Claude Code

Evidence (19)

• 4The New Stack·9/21/2026product_launch
  Grok Build and Claude Code compared on memory in coding agents

  The article tests whether Grok Build and Claude Code can remember project facts and rules across coding sessions. Both tools handled within-repository memory well, but Grok was cheaper in the author’s tests, while Claude performed better on a cross-project rule only after using its global memory support.

• 4Simon Willison·9/20/2026product_launch
  Employee describes heavy Claude Code dependence at a large company

  A new employee at a large company says most documentation, code, tests, PRDs, tickets, and reports are being generated with Claude Code. The post claims staff are under pressure to ship faster, with long work hours and limited review of the generated material.

• 4Hacker News·9/18/2026framework_update
  Claude Code Adds AGENTS.md Support

  Claude Code version 2.1.277 will now look for and use AGENTS.md in a folder when CLAUDE.md is not present. The feature is based on Claude Code mods, which Anthropic says will also support custom project instructions in the future.

• 5Hacker News·9/4/2026product_launch
  Spotify Portal Mode Router Cuts Claude Code Token Use

  The article describes a workflow using Portal by Spotify's AiKA Modes to route repetitive coding tasks away from Claude Code and into cheaper specialized modes. By separating file reading and boilerplate generation from higher-cost reasoning, the author says token usage fell by about 90%.

• 7The New Stack·7/23/2026product_launch
  OpenAI and Anthropic unveil competing voice upgrades

  OpenAI expanded ChatGPT Voice to control desktop apps and AI tasks across macOS and Windows, including support for background task execution and app context on Mac. Anthropic, by contrast, updated Claude Voice Mode and Claude Code Voice to support longer, iterative problem-solving and voice-driven coding workflows.

• 6Hacker News·7/12/2026research
  Study Finds Claude Code Uses More Tokens Than OpenCode

  A Hacker News post describes a small measurement study comparing Claude Code and OpenCode by logging requests to Anthropic's endpoint. The author reports that Claude Code sent far more pre-prompt and harness tokens and appeared less efficient in cache usage than OpenCode, with one caveat noted in the methodology.

• 4Hacker News·7/9/2026product_launch
  LazyPi adds opinionated one-command setup for Pi.dev

  LazyPi is a community package that automatically installs and configures Pi.dev with a curated set of skills, themes, MCP support, sub-agents, memory, and other extensions. The project aims to remove setup overhead by giving users a ready-to-use starting point while still allowing selective installation.

• 6The New Stack·7/6/2026research
  Study Finds AI Coding Tools Haven't Hurt Open Source Newcomer Participation

  A Peking University study of 1,888 GitHub repositories found that adopting AI coding agents such as Cursor and Claude Code did not meaningfully reduce newcomer participation. The projects saw modest increases in code complexity, but contributor entry and retention generally held steady in the established repositories analyzed.

• 5The New Stack·7/6/2026research
  JetBrains finds Claude Code caveman mode saves far fewer tokens than claimed

  As AI coding tools increasingly use usage-based pricing, developers are trying to reduce token consumption by making assistant replies more terse. JetBrains tested a popular Claude Code skill that forces blunt, caveman-style output and found it reduced tokens by about 8.5% in real coding tasks, far below its claimed 65% savings because most agent output is still code, diffs, and exact tool output.

• 8Ars Technica AI·7/6/2026security
  Anthropic removes hidden Claude Code tracking after backlash

  Anthropic removed a hidden tracker in Claude Code after a researcher found code that quietly flagged some users in China and sent related signals back to the company. The company said the feature was an experiment aimed at preventing abuse and unauthorized reselling, but critics described it as a serious breach of user trust.

• 7The New Stack·7/6/2026framework_update
  Markdown emerges as the shared format for AI agent context

  Andrej Karpathy’s LLM Wiki, Google’s Open Knowledge Format, and Garry Tan’s gstack all point to Markdown as the practical format for agent memory, enterprise knowledge, and role-based coding workflows. The article argues that the competitive advantage is shifting from model ownership to the Markdown-based knowledge teams accumulate and can carry across tools and vendors.

• 5Hacker News·7/6/2026product_launch
  Criticism of Anthropic’s Claude pricing and ecosystem limits

  The article argues that Anthropic has reduced goodwill among developers by tightening subscription restrictions, splitting billing between first-party and third-party tools, and relying on extra usage charges. It also criticizes Claude Code’s reliability and lock-in, while noting that alternative open-source coding tools are becoming more competitive.

• 5Simon Willison·7/2/2026product_launch
  llm-coding-agent 0.1a0 ships as an alpha coding agent

  Simon Willison released llm-coding-agent 0.1a0, a new Python coding agent built on the evolving llm agent framework. It can read and edit files, run commands, and expose a CLI and Python API for agentic coding workflows, with the package published to PyPI as a prerelease.

• 4Simon Willison·7/2/2026research
  Using DSPy to improve Datasette Agent SQL prompts

  Simon Willison describes using DSPy to evaluate and refine the system prompts behind Datasette Agent's read-only SQL feature. An initial pass found that prompt wording around schema guidance could lead to column-name guessing and error-retry loops, suggesting prompt changes or adding column names to the schema listing.

• 8Hacker News·6/30/2026security
  Claude Code Allegedly Adds Steganographic Request Markers

  A Hacker News discussion highlights claims that Claude Code is steganographically marking requests, suggesting the tool may be embedding hidden identifiers or signals in its prompts. The report has drawn significant attention because it raises questions about transparency, prompt handling, and possible covert metadata in AI developer tools.

• 7Hacker News·6/28/2026research
  GLM 5.2 Outperforms Claude in IDOR Benchmarks

  Semgrep reports that GLM 5.2, an open-weight model from Zhipu AI, achieved a 39% F1 score on IDOR detection, ahead of Claude Code's 32% in the same benchmark setup. The test highlights how much vulnerability-detection performance depends on the surrounding harness, while Semgrep's own purpose-built pipeline still performed better overall.

• 4Hacker News·6/28/2026research
  Using Claude Code to review an MRI

  A Hacker News post describes using Claude Code with Opus 4.8 to analyze a shoulder MRI and generate a second opinion, then comparing that output with a human radiology report. The author also discusses earlier AI-assisted review of the recommended treatments, highlighting discrepancies and uncertainty rather than a clinical conclusion.

• 9Hacker News·6/28/2026security
  Nearly one million passport images exposed online

  Security researcher Sammy Azdoufal found nearly 985,000 photo IDs and passports exposed at public URLs tied to Cannabis Club Systems’ cloud services. The issue also affected member profiles, admin access, and private chat data, and the company has begun shutting down vulnerable systems while fixes are made.

• 4Hacker News·6/27/2026open_source
  Adrafinil keeps a Mac awake only while AI agents are active

  Adrafinil is an open-source macOS app designed to keep a MacBook awake with the lid closed only while connected agents are actively working. It integrates with tools such as Claude Code and Codex, automatically re-enables sleep when work finishes, and includes safeguards for overheating and manual control via MCP.