# ai

Cursor's Head of Design, Ryo Lu, wrote this late at night: the sweetness and pain of turning a dream into a job, machines getting good at the things that used to feel close to the heart, and why the only thing that matters is the relationship with the source.

Mitchell Hashimoto tries to define "taste": consistently making high-quality qualitative judgments where no objective metric exists. People say taste is worthless because it is easy to copy — but that proves the opposite: without someone with taste making the thing first, there is nothing to copy.

You can't tell a model 'make it premium and smooth' and get a premium UI. kvnkld's full system behind his polished components — easing curves, design tokens, real physics, the 98% press — reduces to one move: trade adjectives for numbers. The model is the hands; the last 10% of taste is still yours.

The 2026 data all points one way: AI pushes raw code output up about 4x, but real delivered value only rises about 10%. The gap in between is all review debt. Writing code got cheap; being sure it is right did not. Code review went from a side effect of engineering to its most leveraged front line.

When machines start writing code, the scarce skill is not tool fluency. It is architectural judgment: digging below abstractions, defining boundaries, writing specs, falsifying claims, and deciding where human judgment still matters.

Addy Osmani warns that default AI coding workflows help people close tasks, but do not automatically make them sharper. The difference is not whether engineers use AI; it is whether they use it to test and grow their own mental models.

Popular AI library LiteLLM was hit with a malicious backdoor — just installing it could trigger credential theft of SSH keys, cloud tokens, and crypto wallets.

Cursor announced on X that Composer 2 is now available. The official post was just one sentence and a screenshot — no changelog, no details.

A thread summarizing an Anthropic livestream interview with Adam Hooda, head of Uber's AI Foundations team. It covers how Claude Skills organically grew from 2 to 500+ inside the company — through dual-layer governance, deterministic outputs, and meta-skills that make skills that make skills.

Alexey Grigorev compiled 200+ resources from major AI labs, engineering teams, practitioners, and community discussions into Awesome AI Engineering — a curated list that connects how companies build AI, how they hire, and what skills keep showing up.

Thariq announced a new session-level feature: you can now set effort to max, letting the model reason longer and use as many tokens as needed. The catch? It burns through your usage limits faster, so you have to enable it manually each session.

DevvMandal announces the release of what they claim is the world's largest open-source computer-use recording dataset — over 10,000 hours of Salesforce, Blender, Photoshop and more, aimed at automating the next level of white-collar work.

Dan McAteer's tweet questions a common leap in logic: if AI automates software engineering, software demand keeps growing, and trained engineers are best positioned to ride the wave — how does anyone conclude that engineers are screwed?

Simon Willison published notes on Tobi Lütke's autoresearch PR — an AI-driven experiment loop that made the Liquid template language (which Tobi created for Shopify 20 years ago) 53% faster with 61% fewer memory allocations.

Deirdre Bosa points out a notable shift: the same private equity firms that pushed cloud software into portfolio companies a decade ago may now use AI as a reason to rip those SaaS tools back out.

How hard is it for a Data Engineer to become an AI Engineer? Alexey Grigorev says way easier than you think — because AI engineering is still engineering, and the hard parts are exactly what DEs already do every day.

When an $8,000 legal memo becomes a $20 subscription, GDP drops 99.7% while actual output stays the same. SemiAnalysis argues we need tools from feminist economics to measure AI's real value.

DataTalksClub founder Alexey Grigorev shared the AI Engineering Buildcamp syllabus: LLM APIs, RAG, Agentic Flows, Monitoring & Guardrails, Evaluation, and a capstone. A practical learning path for building agentic AI in production.

Everyone stares at the first 5 years of AI compute profits. But the real game is decided by what happens after the contract expires.

Claude Code shipped Auto-Memory — AI can finally take its own notes. But we've been doing this with OpenClaw for months. A hands-on comparison of two memory architectures: design philosophy, real pitfalls, and why memory is a trust problem, not just a tech one.

Chris Lattner reviewed the Claude C Compiler built by 16 parallel Claude Opus agents. His verdict: AI can assemble serious engineering systems, but it mostly recombines known knowledge. His Modular team expectations read like an AI coding strategy guide.

A Tech Lead uses his blog as a training ground, learning 12 quality metrics with an AI tutor through RPG-style Level-Up teaching while subagents implement in parallel. The real takeaway is not the metrics, but a reusable AI-assisted learning method.

Vibe Coding is refined sugar for creation, compressing months of effort into seconds. The rush is not it works, but I cannot believe it works. The author dissects dopamine, consumption disguised as creation, and the vertigo of infinite possibilities.

Anthropic's 2026 Agentic Coding Trends Report highlights multi-agent adoption, the Papercut Revolution for tech debt, self-healing code, and Claude Code hitting $1B ARR. TELUS and Rakuten case studies show developers shifting from code writer to system orchestrator.

Junde Wu built OneContext after getting fed up with coding agents forgetting between sessions. It uses filesystem, Git, and knowledge graphs to work across sessions, devices, Claude Code, and Codex; the GCC paper hits 48% on SWE-Bench-Lite.

Dr. Alex Wissner-Gross's daily tech briefing: AI agents as full-time employees in China, OpenAI banning human coding, Claude Opus 4.6 topping benchmarks, rabbit brain cryopreservation, $1 trillion chip sales, SpaceX dismantling the Moon for data centers — and a pig that actually flew

Michael Bloch asks what becomes more valuable when AI intelligence is nearly free. His 12 endgame positions include energy, atoms, capital, regulation, trust, proprietary data, attention, network effects, operations, security, physical space, and intelligence itself.

OpenAI co-founder Greg Brockman publicly reveals how OpenAI is transforming to agentic software development internally. By March 31st, agents should become the first resort for all technical tasks. Includes six concrete recommendations, including 'Say no to slop' on code quality.

Cursor's Head of Design Ryo Lu says AI coding creates a new trap — the 'illusion of speed without structure.' People who can't think clearly just generate slop at scale.

Cursor CEO claimed they built a browser from scratch using GPT-5.2, but developers discovered it was just stitching together existing open source components

Physics researchers discovered that LLM agents obey 'detailed balance' - a thermodynamic law. This isn't a bug, it's a feature.

Vercel CEO Guillermo Rauch shares their AI support system achieved an all-time high of 87.6% autonomous resolution rate — and customers love it

Private data × Untrusted content × External communication = Perfect security disaster, and it's already happening everywhere

Finally someone built a 'package manager' for AI agent skills, so agents stop flying around like headless chickens

Jaana Dogan, principal engineer on Google's Gemini team, admits Claude Code replicated her team's year-long distributed systems work in 60 minutes

From RLVR to Vibe Coding, Karpathy breaks down 6 key LLM developments in 2025

From 80% manual coding to 80% AI agents, Karpathy calls this the biggest change in his 20-year programming career

When you instinctively think 'neat feature idea, but not worth the time' and then prompt Claude Code anyway — because your 25 years of intuition no longer match reality

The minimalist agent definition (LLM + tools + loop) makes you forget what really matters: planning, memory, trust, and evals