Business, Deals & Funding
Claude Code Changelog
v2.1.141
Version 2.1.141 of Claude Code adds several features: a `terminalSequence` field to hook JSON output for emitting desktop notifications/window titles/bells without a controlling terminal, `CLAUDE_CODE_PLUGIN_PREFER_HTTPS` environment variable for cloning GitHub plugin sources over HTTPS instead of SSH, `ANTHROPIC_WORKSPACE_ID` environment variable for workload identity federation to scope tokens to specific workspaces, and a `claude agents --cwd` command (description appears truncated in the changelog).
Why it matters
This appears to be a solid incremental release focused on improving flexibility and configurability. The terminal sequence hook addition is clever for headless/CI environments. The HTTPS plugin cloning option addresses a real pain point for environments without SSH keys configured. The workspace ID for identity federation shows continued enterprise-focused development. The changelog entry appears truncated, so there may be additional features not captured here.
TechCrunch AI
Who decides what AI tells you? Campbell Brown, once Meta’s news chief, has thoughts
Campbell Brown, former Meta news chief, has launched Forum AI, a 17-month-old New York-based company that evaluates how foundation AI models handle high-stakes topics like geopolitics, mental health, finance, and hiring. The company recruits world-class experts—including Niall Ferguson, Fareed Zakaria, Tony Blinken, and Kevin McCarthy—to architect benchmarks, then trains AI judges to evaluate models at scale, aiming for roughly 90% consensus with human experts. Brown was motivated by watching ChatGPT's release while at Meta and realizing AI would become the primary information funnel despite significant accuracy problems. Forum AI's evaluations have found issues including models pulling from Chinese Communist Party websites for unrelated stories, left-leaning political bias across nearly all models, missing context, and straw-manned arguments. Brown believes enterprise clients—companies…
Why it matters
This is a genuinely important initiative addressing one of the most consequential questions of the AI era: who ensures the accuracy and balance of information that AI systems deliver to billions of people. Brown's credibility is bolstered by her direct experience watching Facebook's failures with misinformation—she saw firsthand how optimizing for engagement degraded public discourse. The expert-driven benchmarking approach is smart, though the choice of experts will inevitably invite scrutiny…
TechCrunch AI
Clio’s $500M milestone arrives just as Anthropic ups the ante
Clio, an 18-year-old Canadian legal tech company, has reached $500 million in annual recurring revenue (ARR), accelerating rapidly after integrating AI in 2023—growing from $200M ARR in mid-2024 to $400M by late 2025 to $500M now. CEO Jack Newton argues legal tech is poised to be the next major AI winner after code writing, given the vast corpus of legal text available for training. Other legal AI companies are also surging: Harvey reached $190M ARR by end of 2025, and Legora hit $100M ARR just 18 months after launch. However, Anthropic has complicated the landscape by launching Claude for Legal, a law-focused suite that makes it both a key supplier and competitor to companies like Harvey and Legora that rely on Claude. Clio was valued at $5 billion after raising a $500M Series G last November and acquired data intelligence platform vLex for $1 billion to expand its AI research capabili…
Why it matters
This article highlights a genuinely compelling AI use case. Legal work is text-heavy, repetitive, and expensive—exactly the kind of domain where LLMs can deliver transformative value. Clio's revenue trajectory from $200M to $500M ARR in roughly a year is remarkable and suggests real product-market fit, not just hype. The most interesting tension in the piece is Anthropic's move into legal-specific features, which creates a classic platform risk problem for Harvey and Legora. This dynamic—where…
The Verge AI
Microsoft’s Edge Copilot update uses AI to pull information from across your tabs
Microsoft Edge is updating its Copilot AI chatbot to gather and use information from all open browser tabs. Users can ask Copilot to compare products, summarize articles, and answer questions based on tab content. Microsoft is retiring Copilot Mode, folding its agentic features into 'Browse with Copilot.' New features include a 'Study and Learn' mode that creates quizzes from articles, AI-generated podcasts from tabs similar to NotebookLM, and an AI writing assistant. Users can also grant Copilot access to browsing history for more relevant answers, and the chatbot will feature 'long-term memory' to personalize responses based on past conversations. A redesigned new tab page will combine chat, search, and web navigation.
Why it matters
This update represents a significant deepening of AI integration into the browser experience, which is both impressive and concerning. The ability to synthesize information across tabs is genuinely useful for research and shopping. However, the combination of tab access, browsing history access, and long-term conversation memory creates a remarkably comprehensive surveillance profile of user behavior. Microsoft framing this as opt-in is important, but the trend toward AI that knows everything a…
NY Times
U.C.F. Students Boo Commencement Speaker for A.I. Comments
Students at the University of Central Florida's commencement ceremony booed a speaker after she praised artificial intelligence as 'the next industrial revolution.' The speaker, addressing graduates from creative fields, attempted to continue her remarks about AI's potential to solve humanity's problems, but was met with sustained negative reactions. A student interviewed explained that many art students oppose generative AI because it can only reproduce existing work, whereas artists want to create original stories from personal experiences. The incident highlighted tensions between tech industry enthusiasm for AI and the concerns of young creative professionals entering the workforce.
Why it matters
This incident captures a genuine and understandable frustration among creative graduates who are watching their career prospects be actively undermined by the very technology being celebrated at their graduation. The speaker's tone-deafness is remarkable — telling a room full of arts graduates that the technology threatening to devalue their skills is exciting is almost comically oblivious. The student's response was articulate and important: generative AI recombines what already exists, while…
TechCrunch AI
Notion just turned its workspace into a hub for AI agents
Notion has launched a new developer platform that transforms its workspace into a hub for AI agents. The platform introduces several key features: Workers (a cloud-based environment for running custom code in secure sandboxes), database sync capabilities that pull data from external sources like Salesforce and Zendesk, and the ability to chat with and manage external AI agents like Claude Code, Cursor, Codex, and Decagon directly within Notion. Building on its Custom Agents feature launched in February (which has already seen over 1 million agents created), the platform addresses previous limitations by allowing external data connections, custom logic, and webhook-triggered workflows. CEO Ivan Zhao acknowledged Notion hasn't historically been developer-focused but signaled a strategic shift. Workers use the same credit system as Custom Agents and will be free through August. The platfor…
Why it matters
This is a significant and well-timed strategic move by Notion. By positioning itself as an orchestration layer for AI agents rather than just another productivity tool with AI bolted on, Notion is making a credible play to become the central nervous system of knowledge work. The numbers speak for themselves — 1 million custom agents created since February suggests genuine product-market fit. The Workers feature is particularly smart because it lowers the barrier for developers while keeping exe…
TechCrunch AI
Musk’s xAI is running nearly 50 gas turbines unchecked at its Mississippi data center
Elon Musk's xAI is operating 46 natural gas turbines at its Colossus 2 data center in Mississippi, exploiting a state loophole that classifies the turbines as 'mobile' because they sit on flatbed trailers, allowing them to avoid air pollution regulations for one year. The NAACP, represented by the Southern Environmental Law Center, has filed a lawsuit arguing the turbines violate federal law, which can classify trailer-mounted power plants as stationary sources subject to air-pollution regulations. The organization has asked the court for an injunction against xAI, citing worsening air quality in an already polluted region. While xAI has permits for only 15 of the turbines, the company has continued installing more beyond the roughly 35 that were previously reported operating.
Why it matters
This is a troubling example of a powerful tech company exploiting regulatory loopholes to prioritize rapid infrastructure deployment over environmental responsibility and community health. Calling industrial gas turbines 'mobile' simply because they sit on trailers is a transparently cynical workaround, and the fact that xAI has permits for only 15 of its 46 turbines makes the situation even more egregious. The AI industry's insatiable demand for compute power is creating real environmental ext…
TechCrunch AI
Anthropic’s Cat Wu says that, in the future, AI will anticipate your needs before you know what they are
Cat Wu, Anthropic's head of product for Claude Code and Cowork, discussed the company's product strategy and vision for AI's future at the Code with Claude conference. She emphasized that Anthropic focuses on staying at the frontier rather than reacting to competitors, and expects the rapid pace of model releases to continue. Wu believes the future of work involves humans managing fleets of AI agents, but stresses that managers must remain domain experts. The article also notes Anthropic's strong business performance, potentially raising funds at a $950 billion valuation, and its growing market share over OpenAI among business customers. The company's Glasswing initiative, which provides limited access to its cybersecurity model Mythos, was also discussed as an example of balancing capability with safety.
Why it matters
This article provides an interesting glimpse into Anthropic's product philosophy and trajectory, though it reads more like a promotional piece than critical journalism. Wu's insight that designing reactively to competitors puts you perpetually behind is a genuinely valuable strategic observation. The claim that AI will anticipate needs before users know them is a bold but somewhat cliché prediction that deserves more scrutiny — the privacy and autonomy implications are significant and largely u…
From X/Twitter
- Google open-sourced design.md, a standard file format that tells AI agents how your brand looks — because 70% of brands won't show up when an agent goes shopping.
- OpenAI data shows professional and agency services make up 22% of all active entrepreneurs on ChatGPT — the single biggest category, ahead of retail, healthcare, and tech.
- The real skill with Claude Code isn't prompting — it's structuring your repo so Claude thinks like an engineer instead of a chatbot.
- Claude Opus 4.7: low-effort mode matches medium-effort Opus 4.6 in output quality — same results, fewer tokens.
- The Obsidian founder's own setup: barely any folders, heavy internal linking, categories as properties on the note itself.
- Aakash Gupta explains why your Claude skills never fire: the model only scans the description to decide routing — it never reads the full instructions first.
From Reddit/HN/YC
- [Hacker News] A sharp argument that LLMs are breaking 20-year-old system design assumptions we've taken for granted.
- [Hacker News] A Nature study finds state media control shapes LLM behaviour by skewing the training data models learn from.
- [Hacker News] Tenstorrent unveils Galaxy, an AI platform designed to compete on scale and efficiency rather than raw parameter count.
- [Hacker News] After eight years, Noah Golmant rewrote his open-source PyTorch curvature library from scratch.
- [Hacker News] Kubesearch lets you search through Kubernetes YAMLs scraped from real homelabs.
- [Hacker News] Poppy — a dynamic instrumentation pipeline for macOS aimed at security research.