EPISODE 2026-06-22

AI:AM LIVE — June 22, 2026 — The State of AI Engineering with swyx

The open covered Dean Ball joining OpenAI to lead a new 'Strategic Futures' team, tomorrow's NY-12 Democratic primary as a referendum on AI safety (RAISE Act author Alex Bores versus an a16z-backed super PAC), GLM 5.2 emerging as the new best open-weights model, and Sakana's Fugu orchestration launch. Then a long-form conversation with swyx — Shawn Wang, who coined 'the AI Engineer' — on the state of AI engineering and whether coding agents are automating or leveling up the role.

𝕏 Live broadcast

The June 22 show opened on a dense morning of frontier news — Dean Ball leaving the White House for a new 'Strategic Futures' team at OpenAI, the NY-12 Democratic primary shaping up as the first real electoral referendum on AI safety, GLM 5.2 emerging as the new best open-weights model, and Sakana's Fugu launch reframing orchestration as the next frontier — before a long-form conversation with swyx on the state of AI engineering.

Shawn Wang (swyx) — who coined 'the AI Engineer' as a job title — joined for an extended interview on whether coding agents are automating the role or leveling it up, the quality-versus-quantity problem in agent-written code, and the upcoming AI Engineer World's Fair.

The rundown

  1. --:--Opening30 min planned
    Opening: Dean Ball to OpenAI, AI Safety on the Ballot, GLM 5.2, Sakana's FuguFour threads from the morning's feed: Dean Ball leaving the White House to lead a new 'Strategic Futures' team at OpenAI, the NY-12 primary becoming the first electoral referendum on AI safety, GLM 5.2 emerging as the new best open-weights model, and Sakana shipping Fugu as an orchestration layer over a pool of other models.

    Dean Ball goes inside — a frontier lab hires the man who wrote the rules. Dean Ball, a primary architect of the administration's AI Action Plan and until last week a Senior Policy Adviser at the White House OSTP, is joining OpenAI to lead a new team called 'Strategic Futures' — chartered to look six to twelve months out at catastrophic risk, recursive self-improvement, labor disruption, and lab–government relations. The tension worth sitting in: the sharpest external policy thinker finally seeing the real capability curve, or regulatory capture told as a hiring announcement?

    Election eve: AI safety is on the ballot in NY-12. Tomorrow Democrats in NY-12 — the open Manhattan seat Jerry Nadler is vacating — vote in a primary that has quietly become the first real electoral referendum on AI safety. Alex Bores, the NY Assemblymember who authored the RAISE Act, is running, and an a16z-backed pro-AI super PAC named him its first target. Anthropic funded a counter-PAC. A House primary is now a literal lab-versus-VC proxy war — even as Bores polls behind the frontrunner.

    GLM 5.2: the open-weights coding gap collapses to weeks. Zhipu (Z.ai) shipped GLM 5.2 — open weights, MIT license — and the frontier-watcher consensus is that it's the new best open model. The headline isn't a benchmark; it's the harness. Nathan Lambert's verdict is that GLM 5.2 is the first open-weight model that 'feels right' inside coding harnesses as a general agent — landing in US-built harnesses (Claude Code, OpenCode) before Google's own model got there. So where exactly is the moat, and whose is it?

    Sakana's Fugu — is an orchestrator a frontier, or just a router? Sakana shipped Fugu and Fugu Ultra — a small 'conductor' model trained to dynamically route tasks across a swappable pool of other LLMs, including recursive copies of itself, behind one API. The pitch is that the most powerful AI systems won't be isolated monoliths but collaborative ecosystems. The tell is in the release: the strongest export-controlled models can't be in the pool. A router's ceiling is the pool — and the pool is shrinking.

  2. --:--Interview25 min planned
    The State of AI EngineeringShawn Wang (swyx)swyx on whether coding agents are eating or leveling up the 'AI Engineer' role he coined, the quality-versus-quantity problem in agent-written code, and the upcoming AI Engineer World's Fair (June 29 – July 2, San Francisco).
  3. --:--Closing
    Closing

Opening: Dean Ball to OpenAI, AI Safety on the Ballot, GLM 5.2, Sakana's Fugu

The open ran through four threads: Dean Ball's move from the White House OSTP to lead a new 'Strategic Futures' team at OpenAI; the NY-12 primary as the first electoral referendum on AI safety, pitting RAISE Act author Alex Bores against an a16z-backed super PAC; GLM 5.2 emerging as the new best open-weights model and the first that 'feels right' inside coding harnesses; and Sakana's Fugu, an orchestration layer that routes across a pool of other models.

  • Dean Ball joins OpenAI to lead 'Strategic Futures' — catastrophic risk, recursive self-improvement, labor disruption, and lab–government relations.
  • NY-12 primary: RAISE Act author Alex Bores versus an a16z-backed pro-AI super PAC, with Anthropic funding a counter-PAC.
  • GLM 5.2 (open weights, MIT license) lands as the new best open model and the first to 'feel right' in US-built coding harnesses.
  • Sakana's Fugu reframes orchestration as a frontier — but its pool can't include the strongest export-controlled models.

The State of AI Engineering — swyx

swyx — Shawn Wang, who coined 'the AI Engineer' in his 2023 essay 'The Rise of the AI Engineer' — joined for a long-form conversation on whether coding agents are eating the role or leveling it up, the quality-versus-quantity problem in agent-written code, and the upcoming AI Engineer World's Fair (June 29 – July 2, San Francisco).

Closing

The show closed with an extended wrap on the morning's threads and the conversation with swyx.