TLDR Dev 2026-11-17
Smoother CSS corners 🔲, smaller Valkey indexes 🌳, your own AI crew 🤖
Together With
Aperture by Tailscale: Build and scale AI agents faster, with security already built in (Sponsor)
Building agents should be fun. Setting up skills, MCP servers, and system access every time? That's just…work.
Aperture by Tailscale puts models, agents, and MCP servers behind one private gateway in your Tailscale network. Point Claude Code, Codex, or Gemini CLI at Aperture and start building, no API keys to hand out. Set up skills and MCP servers once, then reuse them across agents and devices.
Scaling from one agent to 50? Aperture logs every request, ties it to an identity, and checks it against your policies and budgets.Try Aperture by Tailscale for free
Squircles (6 minute read)
Squircles that look similar can have very different curvature, and Figma's approximation contains discontinuities where its curves join. This article contains an interactive comparison that explains the tradeoffs between Apple, Figma, clothoid, and superellipse shapes, along with the CSS for approximating Apple's corners using corner-shape.
Staggering Scroll-Driven Animations with animation-range-start (7 minute read)
Staggered scroll animations can use animation-range-start and sibling-index() instead of a custom progress formula. This post contains a worked example that replaces the complicated stagger calculation with one CSS declaration while retaining the browser's built-in easing behavior.
From Skiplists to B+ Trees: Making Valkey Sorted Sets More Memory Efficient (7 minute read)
Valkey 9.2 replaces large sorted sets' skiplist index with a B+ tree that packs entries together and uses SIMD-assisted searches. In a five-million-member test, per-member index overhead fell by 36% to 43%, while command behavior stayed the same. The release is still in candidate testing.
What Does It Mean to Understand What We Built? (2 minute read)
Understanding individual lines of code does not necessarily explain a system's behavior, especially emergent failures. AI could help engineers focus on system properties and develop better tools for understanding what they actually built.
From Tasks to Goals (6 minute read)
Giving an agent an ongoing goal means delegating decisions about which tasks to do and when to act. That requires tool boundaries, feedback, persistent memory, and ways to ask questions, plus stopping rules that let the agent do nothing when no action is useful.
How Rust Changed Me as a Developer (10 minute read)
Rust's enums make variants explicit and let the compiler enforce that every case is handled. This essay explores how sum types and static dispatch change abstraction choices, including when a straightforward enum is more useful than designing an interface.
Your AI software factory shouldn't stop at deploy (Sponsor)
AI software factories can fall short without automated release controls for AI-led development. This guide explores:
- What most software factories are missing.
- What LaunchDarkly engineers learned from building our own.
- Why automated release controls are the key to faster, safer delivery of agent-written code.
Get the guide
dusk bot (Website)
dusk bot runs a crew of AI helpers, each with its own chat, computer, and memory, using models or subscriptions you already have. Bots can work in parallel, hand off tasks in group chats, and run scheduled routines locally, on your own server, or through dusk's hosted cloud.
BrowserPod 4.0 (9 minute read)
BrowserPod runs a Linux-like WebAssembly microVM on the user's device, with tools including Node.js, git, bash, and Rust programs. Version 4.0 makes the CDN-loaded runtime free for commercial and production use without an account, while unrestricted networking is a separate service and self-hosting the runtime requires an enterprise license.
Jolt (Website)
Jolt is a self-hosted Clojure implementation that compiles through Scheme and can produce a standalone native binary with statically linked C dependencies. It supports persistent collections, native C interop, and OS-thread concurrency, but does not provide general JVM interop.
Become a founding member of TLDR’s AI community (2 minute form)
TLDR is launching a curated community for AI builders to discuss major news and launches. The best links and comments will be featured in TLDR AI.
Apply here to join.
BenchProofer: Formally Verified Coding Benchmarks, on Demand (11 minute read)
BenchProofer adds formal specifications, verified reference implementations, and audits to coding benchmarks, with SWE-Proof covering all 500 SWE-Bench Verified tasks. Its audit flagged behavioral differences in 134 test-passing Opus 4.8 patches, but human review found actual bugs in only 21 of 41 sampled flags. The distinction matters because a different implementation can still be a valid fix.
Golden Board: Teaching Chess Without a Shared Language (25 minute read)
Golden Board packs a visual introduction to chess into a 512 KiB binary message, starting with a bootstrap that teaches the recipient how to decode it. The message includes lessons, exercises, and complete games, while its LLM trials explore ambiguities rather than proving an unknown intelligence would understand it.
Build, evaluate, deploy, and improve AI agents on Chalk. (Sponsor)
Stop wiring together databases, sandboxes, tracing, and evals by hand. Chalk brings the entire agent development loop into one project, with no infrastructure to provision.
Try Chalk now.React Flow 13 (5 minute read)
React Flow 13 splits its internal store into specialized subscriptions to reduce unnecessary renders in large node-based interfaces.
How to Make (Almost) Anything (Website)
MIT's 2026 fabrication course links lessons and recitations on CAD, electronics, embedded programming, machining, networking, and system integration.
Turso Sync: Local-First SQLite with Push and Pull (9 minute read)
Turso Sync keeps writes local and replicates logical row changes through separate push and pull calls.
Coding Agent, Meet ngrok: The ngrok MCP Server Is Live (6 minute read)
ngrok's MCP server lets coding agents inspect traffic, configure connectivity, and validate Traffic Policy rules before applying them.
The most important software engineering news in one daily email
Join 480,000 readers for
one daily email