Frontend & Web

Browser-side engineering: React, Vue and TypeScript, build tooling and bundle size, CSS and design systems, accessibility and rendering performance. Mostly teams explaining a framework migration, a design-system rollout, or how they got a slow page fast.

Loading
Introducing eve 1.0
Vercel announces eve 1.0, an open-source framework for production AI agents, and details the engineering behind it. Median hosted turn latency fell from 3.04s to 414ms through a single-orchestrator workflow design and WebSocket event writes, and estimated workflow cost for a 50-turn session dropped about 76% by storing only changed conversation data. The post also covers OpenTelemetry-based tracing, multi-tenant agent configuration, and dynamic tool loading.
Single-second static deployments now support more assets
Vercel's Instant deployments now accept static sites that include CSS, images, fonts, and other assets, not just HTML and Markdown. Eligible deployments skip the build step and go live in about a second via the CLI or REST API, subject to limits of 100 files and 25 MB total. Deployments with JavaScript, framework presets, custom build settings, or Git triggers still build as before.
After Binance, check what your signing systems can approve
Trail of Bits analyzes the October 2026 Binance incident, in which an attacker abused an EIP-7702 delegation to set a standing authority that enabled repeated transfers without fresh signatures. The post gives a four-step checklist covering restricting signing access, treating delegation as an administrative privilege, limiting damage if signing is compromised, and detecting wallet authority changes, each with verification tests.
Kitesurf is now open source
Cloudflare has open sourced Kitesurf, its experimental stateless browser engine for AI agents that runs on Cloudflare Workers, under the Apache 2.0 license. The post outlines the engine's components, including Blitz, Stylo, and Boa, and explains how to run it locally or deploy it to a Workers account. Developers can inspect, modify, and contribute to the codebase.
How AI-assisted development helped us build the new Dropbox experience
Dropbox engineers describe how AI-assisted development helped them ship its new desktop and web experience on a fixed launch timeline. The post covers two desktop integration approaches (an Electron embedded web view versus importing individual React components), shared cross-platform code with adapters, and a coding-agent workflow with written plans and human review. It also reports on accessibility fixes and a test-suite cleanup split into 53 tasks.
Don’t Build the Reusable Component Yet
The author describes a frontend case where merging three similar item workflows into one component produced a growing set of mode and variant flags, coupling unrelated decisions and slowing changes. The post proposes a practical test for reuse: abstract only when copies share the same reason for change, such as shared date formatting, design tokens, accessibility behavior, or standards-defined shapes. It also outlines how to undo a premature abstraction by inlining and re-splitting the code.
How we’re training a Figma-fluent editing model
Figma's engineering team describes training a first-party design editing model, starting with supervised fine-tuning on synthetic edit data and teacher-model validation, then moving toward reinforcement learning to improve generalization. The post also covers how they evaluate design edits across separate structural and visual dimensions, and reports speed, undo-rate, and preference results from the Figma agent's rollout.
Improving site performance by shipping more CSS
GitHub details its multi-year migration from CSS-in-JS to CSS Modules across Primer and github.com. The post explains the incremental feature-flagged rollout, automated sx-prop conversions, theming changes, and measured gains including 55% faster server-side rendering and 25% faster component initialization.
How we used Agentic Pipelines to turn test coverage into a daily habit
An Atlassian commerce engineering team describes a scheduled Bitbucket Pipelines agent that selects one under-tested TypeScript or React file per run, writes focused Jest tests, and opens a single pull request for review. Coverage reportedly rose from about 70% to 82% over four weeks, and the prompt rules evolved from reviewed PRs, such as deleting mocks that a test still passes without. The post stresses keeping each change small and keeping engineers responsible for merges and prompt...
Automating Data Ingestion with a RAG Flywheel
Zalando details an LLM-powered schema-mapping pipeline that combines AWS Bedrock embeddings, pgvector similarity search, constrained tool use, and human validation. The RAG flywheel moves from zero-shot predictions to historically informed suggestions while preserving API-level safeguards, rebuildable embeddings, and graceful degradation.
Scaling AI in Legal: Building Uber’s Redlining Agent
Uber engineers describe the evolution of a legal redlining agent from RAG to feedback-driven retrieval, tone refinement, and agentic clause modification. The post covers its Word add-in architecture, hybrid rules and learning-based decisions, time-weighted feedback sampling, vector search, and measured reductions in contract review time.
Supabase Middleware: composable, reusable middleware for Web Fetch handlers
Supabase introduces @supabase/middleware, a composable, type-safe pipeline for Web Fetch handlers that lets per-request logic such as CORS, JWT verification, and scoped clients be defined once and reused across runtimes and frameworks. The post explains the shared typed context model, the defineMiddleware API, and the Postgres limitation on Cloudflare Workers caused by pg's TCP requirement. It also shows how an OAuth protected-resource entry supports MCP server authorization with RLS-scoped...
How Commerce.com and Cloudflare are getting storefronts ready for AI agents
Cloudflare and Commerce.com enabled WebMCP across nearly 100,000 storefronts by injecting a bridge script at the edge. The post explains how per-merchant opt-outs were built with Configuration Rules and custom hostname metadata, how a configurable MCP endpoint path was added, and how the script fails safely so human shoppers are unaffected.
From Tools to Teammates: How AI Empowered Our UX Team to Level Up
The trivago UX team details how AI agents and orchestrated workflows transformed copywriting, localization, research, and product design. The team reports 75% faster copy turnaround, reduced prototyping time, reusable domain systems, and lessons for designing AI-native workflows with human quality controls.
ShopGym: Realistic, reproducible sandboxes for shopping agents
ShopGym converts live storefronts into anonymized, resettable sandbox shops and generates grounded shopping tasks for agent evaluation. The post explains its exploration, staged generation, verification, and benchmarking workflow, including structural and behavioral comparisons across real and synthetic stores.
Getting started with Claude Code mods
Addy Osmani explains how Claude Code mods work as plugin-hosted hooks, covering event middleware, state that survives hot reloads, custom terminal UI, validation, and testing. The tutorial builds a context-window monitor and examines safeguards and edit-replay workflows using concrete JavaScript and TypeScript examples.
How Stripe is designing Checkout for AI agents
Stripe explains how it integrated WebMCP into Checkout to give browser-based AI agents state-aware tools instead of requiring DOM interpretation. Progressive tool disclosure and shared form logic reduced token consumption by 42%, tool calls by 38%, and checkout time by 39% in tests across six models.
What the TDOE interns have wrought, 2026 edition
James Somers details three Jane Street intern projects: matching voice trades to anonymous public prints, implementing exchange-specific primary-peg order types, and reconciling global ETF holiday calendars. The projects illustrate how engineers translate complex domain rules into tested OCaml, FIX, and internal workflow tools.