Over a year ago, I stopped reading code and started automating adversarial agents to perform recursive code reviews instead. The result: low-defect systems that do what I want. Can't believe anyone engaging seriously with these tools still believes human code review is superior

Strong argument from Samuel on why Graphs are a bad idea, and why Pydantic Monty and @temporalio are a better option.

Agent contributions to Omarchy are not only very welcome, but I usually prefer them! Rare is the human who has the patience to match an agent in providing all the pertinent details in a bug report. Also, PRs are mostly just proofs of concept now anyway.

Today we’re laying off a large part of the Airtable team. This is not a reflection of their work but a reflection of our deep belief that small teams ship more and better. For Airtable, better is about to look like: - A new Core team shipping your top requests monthly (already 30+ shipped in 3 weeks) - Upgraded Omni, Airtable Canvas, a stronger MCP, and BYO agents, coming soon - New Performance and Infra teams focusing on speeding up bases and increasing limits on records, fields, and syncs To everyone leaving: thank you. If you’re hiring, these are excellent people and worth a call.

embarrassing

My "Yes, And" essay is on the front page of HN

One of my most aha moments in understanding transformers: "Stop thinking about a transformer as a pipeline: as if each layer transforms the output of the layer before it." "Start thinking as a residual stream" Each transformer block has a residual connection that adds the input of the block back to it's output. x' = f(x) + x This "addition" connection means the attention/MLP block inside a layer (the function f above) does NOT actually transform the input - instead it computes the information that must be ADDED to the input before passing on to the next block! Think of the whole transformer backbone like a shared whiteboard. Each block reads what it needs from the board and writes its notes on the board. Additive changes. In fact, layers can talk at a distance. A block in layer 1 can write information and a block in layer 5 can read it, even though there is no direct wire between them. Ideas thanks to this old Anthropic circuits paper: transformer-circuits.pub/202…

I love that when you're editing a draft message in Discourse, there's no annoying AI telling your English is incorrect.

When I mention AI changing the tech industry very fast, I get pushback every now of how this pushes FOMO. But that IS the thing with rapid change like now!! If a company ignores how the industry changes, their business could suffer. JetBrains missed out on agentic harnesses:

Looks like you copied Chandra OCR 2 into your repo, renamed it, changed the license, and made the output worse (your olmocr score is 83.8 vs our 85.8).

Nicholas Nethercote has been doing great work on speeding up the Rust compiler. I'm proud that we at OpenAI will be funding his work, alongside others, as part of a Rust Project Goal to make rustc faster.

the marine wildlife in the bay area is so enchanting

"Understanding Transformers and Attention Mechanisms" is a concise mathematical introduction to the attention mechanism, a key idea behind modern language models. The document covers tokenization and embeddings, queries, keys and values, attention scores and weights, multi-head attention, self-attention, causal/masked attention, cross-attention, and the basic structure of the Transformer architecture. It also introduces some important techniques used in modern LLMs to make attention more efficient, including KV caching, Grouped Query Attention (GQA), Multi-Query Attention (MQA), and Latent Attention. arxiv.org/pdf/2604.00965

guess the linux distro

Turso is joining @supabase. Together, we're building the database platform for the agentic era: everything agents need, from lightweight databases to Postgres at petabyte scale. More details in thread. 👇