearly $48 billion in a single year, and an amateur mathematician used GPT-6 Astra to close a 57-year-old proof gap.
Epic Games has told staff there will be "no more hand-written code" on its long-delayed Disney game. That's the headline line from a Bloomberg report by Jason Schreier, which Epic confirmed on the record: the Marvel-and-Star Wars-themed shooter — part of the $1.5 billion Epic-Disney collaboration announced in early 2024 and originally targeted for March 2027 — is delayed indefinitely, and its art will be "generated with full AI workflows" using models such as Anthropic's Claude. Employees were reportedly stunned when the mandate landed last week, many hoping it was a joke, which is unsurprising at a company that laid off 1,000 people in March and then discovered it no longer has the headcount to ship the project. Tim Sweeney didn't deny any of it, posting hours later that "we have a huge pivot underway here at Epic" and framing it as a boon for Unreal Engine users rather than a headcount cut. This is the clearest signal yet that "AI first" has moved from startup pitch to survival strategy inside a AAA studio — and the honest read is that Epic is using AI to cover a staffing hole it dug itself, not because the technology is ready to replace game developers. The Disney relationship makes it higher-stakes: if AI-generated content stumbles on Marvel IP, both companies wear it.
Robotics and "physical AI" companies have raised nearly $48 billion so far in 2026 — about 1.7 times all of 2025. The Financial Times, citing PitchBook data through October 8, puts the year-to-date total at roughly $48 billion, against a 2022–2024 baseline of $10–15 billion a year. The money is heavily top-weighted: Anduril's $5 billion round, NEURA Robotics' $1.4 billion Series C, and Physical Intelligence's $1 billion together made up 39.8% of Q2's record $18.6 billion quarter — though deal count actually fell 14% that quarter, so fewer, bigger bets. Two caveats before you quote the number: it's PitchBook's broad taxonomy (defense-tech included), and Crunchbase's narrower robotics count is roughly a third of it. What's unambiguous is where the spend goes — training infrastructure. The FT's "robot gyms" (teleoperated arms collecting motion data) and Figure's paid video-harvesting app are becoming an industry unto themselves; we covered one symptom last week, Mecka AI nears $500 million valuation as Sequoia bets on robot training data. The open question is whether data volume actually converts to capability the way it did for language models.
A non-professional mathematician, working with GPT-6 Astra, says he has proven the remaining open cases of a circle-packing problem that has stood since 1969. The blogger known as nor claims machine-checked proofs that the classical arrangements for packing 15, 16, 17, 18 and 20 equal circles into the smallest enclosing circle are optimal — closing every case up to 20, where only N≤14 and N=19 had been rigorously settled. The approach leans on the model to reduce infinitely many arrangements to a finite, computer-exhausted set, with nor steering, hand-verifying, and cross-checking results against Claude Opus 5.5. It fits the pattern we noted when GPT-6 Astra solves the last FrontierMath Tier 4 problem: the model is strongest where proof reduces to constrained search. The asterisk is real — this is a self-published blog result with no peer review and no independent expert validation yet, so treat it as a strong claim, not a settled theorem.
What to watch: whether any working mathematician publicly checks nor's proofs — and whether Epic's AI-first mandate leaks into other publisher projects.
Is "no more hand-written code" a rational response to a staffing hole or a tech-washing of layoffs? Tell us in the comments.
