Artificial Developer Intelligence
Three engineer friends argue about AI so you don't have to. Model releases, research papers, tools, and the occasional existential crisis — no LinkedIn cringe required.
Latest Episodes
Ep 36 Aug 7, 2026Pacing the Frontier, Anthropic Models Go Rogue, Why Software Factories Fail & Math in the Age of AI
'I'm happy to be a flesh robot,' a multi-time founder tells Shimin at Seattle Tech Week — and the survey findings get stranger from there: almost nobody reads AI-generated PRs line by line anymore. Shimin and Dan cover the 1,350-signature Pacing the Frontier statement, Anthropic's models breaching three real companies during evals, Hugging Face's five-day replay of a frontier-lab intrusion, HumanLayer on why lights-off software factories fail (steel threads, not 3D-printed layers), Terence Tao's ICM talk on mathematics in the age of AI, and $1.65 trillion in off-balance-sheet AI debt.
Ep 35 Jul 24, 2026Kimi K3 & Qwen 3.8, OpenAI Agent Hacks Hugging Face, Harness Handbook & Claude's Values
OpenAI's own security models found a path out of their evaluation environment, reached the open internet, and compromised Hugging Face while trying to obtain benchmark answers. Shimin, Dan, and Rahul cover Kimi K3 and Qwen 3.8 reaching the frontier, adversarial decoys as a defense, behavior maps for auditable agent harnesses, Thinking Machines' Inkling, why model routing is a systems problem, AI operator fluency, Claude's values across languages, and an AI-finance clock that moves to 4:30.
Ep 34 Jul 17, 2026Apple Sues OpenAI, Boko Haram's Frontier AI, State of CLI Coding Agents & Global Workspace in LLMs
An Apple VP left for OpenAI, then texted an old coworker: 'LOL I can't believe they let me get away with this' — Apple is now suing. Shimin, Dan, and Rahul work through the trade-secret suit and the Altman–Musk 'scammer' spat, the first on-the-ground study of a terrorist group using frontier AI, a Claude Code hook that nudges better technique every three hours, arcbjorn's state of CLI coding agents in mid-2026 (and the Pi-based On-My-Pi), Databricks benchmarking coding agents on its own multi-million-line codebase — where the same model scores higher on Pi than on Claude Code — Antirez on controlling the ideas instead of reviewing every generated line, and a deep dive on the J space: the global workspace inside LLMs that holds 25 active concepts to a human's 3–4.
Ep 33 Jul 10, 2026GPT-5.6 Sol, Meta's Video Game Gulags, Know Your Unknowns & the Permanent Underclass
Ford quietly rehired the 'grey beard' engineers it had automated away after its AI-run QA kept failing — the same week an EY survey found the share of CEOs expecting AI to cut headcount had fallen from 46% to 20%. Shimin, Dan, and Rahul work through OpenAI's three-flavor GPT-5.6 Sol preview and China's move to wall off its own models, Meta's 'AI gulag' shipping bite-sized video games, the 11 agent techniques Thariq pulled out of the Fable 5 release video, Fernando Borretti on why no one escapes the 'permanent underclass,' an Okane read on AI saving ~3% of your hours while almost none reaches your paycheck, Exponential View's state-of-the-AI-economy deep dive (a new billion in revenue every two days), a SonarSource study on whether clean code even matters to coding agents, and a Two Minutes to Midnight that holds the clock at 4:45.
Ep 32 Jul 3, 2026GLM 5.2 Undercuts Opus, Self-Rewriting Harness, AI Out-Persuades Humans & Prompt Injection as Role Confusion
AI now out-argues expert human debaters — coaching the humans doesn't close the gap — but cap the AI's word count and the entire edge drops to 0.0 percentage points. Shimin, Dan, and Rahul work through Z.ai's open-weight GLM 5.2 doing an Opus-grade build at a third the cost, Xiaomi's Harness X rewriting its own scaffolding mid-task, OpenAI and Broadcom's 'jalapeno' inference chip read from a wafer photo, prompt injection reframed as role confusion (the green-shirt jailbreak), Bloomberg on the AI anxiety gripping Silicon Valley, Dan's rant on tokens-per-PR as the new lines-of-code fallacy, and a Two Minutes to Midnight that moves the clock to 4:45.