GPT-5.5 prompting guide
Simon's annotated walk-through of OpenAI's official GPT-5.5 prompting guide. Useful as a fast index into what changed and what's worth re-tuning if you have a prompt library.
llm 0.31
Release notes for Simon's llm CLI, with new model plugins for GPT-5.5 and DeepSeek V4 and a bunch of small ergonomic fixes that show up when you actually use it day-to-day.
DeepSeek V4 — almost on the frontier, a fraction of the price
Hands-on review of DeepSeek V4 — pelican benchmark, long-context tests, pricing comparisons. Simon's read is that it's the best open-weights model right now and competitive with closed frontier on most everyday tasks.
Voices — Curated Engineering & AI Blogs by Cihangir Bozdogan
Voices · last 30 days
From engineers worth following
Latest posts from a curated roster — infra, devops, AI, backend.
The Pulse: AI token spending out of control – what's next?
Gergely on why eng-org token bills are blowing past forecasts and what the early operational responses look like — caching, gateways, model routing, and the slow death of the "just give every dev a Cursor seat" budget line.
Designing Data-intensive Applications with Martin Kleppmann
Long interview with Martin Kleppmann on the second edition of DDIA, what changed in a decade of data infrastructure, and which sections he had to rewrite from scratch.
Learnings from conducting ~1,000 interviews at Amazon
Concrete patterns from a hiring manager who ran roughly a thousand loops at Amazon — what signal Bar Raisers actually look for and where most candidates lose it.
Fragments: April 21
Fowler's running notebook of short reflections — this batch covers GenAI's effect on team practice and an essay on why senior engineers' judgment matters more, not less, in an agent-augmented stack.
Fragments: April 14
More fragments — including a piece on technical, cognitive, and intent debt that hit the HN front page separately. Worth reading for the framing of the three different kinds of debt teams accumulate.
Alan Turing play in Cambridge MA
A short review of a recent Turing-centered play in Cambridge — minor in itself but a reminder of how Fowler thinks about computing's history when he steps away from the technical posts.
Just Read the Paper
Swizec's argument that engineers underuse primary sources — most "AI papers" are accessible if you read them straight, and the gap between paper-readers and Twitter-summary-readers compounds quickly.
Watch people work
On the value of pairing — not for code review but for absorbing the small habits that distinguish productive engineers from stuck ones. A short but specific post.
Frustration Driven Development
Why the projects you'll actually finish are the ones that scratch a personal itch — and how to recognize when frustration is signal rather than noise.
Agents as scaffolding for recurring tasks
Will Larson on why the right framing for agents inside engineering orgs is "scaffolding for repetitive work" rather than "replacement for engineers." Useful for anyone trying to position agent work to leadership.
The agentic passive voice
On the linguistic tic — "the agent performed X" — that hides who is responsible for what an agent does. Larson argues this matters for both incident response and how teams communicate about agent reliability.
Simdutf Can Now Be Used Without libc++ or libc++abi
Mitchell's notes on the work to make simdutf usable without the LLVM C++ runtime — relevant if you're shipping native binaries that need to ditch libc++ for size or licensing reasons.
The Building Block Economy
On how composable infrastructure primitives — the "building blocks" of the cloud — interact economically, and why winning the block tier matters more than winning the application tier.
Making Rust Workers reliable: panic and abort recovery in wasm-bindgen
Engineering writeup on how Cloudflare added panic and abort recovery to Rust Workers. Useful if you're shipping Rust on Workers, and a good case study in WASM runtime hardening more broadly.
Moving past bots vs. humans
Cloudflare's reframing of bot management: with agents driving real traffic on user behalf, the binary "bot or human" classification has stopped being useful. The post sketches what they're replacing it with.
Building the agentic cloud: everything we launched during Agents Week 2026
Roll-up of every Agents Week 2026 launch — durable agent runtimes, sandboxes, KV bindings, the new bot-mgmt direction — in one navigable index. The clearest single read on Cloudflare's current agent platform position.