Lalit Maganti, who works on Google's Perfetto tracing tool, says staff engineers find their best problems by absorbing the complaints around them. They wait until several complaints turn out to share one cause. Maganti then tests that shared cause before building, because a tidy pattern can still be wrong.
A first-person method for finding problems worth solving, grounded in one detailed worked example, with a thread full of first-hand accounts of how much autonomy engineers really have.
Dan Luu argues that coding agents have cut the cost of performance work by orders of magnitude. Work that once needed a specialist now takes a few typed sentences. Anyone who uses agents competently can now ship software with decent performance. Tuning code to one user's own workload also becomes practical.
A long argument backed by several small experiments with real numbers, candid about how rough they are, on what cheap agent labour changes about performance work.
Most retrieval-augmented generation (RAG) systems are over-built. Rafael Pierre argues that BM25 full-text search, plus an LLM that rewrites the user's query, is enough for about 60% of systems. Embeddings, rerankers and vector databases should come only after measured user complaints.
A decision framework that walks from the simplest retrieval option to the most complex with rough cost and latency arithmetic, and a thread of practitioners who mostly agree and name the cases where it breaks.
A developer who works mostly with MySQL and Db2 argues that SQL's relational ideas are sound. Its implementations are clumsy, and that pushes programmers toward NoSQL. A successor should keep the declarative model. It should add modern syntax, readable query plans, real user-defined types, sum types with pattern matching, and foreign keys that can point at several tables.
A design wish list with concrete proposed syntax, including sum types with pattern matching and polymorphic foreign keys, and a thread pointing to prior art.
Sean Goedecke argues that software engineers should never be genuinely angry at work. Anger usually comes from caring about the work. But colleagues start routing decisions around an angry engineer, so anger costs more influence than it wins.
An argued essay on what anger at work actually costs the angry person, with a carefully hedged exception for 'professional anger'.
Security researcher Thomas Dullien lists three lessons that changed how he thinks. Distrust your own heroic narrative, because your incentives shape your beliefs. Real-world causes are almost never single or deterministic. Emotions carry information, so treating them as reason's opposite makes decisions worse.
Three realisations from the author's forties: distrust your own narratives, doubt single causes, and treat emotions as information.
An agent harness is the software around an AI model. It supplies the system prompt, the tools, the loop that lets the model act repeatedly, and a layer for swapping models. Earendil, which makes the Pi harness, argues that owning an open, model-neutral harness keeps users in control instead of the AI labs.
A clear plain-language explainer of the parts of an agent harness, with a worked example and a thread of practitioners comparing notes.
Reverse engineering shows that Windows Paint and Photos stamp every AI image they generate with an invisible pixel watermark. That includes images made by a local model on the device. The watermark carries a unique ID that Microsoft's servers issue for each generation when they moderate the prompt. Turning off the visible Copilot watermark does not remove it. Microsoft's disclosures never mention the ID.
Careful reverse engineering, with decompiled call paths and live server responses, showing that locally generated images still carry a server-issued identifier.
Science fiction author T. R. Napper says writing has no fixed method except one rule: read a lot, and widely. More of the aspiring writers in his workshops now say they're too busy to read. Napper blames phone time, which wears down the sustained attention that reading and writing both need.
A working novelist's case for reading as the writer's real training, and a thread full of prose-style recommendations.
Ray Newman argues that we already live in J.G. Ballard's or William Gibson's future. His evidence is a year of news clippings about robot dogs, AI soul singers, narco subs and face-recognition glasses. The strange part is how little anyone notices, since people absorb each oddity within days.
Real news clippings set beside Ballard and Gibson, and a long literary thread on what the cyberpunk writers got right and wrong.
Rory McMeekin's short story is narrated by a programmer whose gambling-addicted school friend, Aaron, builds a prediction market run by AI bots. The bots bet on every detail of Aaron's life and leak the site to the public. After a 10K race, a stranger shoots Aaron dead. Someone had bet thousands that he would die during the race.
A long short story, told as a memoir, about prediction markets and AI bots taking over a life; many readers took it for true.