DeepSeek's three code-agent job postings decoded: hiring graders, games the only surviving track
teortaxesTex · x · 2026-11-20
teortaxesTex relays Fable's analysis of DeepSeek's three versions (July/Sept/Nov) of code-agent job postings, with a detailed breakdown in the attached images.
Key findings across versions
- Scope shifted from "code" to "engineering": July's mission was all about code as the route to AGI; by November "code" vanished from title and mission, replaced by "complex, real engineering settings," plus non-software chip and mechanical roles.
- The common thread is checkable work: chips come with simulation/DRC/LVS, mechanical with CAE and physical validation, security with working exploits — fields where machines can score results, exactly what an RL reward needs.
- Hiring graders, not builders: September's posting says it "does not require writing product code" and wants 3+ years of experience distinguishing good work from plausible-looking work.
- Disappearing tracks likely mean "filled," not abandoned: security sub-specialties rotate each round (web/kernel/browser in July; crypto/intranet in Sept; OS/mobile/firmware in Nov), suggesting wave-based hiring.
Why games
Games is the only coding track that survived all three versions. Ranked reasons: (1) games are where coding models are weakest — engine editors and binary assets barely appear in public text; (2) correctness can't be unit-tested, needing experts to define checks; (3) game developers can build simulated environments for agents; (4) "make me a game" is the staple request and standard demo; (5) China has a deep talent pool — GameLook saw listings paying 40–70K RMB/month at 14 months' salary. September also saw heavy game-designer hiring, which the author links to rumors that OpenAI found great reasoning uplift from game-playing: "I don't think it was for demos on X."
More from AGI Musings
- The hardest work skill by 2028 won't be prompting — it's writing goals you can accept — evisoft · 2026-11-20
- Does AI make building small profitable businesses easy but big durable ones harder? — vaibhavbetter · 2026-11-20
- US and China Account for 74% of AI Models, Raising Tech Dependency Concerns — mikeflache · 2026-11-20
- Google Documents First Fully Autonomous Attack System, Exploit Agent Turns Patches Into Attack Maps — mikeflache · 2026-11-20
- Nokia Bell Labs talk: AI and the Restructuring of Jobs, Nov 23 — danielequercia · 2026-11-20
- Anthropic's 5,300 founders and employees now worth more than bottom third of US households — jathansadowski · 2026-11-20