Artificial Developer Intelligence

Shimin Zhang, Dan Lasky, & Rahul Yadav
  • 5.0 (8)
  • Technology
  • Updated Weekly

Three engineer friends argue about AI so you don't have to. Shimin Zhang, Dan Lasky, and Rahul Yadav are working developers who've been watching AI transform their profession in real time, and they got opinions on the robot takeover. Every week the three get together to riff on the latest AI news, geek out over research papers, roast each other's tool choices, and occasionally have an existential crisis about whether the craft is dying or just getting weird. What you're signing up for: - AI news without the LinkedIn cringe: model drops, acquisitions, open-source drama, and the other stuff that actually matters if you write code for a living. - Technique corner: real tips from the trenches: spec-driven development, multi-agent orchestration, Claude.md tricks, and all the ways they've wasted hours so you don't have to. - Two Minutes to Midnight: the show's running AI bubble tracker, complete with circular funding diagrams, hyperscaler CAPEX math, and a doomsday clock they keep arguing about moving. - Deep dives that (occasionally) go deep: hallucination neurons, agentic memory, workflow automation economics, LLM architectures the papers nobody else is covering because they're hard. - Dan's Rant: Dan frequently gets mad about things. It's a whole thing. - The feelings segment: Yes, Shimin reads Tennyson on a tech podcast. Yes, Rahul wrote an AI-generated country song. No, they're not sorry. Three friends with strong opinions, questionable metaphors, and genuine love for the craft they're also mourning for. If you want to understand AI deeply, use it without embarrassing yourself, and laugh at the absurdity of it all, pull up a chair.

  1. 2d ago

    GPT-5.6 Sol, the State of AI, Know Your Unknowns With Agents & the Permanent Underclass

    Ford quietly rehired the "grey beard" engineers it had automated away — the AI running its QA kept failing. The same week, the share of CEOs who expect AI to cut headcount dropped from 46% to 20%. This week: GPT-5.6 Sol, China walls off its own models, Meta's "AI gulag" ships mini video games, 11 agent techniques from the Fable 5 release, and AI revenue adding $1B every two days. Clock holds at 4:45. Co-hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. ▸ GPT-5.6 "Sol" — OpenAI's answer to Mythos and Fable, in three flavors: Sol (max thinking), Terra (workhorse), Luna (fast/cheap). On the unsaturated Gene Bench V1 it's still climbing at 40K tokens — the headroom is in the budget, not the model. ▸ China walls off its models — Reuters (Jul 7): Beijing weighs curbing overseas access to Alibaba, ByteDance, and Z.ai models. US locks its models down, China locks its down, everyone ends up on a VPN. ▸ Meta's "AI gulag" — Zuckerberg concedes the new AI org's bets "have not come to fruition," even at ~$145B infra spend (TechCrunch). The one product we'd try: prompt-to-mini-video-game with a shareable feed. ▸ Hardware Hut — AMD's Ryzen AI Halo Developer Desktop pairs the Ryzen AI Max+ 395's big unified memory with preinstalled isolated-PyTorch scripts — a real fix for AMD's out-of-box pain. Beat the G1A on productivity, lost on GPU. ▸ Technique Corner: Know Your Unknowns — Thariq (@trq212), an Anthropic Claude Code engineer, distilled 11 agent techniques from making the Fable 5 release video — from the "blind-spot pass" to "quiz me before I merge." Full list linked below. ▸ The Permanent Underclass — Fernando Borretti dismantles the Valley's work-or-be-left-behind doom: if AI does everything, the "overclass" is as useless as a modern aristocrat, and even perfect alignment doesn't save the pyramid. Rahul's white whale, finally on the show. ▸ AI Saves ~3% of Your Hours — An Okane read on Humlum & Vestergaard's Denmark data: ~2.8% of hours saved, almost none reaching pay. The 2026 revision says work is being reorganized below the surface. Solo builders capture the gain; converting the speedup to cash is the job. ▸ The State of the AI Economy — Exponential View, no double-counting: Gen AI scales revenue ~3× faster than internet/mobile/cloud and adds $1B every ~2 days (vs 180 in 2023) — yet it's ~0.42% of US GDP, backlog nears $2T, and CapEx is shifting from cash to debt. ▸ Does Code Cleanliness Affect Coding Agents? — SonarSource ran one agent (Opus 4.6) over 30 matched clean-vs-"slopified" repos. Pass rates barely moved; clean code just cut tokens ~7–8% (reasoning ~11%). Messy code costs the agent time, not correctness. ▸ Two Minutes to Midnight — The BIS warns runaway AI-data-center debt risks a 2008-style crunch if hyperscalers slow CapEx; an EY survey shows CEOs expecting AI headcount cuts falling 46% → 20%; Ford un-automates its QA. Clock holds at 4:45. ⏱ Chapters 00:00 Cold Open & Welcome 02:32 News: GPT-5.6 Sol, Terra & Luna 07:30 News: China Moves to Curb Overseas AI Access 09:12 News: Meta's AI "Gulag" Ships Bite-Sized Video Games 13:31 Hardware Hut: AMD Ryzen AI Halo Developer Desktop 18:08 Technique Corner: Know Your Unknowns (Thariq) 29:38 Post-Processing: No One Escapes the Permanent Underclass 39:04 Post-Processing: AI Saves ~3% of Your Hours 45:48 Deep Dive: The State of the AI Economy (Exponential View) 1:04:32 Deep Dive: Does Code Cleanliness Affect Coding Agents? 1:08:33 Two Minutes to Midnight: BIS Crash Warning, CEO Jobs Flip 1:13:25 Outro 🔗 Articles we discussed The Treadmill / News: • GPT-5.6 Sol preview — OpenAI: https://openai.com/index/previewing-gpt-5-6-sol/ • China curbs on overseas AI access — Reuters: https://www.reuters.com/world/beijing-is-looking-curbing-overseas-access-chinas-top-ai-models-sources-say-2026-07-07/ • Zuckerberg: AI agents behind schedule — TechCrunch: https://techcrunch.com/2026/07/02/mark-zuckerberg-tells-staff-that-ai-agents-havent-progressed-as-quickly-as-hed-hoped/ Hardware Hut: • AMD Ryzen AI Halo first look — PCMag: https://www.pcmag.com/news/amd-ryzen-ai-halo-first-look-giant-local-ai-power-in-a-pint-sized-box Technique Corner: • Know Your Unknowns — Thariq: https://thariqs.github.io/html-effectiveness/unknowns/ • Thariq on X: https://x.com/trq212/status/2073100352921215386 Post-Processing: • No One Escapes the Permanent Underclass — Borretti: https://borretti.me/article/no-one-escapes-the-permanent-underclass • AI Saves ~3% of Your Hours — Okane: https://okaneland.com/study/ai-productivity-roi-at-work/ Deep Dive: • State of the AI Economy — Exponential View: https://intelligence.exponentialview.co/ • Does Code Cleanliness Affect Coding Agents? (SonarSource) — arXiv: https://arxiv.org/pdf/2605.20049 Two Minutes to Midnight: • AI boom risks a financial crash — Telegraph: https://www.telegraph.co.uk/business/2026/06/28/ai-boom-risks-global-financial-crash-central-bankers-warn/ • Big Tech flips on the AI jobs wipeout — MSN: https://www.msn.com/en-us/money/careersandeducation/big-tech-has-suddenly-flipped-on-the-ai-jobs-wipeout-scenario/ar-AA27hbnR 🎙 About ADI Pod ADI 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 • humans@adipod.ai If something here gave you something to try on Monday, hit subscribe and drop a comment.

    1h 16m

  2. Jul 3

    GLM 5.2 Undercuts Opus, Self-Rewriting Harness, AI Out-Persuades Humans & Prompt Injection as Role Confusion

    AI now out-argues expert human debaters, even coaching doesn't save them. Cap its word count though, and the entire edge drops to zero.This week: GLM 5.2 undercuts Opus, Xiaomi's self-rewriting harness, OpenAI's "jalapeno" chip, prompt injection as role confusion, and the clock ticking to 4:45. Co-hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. ▸ GLM 5.2 — Z.ai's open-weight 750B MoE: Sonnet-to-Opus quality at ~1/3 the cost (~$5 vs $20 a build). Semgrep even had it beating raw Claude Code on security. ▸ Engineering jobs — SignalFire: software engineering was 2025's most resilient role (~55% of hires). Ramp: AI adopters grew headcount 10.2%, entry-level share slid 50%→34%. ▸ Tool Shed — Xiaomi's Harness X uses an AEGIS judge to rewrite its own scaffolding (Qwen3 5.9B, +44% planning). Ornith 1.0 does RL on the weights and the solution together. ▸ Hardware Hut — OpenAI + Broadcom's "jalapeno" inference chip: from the wafer photo, a systolic-array ASIC, six HBM stacks, cost-per-watt beating NVIDIA, ~9 months to tape-out. ▸ Post-Processing — Hackenberg et al.: AI out-argues lay people (~8pp) and trained debaters (~4.6pp). Cap it to a human's word count and the edge hits 0.0pp. It tripled charity donations. ▸ Listener Mail — Bloomberg on Silicon Valley engineers running a dozen agents at their kids' games. Dan's version is cognitive debt; ChainGuard wants managers at the 50th percentile of usage. ▸ Deep Dive — Yu, Cui & Hadfield-Menell: jailbreaks are a model mistaking your words for its own thoughts. The "wearing green" trick breaks GPT-5-mini and o4-mini. ▸ Two Minutes to Midnight — Masa Son doubts Musk's data-centers-in-space (~7% of the cost is electricity). OpenAI may delay its IPO toward $760B; Epoch AI sees capex outrun cash flow by Q3 2026. Clock 5:00 → 4:45. Chapters 00:00 Cold Open & Welcome02:29 News: GLM 5.2 Undercuts Opus at a Third the Cost10:37 News: Engineering Jobs, the Most Resilient?16:55 Tool Shed: Xiaomi's Harness X & Ornith22:35 Hardware Hut: OpenAI x Broadcom "Jalapeno" Chip27:55 Post-Processing: AI Out-Persuades Expert Humans39:09 Listener Mail: AI Anxiety in Silicon Valley46:43 Deep Dive: Prompt Injection as Role Confusion1:02:43 Dan's Rant: Token-Maxing Is Dead1:07:52 Two Minutes to Midnight: Space Data Centers, OpenAI's IPO, Capex Articles we discussed The Treadmill / News:• GLM 5.2 vs Opus — techstackups: https://techstackups.com/comparisons/glm-5.2-vs-opus/• GLM 5.2 beats Claude on cyber benchmarks — Semgrep: https://semgrep.dev/blog/2026/we-have-mythos-at-home-glm-52-beats-claude-in-our-cyber-benchmarks/• Engineering jobs are the most resilient — TechCrunch: https://techcrunch.com/2026/06/24/ai-was-supposed-to-kill-engineering-jobs-but-new-data-suggests-theyre-the-most-resilient/• Companies hire more after AI adoption — Ramp: https://ramp.com/data/heavy-ai-adopters-hire-more Tool Shed:• Xiaomi HarnessX rewrites its own scaffolding — VentureBeat: https://venturebeat.com/orchestration/xiaomis-harnessx-rewrites-its-own-ai-scaffolding-mid-task-and-smaller-models-gain-the-most• Ornith 1.0 — Deep Reinforce: https://deep-reinforce.com/ornith_1_0.html• Simon Willison on Ornith: https://simonwillison.net/2026/Jun/29/ornith/ Hardware Hut:• OpenAI x Broadcom "Jalapeno" inference chip — OpenAI: https://openai.com/index/openai-broadcom-jalapeno-inference-chip/ Post-Processing:• AI systems out-persuade expert humans (Hackenberg et al.) — arXiv: https://arxiv.org/pdf/2606.16475 Listener Mail:• AI anxiety is fueling burnout across Silicon Valley — Bloomberg: https://www.bloomberg.com/news/articles/2026-06-26/ai-anxiety-is-fueling-burnout-across-silicon-valley-s-tech-workers Deep Dive:• Prompt Injection as Role Confusion (Yu, Cui, Hadfield-Menell): https://role-confusion.github.io/ Two Minutes to Midnight:• Betting against Musk's AI vision — MSN: https://www.msn.com/en-us/money/other/why-one-of-tech-s-biggest-gamblers-is-betting-against-elon-musk-s-ai-vision/ar-AA26Fe1f• Archive mirror: https://archive.ph/UdzT4• Hyperscaler capex vs cash flow — Epoch AI: https://epoch.ai/data-insights/hyperscaler-capex-vs-cash-flow About ADI Pod ADI 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• humans@adipod.ai If something here gave you something to try on Monday, hit subscribe and drop a comment. (00:00) - Cold Open & Welcome (02:29) - News: GLM 5.2 Undercuts Opus at a Third the Cost (10:37) - News: Engineering Jobs, the Most Resilient? (16:55) - Tool Shed: Xiaomi's Harness X & Ornith (22:35) - Hardware Hut: OpenAI x Broadcom "Jalapeno" Chip (27:55) - Post-Processing: AI Out-Persuades Expert Humans (39:09) - Listener Mail: AI Anxiety in Silicon Valley (46:43) - Deep Dive: Prompt Injection as Role Confusion (01:02:43) - Dan's Rant: Token-Maxing Is Dead (01:07:52) - Two Minutes to Midnight: Space Data Centers, OpenAI's IPO, Capex

    1h 15m

  3. Jun 26

    Grok Buys Cursor, MidJourney Goes Hardware, Hermes Agent & Evaluation-Driven Development

    MidJourney — the AI image company — just quit image generation to build 50,000 spas that scan your body slice by slice. Then the week got weirder. Co-hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. ▸ SpaceX buys Cursor — Elon's SpaceX (xAI/"Grok Cursor") is acquiring Cursor for $60B in Class A common stock — a ~60x multiple on ~$1B revenue, largely to buy an enterprise foothold. (Shimin: the first real sign of an AI-tool consolidation phase.) ▸ MidJourney goes hardware — the image-gen pioneer is licensing micro-ultrasound chips to build 50,000 body-scan spas (first one: SF, 2027), aiming for a billion scans a month. Fully private, no VC backers, a self-described "community research lab." Terabytes/second — ~500 hours of HD video per single second of scan. ▸ Tool Shed: Hermes Agent (Nous Research) — the plugin-maximalist opposite of a minimal harness like Pi: built-in memory, a self-learning skill loop, cron scheduling, swappable memory providers, and ~20 chat channels out of the box. Dan: "parachuting in with sixteen crates of supplies and a film crew." ▸ Is AI ruining our skills? (Nature) — physicians' precancerous-lesion detection fell from 28.4% to 22.4% once the AI tool was removed; 52 engineers scored 50% on understanding their own code with AI vs 67% without. Cognitive debt is showing up in the data. ▸ Claude Code is a video game (Provi.me) — the "one more prompt" loop that keeps you up three hours past bedtime, and why AI finally made B2B SaaS addictive. Plus the "agent dice" repo: roll a natural 20 and a stop hook makes the agent reflect and write itself a skill. ▸ Evaluation-Driven Development (Decoding AI) — treat every AI feature as a hypothesis and gate the PR on an offline eval pipeline (built on Opik) instead of unit tests. Gold-standard vs synthetic datasets, code-metric vs LLM-as-judge evaluators, and an "aggression" dial for how big a jerk your reviewer is. (Shimin: Newtonian physics → quantum mechanics.) ▸ Two Minutes to Midnight — ChatGPT slips under 50% share (46.4%; Gemini 27.7%, Claude 10.3%), Nvidia raises $25B in its first bond deal since 2021, and Ed Zitron walks OpenAI's FT-verified financials ($38.5B loss in 2025). ~2B users — one in four people on Earth; no 10x left. Clock moved up to 5:00. ⏱ Chapters 00:00 Cold Open & Welcome01:50 News: SpaceX Buys Cursor for $60B04:46 News: MidJourney Pivots to Body-Scan Spas11:45 Tool Shed: Hermes Agent (Nous Research)19:54 Post-Processing: Is AI Ruining Our Skills? (Nature)27:13 Post-Processing: Claude Code Is a Video Game35:23 Post-Processing: Evaluation-Driven Development (EDD)41:44 Two Minutes to Midnight: ChatGPT Under 50%, Nvidia Debt, OpenAI's Numbers55:06 Outro 🔗 Articles we discussed News:• SpaceX to acquire Cursor — CNBC: https://www.cnbc.com/2026/06/16/spacex-spcx-cursor-acquisition-ipo.html• MidJourney's medical pivot — MidJourney: https://www.midjourney.com/medical/blogpost Tool Shed:• Hermes Agent docs — Nous Research: https://hermes-agent.nousresearch.com/docs/ Post-Processing:• Is AI ruining our skills? Early results are in — Nature: https://www.nature.com/articles/d41586-026-01947-1• Claude Code is a video game — Provi.me: https://provi.me/cc-like-video-games• How Evaluation-Driven Development (EDD) works — Decoding AI (Paul Easton & Alejandro Aboy): https://www.decodingai.com/p/5b766861-0001-494f-a37f-4d4eb104dcfa Two Minutes to Midnight:• ChatGPT's market share slips below 50% for the first time — TechCrunch: https://techcrunch.com/2026/06/16/chatgpts-market-share-slips-below-50-for-first-time/• Nvidia seeks to raise over $25B in first bond deal since 2021 — Ars Technica: https://arstechnica.com/ai/2026/06/chipmaker-nvidia-seeks-to-raise-over-25b-in-first-bond-deal-since-2021/• Exclusive: OpenAI's financials — Where's Your Ed At (Ed Zitron): https://www.wheresyoured.at/exclusive-openai-financials/ 🎙 About ADI Pod ADI 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. Hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. New episodes Tuesdays. • https://www.adipod.ai• humans@adipod.ai If something here changed your mind or gave you something to try on Monday, hit subscribe and leave a comment with what you tried.

    56 min

  4. Jun 19

    Fable 5 Ban, Meta's AI Gulag, Elias Thorne & What is Loop Engineering?

    Three days after Fable 5 launched, the US government banned it — for every foreign national on Earth, including Anthropic's own employees. Then it got weirder. This week on ADI Pod: the Fable 5 export ban, Meta's applied-AI "gulag," the Elias Thorne dataset virus, loop engineering, a local DeepSeek V4 demo, the paper that shatters Dunning-Kruger, and NBER's bubble math. Co-hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. ▸ Fable 5 & Mythos 5, export-banned — a national-security order cut access for all foreign nationals (even Anthropic's own staff) in ~90 minutes, reportedly after an AWS jailbreak claim; likely the end of universal frontier-model access. Shimin had a near-"AI psychosis" moment using it to design a novel drone. ▸ Meta's "AI Gulag" — Alexandr Wang's unit drafts laid-off engineers to write puzzles and label data to train Meta's weaker models, on full salary and RSUs; the "gulag" label is a stretch, but the internal drama is real. ▸ The Elias Thorne mystery (404 Media) — a lighthouse keeper seeded by ~111 ChatGPT-3.5 chats became a "dataset virus" now in ~88% of AI stories and "authoring" books on Amazon across every lab (Cornell's Hamilton & Mimno). ▸ AI is fast, the economy isn't (howfastis.ai) — task horizons double every ~6 months, but weak-link / Theory-of-Constraints bottlenecks (Chad Jones; Goldratt) keep growth near 2%/yr; human judgment is the constraint AI can't yet remove. ▸ Loop engineering (Addy Osmani) — six pieces turn a bare /loop (Ralph loop) into a real agent harness: automations, worktrees, skills, plugins/connectors, subagents (split the worker from the reviewer), and memory. It amplifies whatever judgment you bake into your skills. ▸ Deep Dive — "Beyond the Steeper Curve" (Christopher Koch) — AI doesn't steepen Dunning-Kruger, it shatters it: "metacognitive decoupling" unglues output quality from self-assessment. Plus the "slop grenade" and the sycophancy trap (No One's Happy). ▸ Vibe & Tell — Dan runs DeepSeek V4 Flash locally ("DS4," the dwarf star runner) on a Framework Ryzen 395 Max over ROCm, ~14 tok/s, wired to Pi agent — ~$4,000 of hardware, no cloud. ▸ Two Minutes to Midnight — Claude on Apple's foundation-model backend (a commoditization tell), the end of subsidized inference, and an NBER paper pricing genuine insolvency risk into the AI build-out. Clock set back to 5:30. ⏱ Chapters 00:00 Cold Open & Welcome 02:01 News: The US Government Bans Fable 5 & Mythos 5 10:35 News: Meta's "AI Gulag" (feat. Rahul) 14:39 Post-Processing: The Elias Thorne Mystery 21:12 Post-Processing: AI Is Fast, the Economy Isn't (howfastis.ai) 29:22 Post-Processing: Loop Engineering (Addy Osmani) 36:37 Deep Dive: Beyond the Steeper Curve (Dunning-Kruger, Shattered) 43:46 Deep Dive: Appearing Productive & the Slop Grenade 51:51 Vibe & Tell: DeepSeek V4 Flash at Home (DS4) 57:39 Two Minutes to Midnight: Apple Foundation Models, Cheaper Inference, NBER Bubble Math 1:08:44 Outro 🔗 Articles we discussed News: • Fable & Mythos access update — Anthropic: https://www.anthropic.com/news/fable-mythos-access • Anthropic lobbies the White House over the Mythos/Fable ban — Axios: https://www.axios.com/2026/06/14/anthropic-white-house-mythos-fable • Meta's months-old AI unit is a "soul-crushing gulag," say the engineers stuck inside it — TechCrunch: https://techcrunch.com/2026/06/12/metas-months-old-ai-unit-is-a-soul-crushing-gulag-say-the-engineers-stuck-inside-it/ Post-Processing: • Chatbots keep telling stories about lighthouse keeper Elias Thorne — 404 Media: https://www.404media.co/elias-thorne-chatbots-llms-chatgpt-lighthouse-keeper-story/ • How Fast Is AI? — Emory Taziki: https://howfastis.ai/ • Loop Engineering — Addy Osmani: https://addyosmani.com/blog/loop-engineering/ Deep Dive: • Beyond the Steeper Curve: AI-Mediated Metacognitive Decoupling and the Limits of the Dunning-Kruger Metaphor — Christopher Koch (arXiv): https://arxiv.org/html/2603.29681 • Appearing Productive in the Workplace — No One's Happy: https://nooneshappy.com/article/appearing-productive-in-the-workplace/ Two Minutes to Midnight: • Claude SDK for Apple Foundation Models — Claude Platform docs: https://platform.claude.com/docs/en/cli-sdks-libraries/libraries/apple-foundation-models • Can tech companies learn to love cheaper AI models? — TechCrunch: https://techcrunch.com/2026/06/09/can-tech-companies-learn-to-love-cheaper-models/ • What Investment Data Implies About the AI Transition — NBER Working Paper w35290 (Walter & Walter): https://www.nber.org/system/files/working_papers/w35290/w35290.pdf 🎙 About ADI Pod ADI 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. Hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. New episodes Fridays. • https://www.adipod.ai • humans@adipod.ai If something here changed your mind or gave you something to try on Monday, hit subscribe and leave a comment with what you tried.

    1h 10m

  5. Jun 12

    Claude Fable 5 is Here! Plus: Meta AI Hack, LLMs as Black Boxes, and Future of Agents

    "I think this is the first time ever where the user has no control over which model you're actually using." — Shimin on Fable 5's safety fallback, which answers blocked questions with Opus 4.8 instead. This week on ADI Pod: Anthropic ships Fable 5 and Mythos 5, hackers sweet-talk Meta's AI support bot into handing over Instagram accounts (including Obama's), interpretability opens up the black box, Chris Roth's Future of Agents, and Shimin demos Inhabited Design. Rahul's back; the clock moves to 5:20. ▸ Fable 5 & Mythos 5 (Anthropic) — mostly incremental benchmarks, a step change in spatial reasoning (Blueprint Bench 2), scales cleanly with thinking tokens where Opus 4.8 zigzags, and it beats Pokémon FireRed on vision alone. Fable 5's safety classifier blocks cybersecurity and biohazard prompts and answers with Opus 4.8 instead; Mythos 5 stays unguarded for the Project Glasswing cohort. Subscription access ends June 22, then usage-based pricing only. ▸ The Meta AI Hack (Krebs on Security) — the recipe spread on Telegram from May 31: VPN exit near the target's hometown, then ask Meta's AI support assistant to send reset codes to your new email. It complies. Pro-Iranian hackers defaced the Obama White House Instagram and a Space Force account. Why does a support bot hold elevated permissions to replace a boring form? Red teams will make bank. ▸ LLMs Are Not the Black Box You Were Promised (Jay Hack) — on Anthropic's "On the Biology of a Large Language Model." Circuit tracing with a sparse replacement model — one neuron per concept — shows features firing in sequence: Dallas, then Texas, then Austin. Models plan rhymes ahead of the line, and they confabulate how they do math. Shimin: metacognition may be the durable human advantage. ▸ The Future of Agents (Chris Roth) — personal agents, bring-your-own-agent trust boundaries, super apps as agent clients, open standards (MCP, A2A, AG-UI), enterprise open source, generative UI. The hosts push back: code is cheap enough that adapters kill any standard's network effects. Shimin migrated his Pi-agent skills to Claude Code with one prompt. There is no AI moat. ▸ Vibe and Tell: Inhabited Design — Shimin's open-source skill for escaping RLHF attractor states (the same stock ticker and Bloomberg yellow on every finance landing page). Verbalized sampling plus intent-factored generation: sample uniformly over designer, typography, and inspiration, then run two convergence loops. Dan's verdict: "it still feels like a human designer did it." ▸ Two Minutes to Midnight — Google will pay SpaceX $920M a month for compute at xAI data centers; Google already owns a chunk of SpaceX. The S&P 500 won't bend its rules for SpaceX and OpenAI fast-track inclusion. Alphabet upsizes its equity raise to $84.75B against roughly $190B in 2026 capex. Founders Fund's $20M SpaceX check from 2008 is now worth $26–52B. Clock: 5:30 → 5:20. ⏱ Chapters00:00 Welcome & Rundown01:20 News Threadmill: Fable 5 & Mythos 5 Launch10:59 News Threadmill: The Meta AI Support Bot Hack18:39 Post Processing: LLMs Are Not the Black Box You Were Promised28:15 Post Processing: The Future of Agents51:22 Vibe and Tell: Inhabited Design57:49 Two Minutes to Midnight: SpaceX, S&P 500, Alphabet1:07:47 Outro 🔗 Articles we discussed News:• Claude Fable 5 — Anthropic: https://www.anthropic.com/claude/fable• Hackers Used Meta's AI Support Bot to Seize Instagram Accounts — Krebs on Security: https://krebsonsecurity.com/2026/06/hackers-used-metas-ai-support-bot-to-seize-instagram-accounts/ Post Processing:• LLMs are not the Black Box you were promised — Jay Hack: https://www.jay.ai/blog/llms-are-not-a-black-box• The Future of Agents — Chris Roth: https://cjroth.com/blog/2026-06-03-future-of-agents Vibe and Tell:• Inhabited Design — Shimin Zhang (Wolf Peach Labs): https://wolfpeachlabs.com/inhabited-design/ Two Minutes to Midnight:• Google to pay SpaceX $920M a month for xAI compute — CNBC: https://www.cnbc.com/2026/06/05/google-to-pay-spacex-920-million-a-month-for-xai-compute-capacity.html• S&P 500 blocks fast SpaceX entry — Ars Technica: https://arstechnica.com/tech-policy/2026/06/sp-500-blocks-fast-spacex-entry-wont-waive-rule-for-unprofitable-ai-firms/• Alphabet to raise $84.75B in upsized equity offering — Reuters: https://www.reuters.com/legal/transactional/alphabet-raise-8475-billion-upsized-equity-offering-fund-ai-ambitions-2026-06-03/ 🎙 About ADI Pod ADI Pod (Artificial Developer Intelligence) is a weekly podcast about AI and software development for working developers. We read hundreds of links and newsletters each week so you don't have to. Hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. New episodes Tuesdays. • https://www.adipod.ai• humans@adipod.ai If this gave you something to try on Monday, subscribe and tell us. (00:00) - Welcome & Rundown (01:20) - News Threadmill: Fable 5 & Mythos 5 Launch (10:59) - News Threadmill: The Meta AI Support Bot Hack (18:39) - Post Processing: LLMs Are Not the Black Box You Were Promised (28:15) - Post Processing: The Future of Agents (51:22) - Vibe and Tell: Inhabited Design (57:49) - Two Minutes to Midnight: SpaceX, S&P 500, Alphabet (01:07:47) - Outro

    1h 9m

  6. Jun 5

    Claude Opus 4.8, Undocumented Claude Code Features, Eval Harness for AI Skills, Pope on AI

    "Every time you vibe code, you're gonna spend compute to skip the bottleneck of code review." — Shimin's cold open on ep-28. Trading compute for human labor: that's the default that's coming. This week on ADI Pod: Claude Opus 4.8 and Anthropic's dynamic-workflow tool, Pope Leo XIV's AI encyclical, a deep read of the Claude Code source code, a Pinterest method for testing whether your AI skills actually fire, two essays on senior engineering and the "dead economy," and a bubble check full of S-1s. Rahul's out this week; the clock moves up to 5:30. ▸ Claude Opus 4.8 + the Dynamic Workflow Tool (TechCrunch): a 41-day fast-follow to 4.7. The new "dynamic workflow" is extra-high thinking plus a huge fan-out of coordinated parallel agents — the hosts call it "Gastown, by Anthropic." Dan likes it more than 4.7, but it hallucinated file names that don't exist and ate a full token budget in 25 minutes. Likely a Mythos distill, not a new base model. ▸ Pope Leo XIV's AI Encyclical — "Magnifica Humanitas" (Vatican): "On Safeguarding the Human Person in the Time of Artificial Intelligence." The Pope gets that models are grown, not developed, warns against pretending AI is neutral, and ties automation to worker protection. Anthropic's Chris Olah was in the room. Shimin's take: better AI takes than most Fortune 500 CEOs. ▸ I Read the Claude Code Source Code (Building Better): the undocumented stuff. A pre-tool-use hook can rewrite a tool's input mid-flight, return allow/deny with a reason, and inject context. Skills take undocumented front-matter (model + effort). Plus where settings.json really lives, and the auto-memory and "dream" toggles. ▸ Technique Corner — An Engineer's Guide to Better AI Skills (Pinterest): a test harness for skill invocation — 15 positive prompts, 5 negative, 5 runs each. Codex went 73%→95% with everything combined; Claude went 62%→73% on a single change and got worse when you combined them. Asking the AI to improve the skill didn't help. ▸ Post Processing — Is This Sustainable? (Jamie Hurst): seniors absorbed AI's rising stakes before juniors did. You skip the RFC and just build the thing. The scary part: AI depth is perishable in ~18 months; what lasts is taste and judgment. ▸ Post Processing — The Dead Economy Theory (Owen McGrann): a turn-by-turn case that replacing workers with AI eats its own market. Peter Thiel, a Valley misread of Nietzsche, and UBI. Shimin pushes back while half-infected by the inevitability virus. ▸ Two Minutes to Midnight (SEC + Qazinform): SpaceX's S-1 claims a $26.5T market that's mostly "AI" and says "truth seeking" 39 times. Anthropic overtakes OpenAI as the most valuable AI startup on a $65B Series H (~3x its February mark) plus a confidential S-1. Microsoft pulls Claude Code back to Copilot on cost. Clock -> 5:30. ⏱ Chapters 00:00 Cold Open & Welcome 02:09 News: Claude Opus 4.8 & the Dynamic Workflow Tool 08:27 News: Pope Leo XIV's AI Encyclical 14:17 ToolShed: I Read the Claude Code Source Code 22:10 Technique Corner: Do Your AI Skills Actually Fire? 29:15 Post Processing: Is This Sustainable? (Senior Eng in the AI Age) 35:30 Post Processing: The Dead Economy Theory 42:12 Two Minutes to Midnight: SpaceX's S-1 & the $26.5T "AI" TAM 45:50 Two Minutes to Midnight: Anthropic Overtakes OpenAI 53:35 Outro 🔗 Articles we discussed News: • Anthropic releases Opus 4.8 with new dynamic workflow tool — TechCrunch: https://techcrunch.com/2026/05/28/anthropic-releases-opus-4-8-with-new-dynamic-workflow-tool/ • Magnifica Humanitas (encyclical on AI) — Pope Leo XIV / Vatican: https://www.vatican.va/content/leo-xiv/en/encyclicals/documents/20260515-magnifica-humanitas.html • I Read the Claude Code Source Code — Building Better: https://buildingbetter.tech/p/i-read-the-claude-code-source-code Technique Corner: • An Engineer's Guide to Better AI Skills — Pinterest Engineering: https://medium.com/pinterest-engineering/an-engineers-guide-to-better-ai-skills-implementing-a-testing-process-to-optimize-agent-a000c9c9abcd Post Processing: • Is This Sustainable? — Jamie Hurst: https://jamiehurst.co.uk/2026-05-24_ai-sustainable • The Dead Economy Theory — Owen McGrann: https://www.owenmcgrann.com/p/the-dead-economy-theory Two Minutes to Midnight: • SpaceX (Space Exploration Technologies) Form S-1 — SEC EDGAR: https://www.sec.gov/Archives/edgar/data/1181412/000162828026036936/spaceexplorationtechnologi.htm • Anthropic surpasses OpenAI to become world's most valuable AI startup — Qazinform: https://qazinform.com/news/anthropic-surpasses-openai-to-become-worlds-most-valuable-ai-startup 🎙 About ADI Pod ADI 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. Hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. New episodes Tuesdays. • https://www.adipod.ai • humans@adipod.ai If something here gave you something to try on Monday, subscribe and tell us what you tried. (00:00) - Cold Open & Welcome (02:09) - News: Claude Opus 4.8 & the Dynamic Workflow Tool (08:27) - News: Pope Leo XIV's AI Encyclical (14:17) - News / Tool Shed: I Read the Claude Code Source Code (22:10) - Technique Corner: Do Your AI Skills Actually Fire? (29:15) - Post Processing: Is This Sustainable? (35:30) - Post Processing: The Dead Economy Theory (42:12) - Two Minutes to Midnight: SpaceX's S-1 (45:50) - Two Minutes to Midnight: Anthropic Overtakes OpenAI (53:35) - Outro

    55 min

  7. May 29

    OpenAI Beats Musk, Gemini 3.5 Flash & AI Burnout Mitigation

    "Sam Altman won in court against Elon Musk. But, really, we all lost." That's the New Yorker headline Dan brought to ep-27 — and the question under it is whether any one person should own AI safety. This week on ADI Pod: the OpenAI–Musk verdict and who really owns AI safety, Gemini 3.5 Flash in AI Overviews, a $48K home GPU server, AI burnout from two angles, the "$100M startup in your laptop" myth, the slop grenade, and an IPO squeeze that could funnel ~10% of the major indexes into three AI firms. Rahul's out this week; clock moves back to 6:15. ▸ OpenAI v. Musk (The New Yorker): OpenAI wins on a statute-of-limitations technicality. Musk's lawyer argues "we could all die" from AI; the judge notes he'd mean it more if he didn't fund xAI. The courtroom "butt pillows" become the complacency metaphor. ▸ Gemini 3.5 Flash: shipped Flash-only into AI Overviews (Dan's bet: it's on TPUs). Mathier than 3.1 but fewer results; the viral "can't search 'disregard'" bug was a harness failure. The pelican it drew looks dressed for a Miami crypto conference (h/t Simon Willison). ▸ Hardware Hut — was a $48K GPU server worth it? (rosmine.ai): an ex-FAANG researcher's 6× RTX 6000 Ada rig breaks even near 80% utilization, then ~$125/month and constant riser failures. Shimin's version: a 128GB Mac for local models, or keep paying Anthropic? ▸ Technique Corner — AI burnout (Evil Martians + Siddhant Khare): cap parallel agents at 3–4, keep hands on the keyboard, accept 70% and hand-code the rest. Shimin's confession: seven Claude Code sessions after work. Capper: Microsoft cancels Claude Code subs after costs top human devs. ▸ Post Processing — Human Bottlenecks (borretti.me): the $100M startup in your laptop stays there because the limiter was always you — judgment, energy, executive function — not the tools. ▸ Dan's Rant — the Slop Grenade (noslopgrenade.com): paste raw Claude output at a coworker instead of an answer and you've thrown one. The successor to nohello.com. Fix: lead with your one-line take, then attach the output for the full kaboom. ▸ Two Minutes to Midnight (Morningstar + Is AI Profitable Yet?): SpaceX/OpenAI/Anthropic could add ~10% to the Morningstar 100; Nasdaq cut its post-IPO wait from 12 months to ~15 trading days. On isaiprofitable.com only Nvidia is green (+$253B); Amazon leads capex at −$291B. Clock → 6:15. 🔗 Articles we discussed News: • OpenAI Won, But We All Lost — The New Yorker: https://www.newyorker.com/news/letter-from-silicon-valley/sam-altman-won-in-court-against-elon-musk-but-really-we-all-lost • Gemini 3.5: Frontier Intelligence With Action — Google: https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-3-5/#gemini-3-5-flash • Gemini 3.5 Flash hands-on — Simon Willison: https://simonwillison.net/2026/May/19/gemini-35-flash/ Hardware Hut: • Was My $48K GPU Server Worth It? — rosmine.ai: https://rosmine.ai/2026/05/13/was-my-48k-gpu-worth-it/ Technique Corner: • AI-Assisted Engineers Are Burning Out — Evil Martians: https://evilmartians.com/chronicles/ai-assisted-engineers-are-burning-out-is-this-fine • AI Fatigue Is Real — Siddhant Khare: https://siddhantkhare.com/writing/ai-fatigue-is-real Post Processing: • Human Bottlenecks — borretti.me: https://borretti.me/article/human-bottlenecks Dan's Rant: • No Slop Grenade: https://noslopgrenade.com Two Minutes to Midnight: • The SpaceX IPO: How US Index Funds Will Adapt — Morningstar (Zachary Evans): https://global.morningstar.com/en-ca/funds/spacex-ipo-how-us-stock-index-funds-will-adapt • Is AI Profitable Yet?: https://isaiprofitable.com/ 🎙 About ADI Pod ADI 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. Hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. New episodes Tuesdays. • https://www.adipod.ai • humans@adipod.ai If something here gave you something to try on Monday, subscribe and tell us what you tried. (00:00) - Cold Open & Welcome (02:15) - News: OpenAI Beats Musk — "We All Lost" (The New Yorker) (07:20) - News: Gemini 3.5 Flash & the Pelican Test (13:05) - Hardware Hut: Was a $48K GPU Server Worth It? (19:14) - Technique Corner: AI-Assisted Engineers Are Burning Out (28:27) - Microsoft Cancels Claude Code — the Token Pendulum (30:35) - Post Processing: Human Bottlenecks (36:10) - Raising the First AI-Native Generation (40:05) - Dan's Rant: The Slop Grenade (44:49) - Two Minutes to Midnight: The SpaceX IPO Index Squeeze (49:32) - Is AI Profitable Yet? (Capex by the Billions) (55:20) - Outro

    56 min

  8. May 22

    LLM Neuralanatomy with David Noel Ng, Forward Deployed Everybody, Preferences Revealed by AI

    This week on ADI Pod: Mira Murati's Thinking Machines ships its first product (interaction models), Meta employees fight the mouse-tracking program with flyers, and the Palantir-coined "forward deployed engineer" job title quietly takes over the post-AI engineering org chart. Our sit-down is with Dr. David Noel Ng (https://dnhkng.github.io/) — author of the LLM Neuroanatomy series we covered a few weeks back. He explains why he watched action potentials race down rat neurons at 10,000 fps before getting into LLM interpretability. Dan runs DeepSeek-V4 Flash on a 128 GB Ryzen 395 Max box and vibe-codes an ESP32 home dashboard in C. Deep dive on a paper asking whether AI should obey what you say or what you actually do. And in Two Minutes to Midnight: Cerebras pops 108% on IPO day, Anthropic passes OpenAI on Ramp business data, and we read Andy Hall on the politics of jobless prosperity. Clock stays at 6 minutes. Co-hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. ▸ Interaction Models — Thinking Machines' first product. A small Qwen 3.5–class "interaction model" runs the UI; a background model handles the heavy lift. They credit the pattern to Qwen, not themselves. Multimodal by default, and surprisingly snappy in the demo. ▸ Meta vs. its own employees — Meta posted flyers around its offices reading "don't want to work at the employee data extraction factory?" after announcing it would record keystrokes, mouse movements, and screens to train internal AI. Last episode's "Model Capability Initiative" story got worse. ▸ Here Comes Forward Deployed Everybody (Scott Werner / works on my machine) — Salesforce moves to an API-only data model. Palantir's "forward deployed engineer" title (originally called "delta") becomes the new pit-crew role across every department. We argue Jevons paradox vs. just-rebranding-the-least-glamorous-job: 20 marketers + 5 pit crews → 30 marketers + 10 pit crews as productivity rises. ▸ Sit-down: Dr. David Noel Ng on LLM Neuroanatomy — fluorescent dyes that change color with membrane voltage, what brain-microchip interfacing taught him about feature attribution, and why interpretability deserves the same first-principles rigor as wet-lab biology. New posts coming. ▸ Vibe Intel — Dan runs Antires's specialized DeepSeek-V4 Flash fork of llama.cpp. Q2 quant on the front, full experts on the back, SSD-cached prefills, ~10 tokens/sec on a Ryzen AI Max+ 395 with 128 GB unified memory. Output peg: Sonnet 4.5-ish. Plus an ESP32 / ESPHome dashboard with a several-thousand-line vibe-coded C lambda that, against all reasonable expectation, works. ▸ Deep Dive — "Should I State or Should I Show?" (Keaton Ellis & Wanying Huang) — three AIs given the same lottery decisions: prompt-only AI hit 70% match with the human; data-only AI hit 75%; both-AI dropped to the worst of the three because it defaulted to the prompt 66% of the time when prompt and behavior conflicted. Implication: if EU AI Act transparency rules force you to honor stated preferences, you're literally picking the worst-performing model. ▸ Two Minutes to Midnight — Cerebras raises $5.5B in IPO, stock pops 108% on day one (claim: ~80× memory throughput vs comparable NVIDIA GPUs). Anthropic now holds 34.4% of Ramp-card-paying businesses, beating OpenAI for the first time. Andy Hall's "Politics of Jobless Prosperity": 2% unemployment jump is the line in the sand for political stability — opens with FDR's 1944 State of the Union. Clock held at 6 minutes; nothing this week jumped the needle. ⏱ Chapters 00:00 Cold Open & Welcome 02:08 News: Interaction Models from Thinking Machines 05:57 News: Meta Employees Protest the Mouse-Tracking Program 10:53 Post-Processing: Here Comes Forward Deployed Everybody (Scott Werner) 22:50 Sit-Down: Dr. David Noel Ng on LLM Neuroanatomy 40:05 Vibe N Tell: DeepSeek-V4 Flash at Home on a 395 Max 45:37 Vibe N tell: ESP32 Home Dashboards via Vibe Coding 48:51 Deep Dive: Should I State or Should I Show? (Ellis & Huang) 1:04:24 Two Minutes to Midnight: Cerebras IPO, Anthropic vs OpenAI, Jobless Prosperity 1:09:34 Outro 🔗 Articles we discussed News: • Interaction Models — Thinking Machines: https://thinkingmachines.ai/blog/interaction-models/ • Meta employees protest the mouse-tracking program — Engadget: https://www.engadget.com/2172212/meta-employees-are-protesting-the-companys-mouse-tracking-program/ Post-Processing: • Here Comes Forward Deployed Everybody — Scott Werner (works on my machine): https://worksonmymachine.ai/p/here-comes-forward-deployed-everybody Sit-Down — Dr. David Noel Ng: • Substack: https://dnhkng.substack.com/ • Site: https://dnhkng.github.io/* Rest of David's home AI Lab Build Story: https://dnhkng.github.io/posts/hopper/ Deep Dive: • Should I State or Should I Show? Aligning AI with Human Preferences — Keaton Ellis & Wanying Huang (arXiv): https://arxiv.org/html/2603.29317v1 Two Minutes to Midnight: • Cerebras raises $5.5B, kicks off 2026's IPO season — TechCrunch: https://techcrunch.com/2026/05/14/cerebras-raises-5-5b-kicking-off-2026s-ipo-season-with-a-bang/ • Cerebras: Faster Tokens, Please — SemiAnalysis: https://newsletter.semianalysis.com/p/cerebras-faster-tokens-please • Anthropic now has more business customers than OpenAI (per Ramp data) — TechCrunch: https://techcrunch.com/2026/05/13/anthropic-now-has-more-business-customers-than-openai-according-to-ramp-data/ • The Politics of Jobless Prosperity — Andy Hall (Free Systems): https://freesystems.substack.com/p/the-politics-of-jobless-prosperity 🎙 About ADI Pod ADI 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. Hosts: Shimin Zhang, Dan Lasky, Rahul Yadav. New episodes Tuesdays. • https://www.adipod.ai • humans@adipod.ai If something in this episode changed your mind or gave you something to try on Monday, hit subscribe and leave a comment with what you tried. (00:00) - Cold Open & Welcome (02:08) - News: Interaction Models from Thinking Machines (05:57) - News: Meta Employees Protest the Mouse-Tracking Program (10:53) - Post-Processing: Here Comes Forward Deployed Everybody (Scott Werner) (22:50) - Sit-Down: Dr. David Noel Ng on LLM Neuroanatomy (40:05) - Vibe N Tell: DeepSeek-V4 Flash at Home on a 395 Max (45:37) - Vibe N tell: ESP32 Home Dashboards via Vibe Coding (48:51) - Deep Dive: Should I State or Should I Show? (Ellis & Huang) (01:04:24) - Two Minutes to Midnight: Cerebras IPO, Anthropic vs OpenAI, Jobless Prosperity (01:09:34) - Outro

    1h 10m
See All (33)

Ratings & Reviews

5
out of 5
8 Ratings

About

Three engineer friends argue about AI so you don't have to. Shimin Zhang, Dan Lasky, and Rahul Yadav are working developers who've been watching AI transform their profession in real time, and they got opinions on the robot takeover. Every week the three get together to riff on the latest AI news, geek out over research papers, roast each other's tool choices, and occasionally have an existential crisis about whether the craft is dying or just getting weird. What you're signing up for: - AI news without the LinkedIn cringe: model drops, acquisitions, open-source drama, and the other stuff that actually matters if you write code for a living. - Technique corner: real tips from the trenches: spec-driven development, multi-agent orchestration, Claude.md tricks, and all the ways they've wasted hours so you don't have to. - Two Minutes to Midnight: the show's running AI bubble tracker, complete with circular funding diagrams, hyperscaler CAPEX math, and a doomsday clock they keep arguing about moving. - Deep dives that (occasionally) go deep: hallucination neurons, agentic memory, workflow automation economics, LLM architectures the papers nobody else is covering because they're hard. - Dan's Rant: Dan frequently gets mad about things. It's a whole thing. - The feelings segment: Yes, Shimin reads Tennyson on a tech podcast. Yes, Rahul wrote an AI-generated country song. No, they're not sorry. Three friends with strong opinions, questionable metaphors, and genuine love for the craft they're also mourning for. If you want to understand AI deeply, use it without embarrassing yourself, and laugh at the absurdity of it all, pull up a chair.

Information

You Might Also Like