Claude Code

Evidence (20)

• 5Simon Willison·6/11/2026open_source
  Claude Fable Uses Browser and Template Hacks to Test UI Bugs

  Simon Willison describes Claude Fable 5 as extremely proactive in investigating a UI scrollbar bug in Datasette Agent. The agent created scratch HTML pages, opened Safari, triggered UI actions by editing templates with injected JavaScript, and even built a local HTTP server to collect measurements from browser-executed code. The post highlights how aggressively the tool uses local files, browser windows, and automation-like techniques to pursue a debugging goal.

• 2Hacker News·6/11/2026product_launch
  Hacker News discusses staying in flow while coding with AI

  A Hacker News post asks how developers maintain deep focus and a flow state when using slow AI coding agents such as Claude. The discussion centers on workflow habits and focus strategies rather than a specific product announcement.

• 5Simon Willison·6/11/2026framework_update
  Datasette 1.0a33 adds JSON extras for queries and rows

  Datasette 1.0a33 extends the ?_extra= API pattern to cover queries and rows as well as tables. The release also documents the feature and includes a custom API explorer demo built with Claude Code and Codex Desktop.

• 6The New Stack·6/11/2026framework_update
  Designing an AI-first software delivery flow

  The article argues that teams can delegate a significant share of ticket-to-production work to AI if the software delivery process is redesigned around agents. It outlines a five-phase workflow with human approval gates, supported by a context lake, guardrails, and visibility across planning, development, preview, and deploy steps.

• 4The New Stack·6/10/2026product_launch
  AI Agents Aim to Tame Observability Overload

  The article argues that modern observability stacks generate more logs, traces, and alerts than engineers can efficiently handle, slowing root-cause analysis and resolution. It highlights AI agents as a proposed solution, with vendors building systems that can correlate observability data, remediate issues autonomously, or surface guidance inside tools like Codex, Cursor, and Claude Code.

• 7The New Stack·6/10/2026framework_update
  Anthropic’s Claude Code creator says agents now need loops

  Anthropic’s Boris Cherny said he no longer prompts Claude directly and instead designs loops that run the work. The article frames this as “loop engineering,” an orchestration pattern built from scheduled execution, isolated workspaces, verifier agents, and persistent memory for long-running coding agents.

• 8The New Stack·6/10/2026security
  Microsoft Pulls 73 GitHub Repos After Malware Attack

  Microsoft temporarily removed 73 of its GitHub repositories after a malicious commit was found that could steal developers’ credentials when opened in AI coding agents. The company says it notified a small number of customers and later restored the repositories, but it has not disclosed how many developers were affected or exactly how the intrusion occurred.

• 4Simon Willison·6/10/2026research
  Jeremy Howard argues top AI labs should not use their own frontier models for research

  Jeremy Howard argues that if a lab holds the top-ranked AI model and wants to slow recursive AI self-improvement, it should refrain from using that model for frontier AI work while keeping access open for others. He says Anthropic is taking the opposite approach by using its top model for frontier research and restricting others, which he считает increases both frontier advancement and power imbalance.

• 3Hacker News·6/10/2026open_source
  macOS Menu Bar Gauge for Claude Code Quota

  This Show HN item presents an open-source macOS app that displays Claude Code quota usage directly in the menu bar. It is aimed at helping developers monitor their remaining usage more conveniently while working with Anthropic's coding tool.

• 5TechCrunch AI·6/9/2026product_launch
  Anthropic's Fable 5 turns text into playful video games

  Anthropic says its Claude Fable 5 can generate weirdly fun video games with a single click, aiming at developers and the growing community of vibe coders. The item highlights AI-assisted game creation as a new use case for Anthropic's model lineup.

• 3Simon Willison·6/9/2026research
  Karpathy on the rising demand for software

  Andrej Karpathy says that as software becomes easier to generate on demand, people will want far more of it for highly specific tasks. He points to examples such as explainers, visualizers, dashboards, custom single-use apps, test-suite expansion, code optimization, and large research projects with tailored output.

• 7The New Stack·6/9/2026product_launch
  Lindy switches from Anthropic to DeepSeek to cut AI inference costs

  Lindy, an AI agent platform, says it moved 100% of its traffic from Anthropic models to DeepSeek v4. The company says the switch reduces costs by millions of dollars while improving performance on some core use cases such as email triage and draft replies.

• 7The New Stack·6/8/2026framework_update
  Claude Code adds dynamic workflows for parallel agent orchestration

  Anthropic introduced dynamic workflows in Claude Code with Opus 4.8, a research preview feature that lets Claude write its own orchestration scripts and run multiple subagents in parallel. In a test described by The New Stack, the feature completed a local codebase analysis tool with five agents working at once and produced a working CLI with passing tests.

• 6Simon Willison·6/3/2026product_launch
  Uber Limits Monthly Spending on AI Coding Tools

  Uber has capped employee spending at $1,500 per month per AI coding tool to control costs, according to a Bloomberg report cited in the article. The limit applies to agentic coding tools such as Cursor and Anthropic's Claude Code, and is set separately for each tool.

• 2Simon Willison·6/2/2026product_launch
  Pasted File Editor Prototype

  Simon Willison describes a prototype built with Codex Desktop that mimics Claude's behavior of turning large pasted text into a file attachment. The tool also supports opening files directly, previewing images as thumbnails, and dragging files into the text area.

• 4Simon Willison·5/31/2026research
  AI coding agents may be too easy to overuse

  Simon Willison highlights a post arguing that AI coding tools can quickly turn small requests into many partially finished side projects, creating a distraction risk rather than solving the original task. The discussion also notes that some developers with ADHD say these same tools help them focus and complete work more effectively.

• 3Simon Willison·5/30/2026open_source
  Open Source Leader Leaves Tech for an Offline Life

  Chad Whitacre says he is stepping away from tech and open source, citing AI and the pressures of modern digital life as key reasons. Simon Willison notes that Whitacre’s earlier experience with agentic coding tools appeared to reinforce his decision, and says he will be missed in the open source community.

• 6Hugging Face Blog·1/28/2026research
  Claude Builds CUDA Kernels and Teaches Open Models

  Hugging Face reports an experiment where Claude was used to generate CUDA kernels and help teach open models. The post highlights how the model can assist with low-level GPU optimization as well as model distillation or training workflows. It is presented as a practical demonstration of AI helping with both systems programming and model development.

• 7VentureBeat AI·1/22/2026funding
  Railway raises $100M to build AI-native cloud infrastructure

  Railway raised $100 million in a Series B led by TQ Ventures to expand its cloud platform as AI-driven development increases pressure on legacy infrastructure. The company says its vertically integrated system enables sub-second deployments, lower costs, and faster workflows than traditional providers such as AWS and Google Cloud.

• 7VentureBeat AI·1/19/2026open_source
  Goose Offers a Free Offline Alternative to Claude Code

  Anthropic's Claude Code has drawn criticism over monthly pricing and usage limits, prompting some developers to look for alternatives. Goose, an open-source coding agent from Block, is gaining attention for offering similar functionality on a local machine with no subscription fees, cloud dependency, or rate limits.