TLDR Dev 2026-11-09
X rebuilds its web app π οΈ, Uber secures AI agents π, WebMCP in practice π
Together With
What do Cursor, OpenAI, Anthropic, SpaceX, Entire, Airbnb, Nvidia, Uber and Meta SI have in common? (Sponsor)
Buildkite, the CI orchestration platform for AI.
We provide programmable primitives you co-ordinate, running only what changes, across any cloud, any chip, your secrets, your infra. Control without compromise.
Build faster, more securely, more reliably, with the CI engineered for frontier scale, used by the teams setting the pace.
You deserve better CI. Start with our 30-day trial. Kick the tires β
π§βπ»
Articles & Tutorials
Modern CSS Might Make Your Front-End More Complex (22 minute read)
Replacing JavaScript with container queries or :has() can simplify a component while creating hidden dependencies across the application. The migration checklist covers ownership, scope, cascade, cleanup, measured performance, and maintainability so that fewer lines do not become harder behavior to debug.
Securing AI Agents: Lessons From Real-World Deployment (13 minute read)
Uber secures agents with provisioning checks, runtime enforcement, and continuous testing, adding behavioral controls to traditional authorization. Scoped intents, session context, and prompt provenance help determine whether an authorized action matches the user's request, while small specialized models and parallel checks keep latency manageable.
Rebuilding the Global Town Square from Scratch (12 minute read)
X rebuilt its web app around one server for people, search crawlers, and link previews, replacing three separate servers. On a slow connection, the first post on a profile appears in 2.7 seconds instead of more than 22 seconds, while signed-in traffic is moving over gradually by country.
Writing as a Thinking Tool (8 minute read)
Writing externalizes thought, exposes gaps in understanding, and creates room for connections that can be missed while reading or talking to an LLM. Separate unrestricted drafting from ruthless editing, use plain language, write for yourself, and keep a regular schedule.
We Already Know How to Work with Agents (5 minute read)
Companies offer a starting point for organizing agents through specialization, delegation, management layers, and escalation. Existing communication and project tools can support agent organizations, but identity, authorization, durable state, and human accountability still need to be solved.
Your Coding Agents Made You a Manager (16 minute read)
Working effectively with coding agents depends on clear context, self-verification, early course correction, and explicit boundaries. The advice is to turn repeated corrections into shared instructions, enforce only non-negotiable requirements, and keep humans responsible for product decisions, security boundaries, and what ships.
Run every model from the same endpoint (Sponsor)
Router.com (by Ramp) is the LLM gateway that cuts inference costs. Switch between closed- and open-weight models automatically to get frontier-level intelligence at a small fraction of the cost. One endpoint with automatic fallbacks, request-level observability, built in privacy controls, and ZDR on eligible models.
Get your API key Ink: State Is a Program (Website)
Ink is a typed language that compiles state, derived results, and atomic updates together, keeping declared relationships consistent without hand-written cache invalidation. The MIT-licensed working prototype targets JavaScript, native Rust, Rust/Wasm, and eligible WebGPU computation, with proof-checked transformations and explicitly constrained learning.
PI-Desktop (GitHub Repo)
PI-Desktop brings projects, coding agents, models, and workflows into a persistent desktop workspace built with Electron and a Rust host core. Plugins can add tools, panels, MCP servers, and background services, while the application supports switching between cloud and local models.
Spec Kit (GitHub Repo)
Spec Kit gives coding agents structured processes, reusable templates, and documented outcomes for building features, fixing bugs, and assessing ideas. Its extensions, presets, and workflows let teams adapt those processes to their tools and standards.
WebMCP, Explained with a To-Do App (12 minute read)
A TodoMVC example exposes seven browser tools through HTML forms and JavaScript, showing how agents can operate a page without guessing which controls to click. It covers input validation, clear tool responses, untrusted content, and confirmation hints, while explaining why tool calls do not replace UI tests.
JIT Bitmap Text Rendering for Microcontrollers (7 minute read)
A microcontroller text renderer generates machine code for each string and draws through a compact one-bit intermediate buffer before expanding to RGB565. Packed glyphs, overlapping font data, and instruction scheduling reduce memory pressure and rendering cost, including shrinking one font bitmap from 1,140 to 960 bytes.
Use NVIDIA NeMo to Turn Agent Failures into Tests and Harness Fixes (7 minute read)
NeMo Compass mines agent traces for recurring failures, and NeMo Eval Author turns them into domain-specific tests that can validate harness changes. In a Dassault Systèmes engineering example, the loop reduced malformed tool calls from 142 to zero and improved the evaluation score by 28% with 35% fewer tool calls per run.
The most important software engineering news in one daily email
Join 470,000 readers for
one daily email