Industry and Market Signal

Review of Meta's Muse (consumer personal AI agent): permission-asking UX patterns, an activity feed showing every tool call for transparency, and vocabulary designed for non-technical users (feed/idea…

Cross-post promo pointing to an episode of 'The Skip' about what changed at Meta after a year of going all-in on AI. The email itself contains no substantive content beyond the teaser; the actual conv…

Casey Muratori (Handmade Hero) explicitly avoids AI for his own game development ('I want to program things because I want to program them') and draws a parallel between AI-driven code volume today an…

Deep dive into Microsoft's plan for new agentic OS primitives, a Linux-on-Windows push, local model/GPU support, and reversing developer-hostile decisions to win back developers after being dubbed 'Mi…

A small fast voice model handles conversation while a frontier model (GPT-5.5) is delegated complex reasoning/tool calls asynchronously so conversation never stalls. Evaluation for full-duplex systems…

Roman Ugarte on building Grok Bot: cloud-first execution and per-user cloud computers as the two early product bets, 'unshipping' internal reasoning/memory traces before public launch, building a sepa…

Claire Vo's blind evaluations rank Grok 4.6 alongside GPT-5.6 Sol, ahead of Sonnet 5 and Opus 5 on raw capability, though Sonnet 5 still wins for enjoyable/collaborative conversation. Grok Bot's multi…

Claire Vo tests GPT-6 Astra across coding, computer use, hardware hacking, and 3D game dev — finishes a stuck 6-month feature in one session, does 90 minutes of autonomous QA testing, and reverse-engi…

a16z GP Anish Acharya predicts loops (a bug report becomes a fix/release in minutes) spreading beyond coding into growth, sales, legal, and support; argues moats are discovered, not designed upfront (…

DHH (Ruby on Rails creator) announced 37signals has stopped writing code by hand, moving backend services to Rust because agents write it well, declaring himself 'retired' as a programmer. AWS Disting…

Compile/test/lint workloads from agent tool-use are CPU-heavy (not GPU-heavy), shifting data-center CPU:GPU ratios from 1:8 toward 1:1. Spotify's Claude Code plugin 'shunt' routes cheap I/O work (read…

Mixing small/medium/large models by request difficulty can cut average cost to ~11% of an all-frontier-model baseline. Covers classifier-based routing, cascading, semantic routing, and learned routing…

Sulman Choudhry (Head of Eng, ChatGPT/OpenAI) distinguishes the 'inner loop' (writing/shipping code faster — most teams already do this) from the 'outer loop' (getting real user feedback and iterating…

Eric Sammer (Distinguished Engineer, Redis) defines 'AI-native' as building as if AI already existed from day one; Redis works in 3-4 person pods with a single feature lead. Key insight: the bottlenec…

Argues via Conway's Law that AI amplifies whatever team culture already exists rather than fixing a broken one; executive statements like 'we don't need as many people now that we have AI' destroy psy…

Multiple companies report large AI cost savings moving off frontier models: Uber cut cost-per-request 34% via open-weight models, optimized model selection, cheaper subagents, and prompt caching; Pint…

Reuters-sourced deep dive on Meta's canceled 'Project Organization Transformation' (shrinking 10-20 person teams to 3-5 person 'AI-native' pods). Gergely lists concrete downsides of very small teams (…

Consensus: flat/no-PM structures (Cursor, PostHog, Anthropic, Linear) work because the builder IS the user, and doesn't generalize to B2B2C products. Common pattern for shifting ownership: give engine…

Community consensus: pick exemplars/AI champions to teach peers rather than top-down mandates; weekly 'what I did with AI' demo slots where simple demos land better than flashy ones; remove usage caps…

Deep RAG/retrieval explainer covering chunking tradeoffs, embedding similarity metrics, flat vs. IVF vs. HNSW indexes, metadata filtering, document versioning, and hybrid search plus reranking.

Explainer on LLM memory architecture — trained weights vs. context-window working memory vs. persistent application memory — and memory-extension techniques (sliding window, summarization, structured …

Daniel Blum (PM at Melio) describes a context system that updates its own files on a recurring cadence, a weekly skill that learns from the implicit edits between Claude's drafts and his final sent ve…

Tibo Sottiaux (original Codex engineer, now heads OpenAI's Core Products & Platform) explains Codex CLI is written in Rust (not Python/TS) for millions of concurrent cloud instances, why open-sourcing…

Tara Seshan (OpenAI product lead for Codex/ChatGPT Work) frames the shift as agents taking over execution while humans shift to opinionated direction-setting. Key distinction: coding is verifiable (te…

Warp CEO Zach Lloyd describes 'Warp Factories': Slack request -> triage -> Linear ticket -> implementation -> PR -> computer-use QA -> merge, producing a first PR in ~35 minutes. Uses 'human interacti…

On-site deep dive with 7 OpenAI engineering leaders: non-engineering orgs went from ~0% to 90% Codex usage in 4 months; the full pipeline runs human-defines-outcome -> Codex gathers context (docs/Slac…

Sharadh Krishnamurthy (Stripe) explains that Kai's hardest problem was governance, not the model: 'projects' configure default models/tool access/human-review requirements per team, ~2,000 reusable sk…

Ramp's Inspect (built on open-source OpenCode, not a custom harness) runs on remote sandboxed dev environments with full access to the same internal tools/data a Ramp engineer has (Slack, Linear, Laun…

Reviews the newly released Claude Opus 5 model and covers browser-use capability in Codex, plus a Cursor-on-a-Raspberry-Pi experiment.

Turbopuffer's cofounder Simon Eskildsen on longer founder/engineer tenure, using first-principles thinking to build durable software, and caution around raising VC money.

NVIDIA's VP of Applied Deep Learning Research explains how NVIDIA builds its open models and the reasoning behind their architecture and open-sourcing strategy.

Notes that Chinese open-weight models have reached parity with closed frontier models from Anthropic and OpenAI, alongside other industry notes (AWS billing error, Spotify podcast reliability).

Compares how major dev platforms are repositioning now that AI has solved much of the 'writing code' part of software development.

Roblox's SVP of Engineering discusses the world model Roblox uses to make its multiplayer games photorealistic and the broader bet behind it.

Reports a trend of CTOs, VPEs, and Heads of Engineering walking away from high-status roles, largely tied to AI-driven change and 'founder mode' pressure, with tactical advice for leaders navigating t…

Covers Meta offering $1M+ retainer equity grants to staff who are leaving — largely ineffective — plus whether Grok Bot represents an 'OpenClaw moment' for managed AI agents.

Cursor's Head of Talent on 'funnel of doom' hiring, treating every hire like an executive search, and why caring is the biggest unfair advantage in recruiting.

Covers what it takes to be a great Forward Deployed Engineer at OpenAI, based on a conversation with OpenAI's Global Head of the role.

Cross-posted panel discussion between three CPOs on what comes after 'everyone in the org can build' becomes normal.

Ian Silber, OpenAI's Head of Product Design, discusses why AI makes this a strong era for designers, where humans still beat AI in design work, and how to manage the overwhelm.

Explores what's next for Stripe's product management practice after the 'PMs should build/prototype themselves' era.

Compares three different architectural approaches three food-delivery companies took to integrating LLMs into product search, and the reasoning behind each choice.

Elizabeth Stone, Netflix's Chief Product & Technology Officer, discusses managing AI-era product teams, the rise of systems thinking as a core skill, and how to build organizational excellence as an o…

Interview with Addy Osmani on 14 years at Google and how AI agents are reshaping software engineering, developer workflows, and the skills engineers need going forward.

Covers how high-performing engineers and engineering leaders use AI to multiply their individual impact.

Compares AI-assisted migration case studies (Asana's Enzyme removal in 2 weeks for $12K vs. an estimated $6M/5-years by hand; Airbnb's 3,500-test migration in 6 weeks; Uber's 600K-test JUnit migration…

Gergely's on-site deep dive inside Anthropic: the 6-month Claude Managed Agents project (still needed full PRDs and cross-team planning despite AI), Jarred Sumner's 11-day/$165K Bun Zig-to-Rust rewrit…

Q&A episode where Gergely Orosz answers listener questions on AI, engineering practice, hiring, and careers. Podcast format, content varies based on questions selected.

Nikhyl Singhal (former Google, Facebook exec) notes a new 2026 market dynamic: for the first time, builder-executives who can both lead and build with AI are commanding compensation packages comparabl…

Gregor addresses the perennial question of how technical an engineering manager needs to be, likely updated for the AI era where the technical landscape is shifting rapidly.

Gregor interviews Venkat Venkataramani (VP Applied Infrastructure, OpenAI) and Emma Tang (Data Infrastructure lead) on how OpenAI organizes its infrastructure teams at their scale.

Lenny's podcast episode with Noam Segal covers the same annual sentiment survey as the July 7 written piece but in interview format, with deeper discussion of specific data points. Lower signal since …

Third installment of Pragmatic Engineer's deep series on the 2026 tech hiring market, based on 50+ hiring manager and job seeker interviews. Highlights: AI-related positions are the hottest market, en…

Gregor argues that the nature of software engineering has shifted over the past two years: individual contributors now routinely do work that used to be classified as engineering leadership — prioriti…

Lenny's second annual tech worker sentiment survey reveals a bifurcated workforce: half is thriving with AI, half is struggling, and burnout just hit a record high. The 'splitting in two' framing sugg…

Pragmatic Engineer Pulse shares direct impressions from visiting OpenAI, Anthropic, and Cursor — the three most important players in AI coding tools. Key signals: cloud-based agents are the dominant e…

Instagram head Adam Mosseri shares how product teams are structured at Instagram in 2026, the emergence of the 'product staff' role, and why AI is creating space for authenticity rather than threateni…

The snippet signals this piece is exceptional: 'Most we do AI here now stories are feeling pretty repetitive. Not this one.' Meta is genuinely restructuring how product management works with AI — like…

Lenny shares a framework for how product managers can get more out of AI in their day-to-day work — not general tips, but a structured approach to increasing PM leverage specifically. Neo starred this…

How I AI reviews Claude Sonnet 5 and covers a workflow for running autonomous coding agents from mobile. Preview only — the Sonnet 5 review could have team-relevant insights on what changed with this …

Andrew Ambrosino (OpenAI Codex lead) argues that AI is making software cheaper to build, which means the scarce resource shifts to taste, judgment, and product decisions — not implementation. He discu…

A starred How I AI episode covers how Gusto (a payroll/HR company) used Claude Code to build a completely new product line — not just improve existing code, but create a new offering end-to-end. The e…

The Pragmatic Engineer Pulse reports that the Bun team completed a major Rust rewrite in 11 days using AI, at a cost of $165K in LLM tokens — a rewrite that would have taken a small team a year to com…

Dex Horthy (founder of Humanloop, a leading eval/prompt management platform) explains context engineering — the discipline of structuring what context AI sees so it can produce better code without sac…

Pragmatic Engineer deep-dive surveying how much engineering has changed in the last six months — how various tech companies are adapting their workflows, team structures, and processes. Preview only (…

Antigravity 2.0 launched with bugs, poor UX, breaking existing workflows. Analysis traces this to Google's 'promotion-driven development' culture — internal incentives favor launching new products ove…

Tony Fadell argues the primary AI risk is 'cognitive surrender' — people outsourcing judgment, taste, and understanding to AI without realizing it. This erodes individual and team capabilities over ti…

Two-part series. Part 1: AI engineering job openings up 60% at top companies in past year; software engineering up 7%. Part 2: AI labs now more attractive than Big Tech for talent — Anthropic accounts…

US government issued an export control order banning Fable from access by any foreign nationals. Triggered by an Amazon CEO-raised security concern. Independent security researcher called the ban 'hea…

Fable launched with two major policy problems: (1) mandatory 30-day data retention eliminating zero-data-retention for enterprise — AWS Bedrock cannot offer Fable within AWS security boundaries; Micro…

Multiple first-person accounts of AI token spend becoming a real constraint. Uber CTO: 'our AI budget for 2026 was blown through by mid-March'; difficulty drawing a straight line from AI spend to new …