The Signal
@emostaque is now pinning a specific date to code automation: 2027 for "instantly generate almost all code," 2028 for "folk running out of ideas." This is no longer speculative; it's a delivery commitment backed by $1B training runs and looped transformer architecture shipping at scale. But the gap between "AI writes better code than most devs" and "no developers left" is not a technical problem—it's an economic one that @bindureddy's framing of "zombie employees" actually exposes rather than resolves.
IMPORTANT
The developer extinction thesis has moved from "possible" to "timetabled," but labor replacement isn't automatic; it requires someone to actually fire people, and that only happens when margin pressure exceeds switching costs.
What's Moving
- Developer role bifurcation accelerating — @emostaque claims only "a few" devs will write better code than Astra by 2027; @svpino frames the shift as "anyone can code, but not anyone can build." The gap is architectural thinking and taste. Coding as a skill is dead; architecture as a judgment call remains scarce. (via @emostaque, @svpino)
- Astra's spatial + knowledge-work positioning reshaping task allocation — @bindureddy flags Astra excels at "computer use and data analytics" while Fable 5.1 owns large codebases. This isn't parity; it's functional specialization that forces orgs to route by task type, not by "best model." Zombie employees emerge when humans stop being the routing layer. (via @bindureddy)
- Prompt-to-deployment collapsing into minutes — @sama's observation that Astra generates playable games within minutes is the real signal: latency between ideation and working software has collapsed. This eliminates the review/iteration tax that historically protected junior developers. (via @sama)
- "Better code than me" threshold crossed silently — @emostaque admits shock at AI surpassing him in math; @excel_rator notes Astra/Fable upgraded from "bad code to good code when prompted right." The inflection isn't announced; it's whispered in DMs between engineers who've already switched workflows. (via @emostaque)
Crosscurrents
- Economic displacement lag vs. capability timeline — @emostaque hedges that full replacement won't happen until "fully AI companies come for your margin." Meaning: models can replace devs today, but incumbents won't fire until they're forced. 2028 capability timeline doesn't mean 2028 employment crash.
- "Zombie employee" framing obscures the real transition — @bindureddy's read treats automation as instantaneous, but the actual pattern is hybrid workflows where humans validate, redirect, and integrate AI output. This is deskilling, not elimination.
Tradecraft
BULL
Astra's $1B spend justified only if OpenAI believes orchestration/specialization (not reasoning) is the moat. If true, this is the first frontier model trained explicitly for product fit, not benchmark dominance.
WATCH
Fable 5.2 launch timeline and whether Anthropic reprices down again or holds. Pricing stability = signal that capability parity is durable, not temporary.
Desk Notes
- @emostaque — Developer extinction now has dates (2027-2028); positioning this as labor market shock while quietly building local-owned AI utility framework to capture the transition politically.
- @bindureddy — Task-based routing (Astra for knowledge work, Fable for hard coding) is the immediate operational reality; watching whether orgs adopt this segmentation or wait for one model to own everything.
- @svpino — Shift from "writing code" to "reviewing code" is the actual deskilling pressure; code review is already collapsing under scale of AI-generated output.