TLDR Dev 2026-10-09
ChatGPT's intelligent UI š§©, cost of open source š², software judgment š§
Domains, hosting, and email. Built into your workflow. (Sponsor)
The GoDaddy Developer Platform connects hosting, domains, and professional email in one programmable workflow, accessible from your code, scripts, or AI agents.Deploy Node.js apps through APIs or the gddy CLI. Then use the APIs to register a domain, configure DNS, and set up professional email.One platform identity, consistent API conventions, and shared documentation connect each step. Access developer pricing on eligible domain registrations.ā
Explore the APIsā
See domain pricingš§āš»
Articles & Tutorials
We built our own cloud agents runtime. Here's what we learned (11 minute read)
PostHog describes the shared runtime behind its Desktop, Slack, self-driving, and web agents, built around Temporal workflows, VM sandboxes, snapshots, fresh credentials, and resumable run logs. The hard-won lessons cover queued follow-ups, deterministic network controls, Docker-capable custom images, and designing every run to survive the loss of its sandbox.
How ChatGPT Intelligent UI works (20 minute read)
A reverse-engineering tour traces ChatGPT's interactive responses from a model-written DIL format through server-side compilation, sandboxed client execution, and native rendering. It also explains how streaming recompiles partial output, preserves keyed state, and sends server-resolved data such as images separately from model text.
Chrome Decisions API: the prompts, limits, engines and code reviews behind DecisionModel (10 minute read)
An investigation of Chrome's experimental Decisions API compares the explainer with implementation code and review comments. It surfaces slow Gemma 4 timings, position bias, context limits, probability semantics, and differences between Chrome's prototype and MediaPipe's DecisionMaker.
When code is cheap, judgement becomes the job (11 minute read)
As AI makes code production cheap, the scarce work shifts to choosing priorities, reviewing product behavior, maintaining architecture, and owning outcomes in production. Senior engineers create leverage by turning repeated review feedback into guidance and deterministic lint rules while keeping human judgment for business context.
Don't merge what you didn't run: a test gate for agent-written code (10 minute read)
A practical test gate checks agent-written changes on a clean machine before review by reinstalling from scratch, running the full suite three times, rejecting removed or skipped tests, and restoring the base branch's tests against the new code. Running the same gate inside the agent loop turns failures into immediate repair instructions while an independent CI run keeps the verdict outside the agent's control.
Treg (GitHub Repo)
Treg acts as an OpenRouter for agent tools, giving agents one base URL and token for a catalog of metered provider endpoints plus team-owned APIs, CLIs, and skills. It can inject credentials server-side, expose tools through MCP and agent skills, and let teams share capabilities without placing provider keys in callers.
Plannotator (GitHub Repo)
Plannotator provides a local browser-based review surface for plans, Markdown, HTML artifacts, diffs, and pull requests produced by coding agents. Hooks and commands route annotations back to agents, with support for multiple agent harnesses and local-first storage by default.
A Single POST Freezes Any Next.js Server (3 minute read)
A React Server Components flaw let an attacker force quadratic work while React rebuilt FormData for a server action, allowing a roughly 900 KB request to trigger 100 million string checks and block a Node.js event loop. The issue affects vulnerable React 19.0 to 19.2 releases and is fixed in 19.0.6, 19.1.7, and 19.2.6.
Your skills need an evaluation mechanism (11 minute read)
Expo built a CI evaluation harness that measures whether coding agents discover relevant skills, follow their guidance, and produce working apps. Adding an overview skill raised sessions loading at least one downstream Expo skill from 9% to 55%, while catalog crowding tests showed why names and opening descriptions must survive truncation.
The Slow Formation of Durable Software (16 minute read)
The origin story of Zotero shows how years of conversation, prototypes, and collaboration helped a team of historians discover what durable research software should become. Its path from Web Scrapbook and Scribe to a browser-aware tool used by millions argues that clear product vision can require slow communal formation, even when implementation itself becomes cheap.
Sentry + OTel: You can add connected context to your OTel signals. Here's how (Sponsor)
You can send your OTLP-compatible logs and traces directly to Sentry, with no extra dependencies. This workshop shows you how to use them for performance monitoring and debugging.
Save your spotcalc-input, a custom input element that accepts mathematical formulas (4 minute read)
calc-input lets forms accept arithmetic expressions, shows formulas on focus and results on blur, and uses a safe parser plus native form validation instead of eval().
How much does it cost to use open source software? (3 minute read)
A Google study found most internal open source package updates take four hours or less, while local patches, dependencies, and upstream contributor counts better predicted difficult updates than package age or internal adoption.
video-shotcraft (GitHub Repo)
video-shotcraft turns coding agents into a Remotion-based motion studio with 157 shot recipes, 214 motion previews, editable templates, sound design guidance, and a browser workbench.
The Bug Only My Screen Could See (10 minute read)
A display-only flicker that browser captures could not reproduce was traced to full-width tile repaints, then reduced from 267 page-tile repaints to 9 by promoting only the changing timeline elements to their own composited layers.
The most important software engineering news in one daily email
Join 470,000 readers for
one daily email