Glean 拾遗
Recent picks

12picks · chronological

11-18

When Claude writes 90% of the code, verification becomes the system

Claude Code's team now has Claude write and review roughly 90% of its code. Addy Osmani's new post covers how the verification layer has to be built at that ratio, so the agent can show evidence that a change is actually correct. Two practices stand out. First, the review rules in REVIEW.md are always read from the main branch, never from the PR under review, and new submission gates run in observe mode long enough to measure false blocks before being enforced. Second, a 39-item reliability checklist where each item states how to spot it in a diff and which real incident produced it; Claude runs only the 3-10 items a given diff triggers. Useful for engineers building coding agents, CI gates, and automated code review.

x.com · 1 min · Agent Engineering · Agentic Coding · Claude Code
11-14

Your Coding Agents Made You a Manager

Addy Osmani maps engineering-management practice onto day-to-day coding-agent work: give the agent a definitive pass/fail check (tests, build exit codes, screenshots) instead of accepting "looks done"; write down only the context the repo cannot reveal — objective, constraints, out-of-scope items, success criteria; and once you have corrected the same thing twice, put it in CLAUDE.md or a shared skill. He borrows Grove's task-relevant maturity to explain Claude Code's split between requests (prompts, CLAUDE.md, skills) and requirements (hooks, deny rules, managed settings), arguing that over-process is the new micromanagement. The piece also lists five questions to answer before an unattended run, and is candid about where the analogy breaks: agents have a notebook rather than experience, rework is nearly free, and the human team still needs your time.

x.com · 22 min · Agent Workflows · Agentic Coding · AI Engineering
11-12

How the Claude Code team reviews 30,000 changes a day

Anthropic's Claude Code team describes its internal review pipeline, built for a repo where daily merged changes went from ~2,500 in May to over 30,000 in September. Review is split into six layers: runnable verification and a fresh-context subagent before the PR exists; a Claude reviewer on every PR that reads the diff against the whole repo and the stated intent, with findings voted on by several agents before posting; a security review that can block merge; CI; risk-tiered human approval (0/1/2 people) assigned by a classifier reading the change rather than the author; then merge queue, flags, internal dogfooding and one-step rollback. They also document the failure modes: one CLI PR ran more than fifty review rounds overnight, and git blame showed ~59% of flagged lines predated the reviewer's first pass, prompting a three-round cap and diff-only re-review. Useful for teams wiring agents into their own review flow.

claude.dev · 19 min · Agentic Coding · Claude Code · Code Review
11-12

Shopify's 12-week rewrite: agents wrote code, humans rebuilt the loop

Thirty Shopify engineers moved a 300+ screen app from React Native to Swift and Kotlin in 12 weeks, with agents writing most of the implementation. The post is about what surrounds the code: business logic runnable without a device, the Reflex architecture and a CLI that loads any screen in milliseconds, one shared architecture across iOS and Android, an example and .md file per screen, and CI that fails when the two platforms diverge. Spec-driven development was abandoned because specs went stale in days; the legacy React Native app became the reference. The second half covers the bottleneck shift — laptops saturating on emulators, review and merge queues, tickets lagging behind Slack-reported fixes — and the responses: a remote dev environment for agents, adversarial agent-first review, and a self-healing loop that reproduces crashes every 15 minutes. Results: Android launch 1.71s to 0.87s, binary 80.4MB to 47.6MB.

shopify.engineering · 19 min · Agent Workflows · Agentic Coding · CI/CD
11-11

Inside Anthropic: how its own teams use Claude Code

Anthropic's own blog rounds up internal interviews on how employees use Claude Code. New data scientists on the infrastructure team hand over the whole codebase and lean on CLAUDE.md files to trace data pipeline dependencies; security engineers paste stack traces and docs into it during incidents and report roughly 3x faster triage; when Kubernetes stopped scheduling pods, the data infra team walked Google Cloud's console with Claude's guidance from dashboard screenshots until they found pod IP exhaustion, then got the exact commands to add a new IP pool. Marketing uses two sub-agents to filter hundreds of ad rows and generate variants; legal prototyped an internal phone tree. The thesis: agentic coding blurs the line between technical and non-technical work. Every case is vendor-reported, with no code, configs, or failure write-ups, so treat the numbers as marketing rather than evidence.

claude.com · 8 min · Agent Workflows · Agentic Coding · AI Productivity
11-11

Prompting Claude Fable 5.2: effort, early stops, code ownership

A migration guide for prompting Claude Fable 5.2, organized by the symptom you observe rather than by feature. Anthropic's position: most of the capability gain already lands at high effort, the medium-to-high step is the main one, and gains above high are uneven and task-dependent, so effort must be re-swept alongside the prompt instead of carried over from Fable 5.1 or Opus 5.5. The doc ships internal test numbers: a completion-prompt cut sessions ending with doable work left by roughly half while raising output tokens about 30 percent; telling the model it owns the code base raised repeated-code merging from under a tenth to about half of tries, at 1.5x output tokens and one-in-six-to-ten behavior changes that closing summaries often omit. It also covers subagent cost descriptions cutting delegation by a third, progress-update thinking blocks being empty at display: omitted, preserved thinking invalidating blocks when harnesses edit earlier turns, and the stop_reason: refusal categories. For engineers migrating an agent harness or maintaining long system prompts and memory files.

platform.claude.com · 22 min · Agent Harness · Agentic Coding · Claude
11-09

Plan With Code, Not Prose: pstack Pt. 2

Part two of poteto's pstack guide covers the front half of the agent workflow: research, planning, prototyping, and architecture. She argues agents mostly fail for two reasons — under-specified intent and missing context — so she has the agent restate a reported problem in its own words before touching code, then uses /how, /why, /recall, and /teach to rebuild runtime mechanics and historical motivation from git history, PR comments, Linear, Slack, Datadog, and Sentry. Planning happens in code rather than prose: README-driven development, a tutorial written before the API, throwaway prototypes driven by /control-app with screenshots and timing measurements, and /architect, which spins up parallel design candidates across model families and has a different model cross-judge them. Escape hatches like any, forced casts, or the same workaround at unrelated call sites count as proof the design is wrong. Aimed at engineers already shipping agent-written code to production.

11-09

The Complete Guide to pstack, Pt. 1: Verification Is All You Need

poteto argues that a verification skill — an agent's ability to drive the app, screenshot it, read traces and confirm its own change — is critical infrastructure rather than a nice-to-have, because it takes the human out of the loop. The post walks through /create-verification-skill: wrap dev-environment control in an agent-friendly CLI (doctor, snapshot, screenshot, wait-settle, feature-flag), then catalog every feature in a markdown Feature Map describing what it does, how a user reaches it, and known gotchas. She advises against scaling with git worktrees (roughly 10 agents per machine before resources bite) and recommends Cursor cloud agents plus /swarm for parallel verification runs. Wired into a scheduled routine, the skill can auto-reproduce user bug reports from Slack. The CLI design rules — --dry-run on destructive commands, descriptive errors, JSON output, progressive disclosure via subcommands — are the most transferable part. Claims of 2,000 PRs/month and 100-1000x output come with no measurement.

x.com · 19 min · Agent Harness · Agent Skills · Agent Tool Design
09-29

Claude Sonnet 5.5 launches with 70.6% on Terminal-Bench 4.0

Anthropic announced Claude Sonnet 5.5, the second model in the Claude 5.5 family, positioned as a faster and cheaper complement to Opus 5.5. It generates output 30%+ faster and costs up to 30% less per task than Sonnet 5. On Terminal-Bench 4.0 it scores 70.6% versus Sonnet 5's 10.3%; gains also show on FrontierCode, CursorBench, GDPval-AA and OSWorld 2.1, with some results near Opus 5.5. Token pricing is unchanged at $2 input / $10 output per million, with $0.20 cache reads, but fewer tokens are needed per task. It is the first Sonnet released with cybersecurity safeguards that fall back to Sonnet 5 for high-risk requests, plus anti-distillation classifiers and preserved thinking. Relevant to engineers choosing models and budgeting agentic coding.

www.anthropic.com · 10 min · Agentic Coding · Ai Safety · Anthropic
09-21

A Behavioral Guide to AI-Assisted Programming

A practitioner's behavioral guide to AI-assisted programming, deliberately tool-agnostic. Core claims: AI agents extend capability but bear no responsibility—humans remain the final owner of the code and should never commit what they cannot explain Feynman-style. Prefer collaboration over delegation; do not behave like a PM who only states requirements. Split PRs (under ~600 lines) and run an AI pre-review before opening one, since review cannot catch everything. Use plan mode to investigate before writing code. Prefer mature libraries over letting the model hand-roll utilities. Make changes verifiable and always verify. For junior engineers: quality over speed, participate in bug fixing instead of unsupervised delegation, design for a fixed 30 minutes before comparing with the AI's answer, read official docs, and shore up design patterns, DDD, security and concurrency.

www.piglei.com · 3 min · Agentic Coding · Agents · AI Engineering
09-20

How Anthropic scaled test impact analysis as agentic coding broke CI

Anthropic's CI absorbed a 25x increase in jobs over six months: Claude now writes 80% of the code, per-engineer quarterly output is 8x the 2021-2025 rate, and the test suite grew 10x while headcount barely moved. The bottleneck moved from writing code to PR review to CI, landing on the test impact analysis service that picks which tests run on each change. Because v0 needed a single writer to keep per-test history ordered, it ran as one process and could not be sharded. Three patches followed: doubling cores bought 70 days, per-package sharding 29 days, and daily restarts under a day. The rewrite moved history into an in-memory data store: any listener worker appends results to a journal and exits stateless, a small consumer rolls the journal into per-test history every few seconds, and the selector queries it. One engineer finished in three weeks.

claude.com · 10 min · Agentic Coding · CI/CD · Engineering Productivity
08-22

Introducing Delta: a multiplayer environment for coding with agents

Zed introduces Delta, a multiplayer environment for coding with agents, now in private beta. At its core is DeltaDB, which replicates both the conversation and the worktree in real time to every participant in a thread, while staying compatible with regular Git repositories. Every edit and conversation is captured between commits, so teammates who never open Delta still see a normal repo. Comments can anchor to any line of code and stay in place as the code evolves, connecting review to the conversation that produced it. The browser client is not a simplified port; it is the same Rust application compiled to WebAssembly and rendered through WebGL. Delta also integrates with third-party agent harnesses starting with Claude Code, syncing terminal sessions live into threads. The interface treats the conversation as an editable document, letting you place your cursor anywhere—on a diff line, a plan step, or a thinking block—and leave comments that the agent precisely understands. A product launch post with solid technical direction but limited hard data.

zed.dev · 7 min · Agent Tooling · Agentic Coding · Code Review