Meta's AI Backfires, GLM 5.3 Flash, OpenAI's Jalapeño Chip & the End of Programming

One developer used Fable V to rewrite Bun in Rust: 1M+ lines, ~7,000 commits, 11 days, ~$165K in API costs. It now underpins Claude Code.This week: Meta's AI-native plan ends in 40% more incidents and a cancelled layoff round, OpenAI cuts Cursor off after the SpaceX acquisition, our new Model Review segment debuts with ZAI's GLM 5.3 Flash, OpenAI's Jalapeño chip posts absurd tokens-per-kilowatt numbers, and the AI bubble clock moves backward for the first time in a month.Hosts: Shimin Zhang, Dan Lasky, and Rahul Yadav — back from the Galápagos, where everyone talks in geological time.▸ News: Meta's AI-Native Plan Backfires — Project OT ("organizational transformation") was meant to cut headcount 25% by replacing workers with AI agents. Code changes rose 220% year over year; features reaching users rose only 36%; major technical and security incidents rose 40%, with firefighting time up as much as 70% — and the second round of layoffs quietly died. The supercar analogy applies: the tech wasn't the bottleneck, the organization was.▸ News: OpenAI Cuts Cursor Off — After SpaceX's ~$60B Cursor acquisition, OpenAI will pull its models by November 12, citing terms-of-service history — and, without quite saying it, distillation. Consolidation begets counter-moves; we ask what Cursor's moat is that OpenRouter doesn't already have.▸ Model Review (new segment): GLM 5.3 Flash — OpenRouter's free mystery model "Ox Alpha" turns out to be ZAI's GLM 5.3 Flash: above Sonnet 5 at max thinking and Grok 4 on the Artificial Analysis index, at $0.075 per million input tokens, served entirely on Chinese-made chips it helped optimize. It fails the two-acre farm test (agricultural work does not pay $50/hour), passes the nth-order-effects test with agent-liability insurance, and reflexively cites FOIA — draw your own distillation conclusions. Xiaomi-in-2013 vibes.▸ Hardware Hut: OpenAI's Jalapeño Chip — R&D to tape-out in under nine months. 22,000 tokens per second per kilowatt on a 538B open-weights model versus 427 on NVIDIA's GB200 — a comparison OpenAI keeps in the appendix. The trick: the KV cache never leaves the die, and idle chip sections power-gate off. Gen 2 is already in tape-out, and unreleased GPT Astra has been tuning the chip for 60+ days.▸ Post Processing: The End of Programming — Paul Dix's essay, anchored by the Bun rewrite. Best-case scenario or the shape of things to come? Dan's hex-payload smart-home port says verification is everything, front-end testing still isn't solved, and even Paul expects organizational inertia to buy "another decade of humans writing code by hand." We land on: refinement is the durable skill.▸ Two Minutes to Midnight — NVIDIA reports a $96.2B quarter and forecasts 70% growth while margins slide on memory costs. The skeet-sourced stat of the week: NVIDIA has half of Amazon's revenue and twice its market cap. Anthropic hits $65B annualized revenue — roughly $13M per employee — IPO expected this fall. The clock moves backward to 4:15.⏱ Chapters00:00 Cold Open — Rahul Returns from the Galápagos02:42 News: Meta's AI-Native Plan Backfires08:39 News: OpenAI Cuts Cursor Off13:07 Model Review: GLM 5.3 Flash (Ox Alpha)20:47 Hardware Hut: OpenAI's Jalapeño Chip28:56 Post Processing: The End of Programming42:48 Two Minutes to Midnight: NVIDIA & Anthropic's $65B49:57 Outro🔗 Articles we discussedNews:• Meta's scrapped AI-native plan — Ars Technica: https://arstechnica.com/ai/2026/08/metas-scrapped-plans-to-go-ai-native-included-slashing-teams-by-60-percent/• OpenAI's decision on Cursor — OpenAI: https://openai.com/index/our-decision-on-cursor-following-its-acquisition-by-spacex/Model Review:• GLM 5.3 Flash announcement — Z.ai: https://z.ai/blog/glm-5.3-flash• The Ox Alpha reveal — TechCrunch: https://techcrunch.com/2026/08/26/surprise-z-ai-is-the-ai-lab-behind-the-mysterious-ox-alpha-model/Hardware Hut:• Jalapeño first results — OpenAI: https://openai.com/index/jalapeno-first-results/Post Processing:• The End of Programming — Paul Dix: https://pauldix.com/the-end-of-programmingTwo Minutes to Midnight:• NVIDIA's quarter — archived coverage: https://archive.ph/qX0JT#selection-1572.0-1572.1• The revenue-vs-market-cap skeet — Bluesky: https://bsky.app/profile/sungkim.bsky.social/post/3muakuku3gc2j• Anthropic's annualized revenue surges to $65B — TechCrunch: https://techcrunch.com/2026/08/17/anthropics-annualized-revenue-surges-to-65b/🎙 About ADI PodADI Pod (Artificial Developer Intelligence) is a weekly podcast about AI and software development for working developers. We go through hundreds of links and dozens of newsletters each week so you don't have to. New episodes Fridays.• https://www.adipod.ai• [email protected] this gave you something to try on Monday, hit subscribe and drop a comment.