Business, Deals & Funding
Claude Code Changelog
v2.1.139
This changelog entry for Claude Code v2.1.139 describes several new features: an 'agent view' for monitoring all Claude Code sessions, a '/goal' command for setting completion conditions that Claude works toward across turns, a '/scroll-speed' command for adjusting mouse wheel speed, and what appears to be a truncated mention of another 'claude plu...' feature.
Why it matters
The content is truncated, making it impossible to fully evaluate. From what's visible, the agent view and /goal command seem like meaningful productivity features for Claude Code users. The agent view addresses a real need for managing multiple sessions, and the /goal command adds useful automation. However, this is just a changelog snippet with minimal detail, so it's hard to assess the actual quality of implementation.
Guardian AI
Datacentres should be forced to invest in wind and solar energy, agree all states except Queensland
Australian state and federal energy ministers, with the exception of Queensland, have agreed that datacentres should be required to invest in new renewable energy generation and storage sufficient to fully offset their electricity consumption. The policy targets the growing energy demands driven by artificial intelligence and would mandate that datacentres invest in wind and solar capacity to completely cover their power needs.
Why it matters
This is a sensible and forward-thinking policy proposal. As AI-driven datacentres are poised to become enormous consumers of electricity, requiring them to fund equivalent renewable energy capacity ensures that their growth doesn't undermine climate goals or strain existing grids. It places the cost burden appropriately on the industry profiting from the energy use rather than on taxpayers or existing consumers. Queensland's dissent is notable and likely reflects its closer ties to fossil fuel…
TechCrunch AI
Thinking Machines wants to build an AI that actually listens while it talks
Thinking Machines Lab, the AI startup founded by former OpenAI CTO Mira Murati, has announced 'interaction models' — AI that can process user input and generate responses simultaneously, enabling full-duplex conversation akin to a phone call rather than turn-based exchanges. Their model, TML-Interaction-Small, reportedly responds in 0.40 seconds, matching natural human conversation speed and outpacing comparable models from OpenAI and Google. However, this is currently a research preview with no public release yet; a limited research preview is expected in the coming months, with a wider release later in 2026.
Why it matters
This is a genuinely interesting conceptual advance. The turn-taking paradigm of current AI conversations is one of the most obvious gaps between human-to-human and human-to-AI interaction, and making interactivity native to the model rather than a bolted-on feature is a compelling architectural philosophy. That said, the skepticism in the article is well-placed — impressive benchmarks on paper don't always translate to impressive real-world experiences, and a 0.40-second response time means lit…
TechCrunch AI
Riding an AI rally, Robinhood preps second retail venture IPO
Robinhood has filed a confidential registration for its second venture fund (RVII), which will invest in growth-stage and early-stage startups, expanding beyond the late-stage focus of its first fund (RVI). RVI, which debuted on the NYSE in March at $21 per share, has more than doubled to $43.69, driven by AI enthusiasm. The first fund holds stakes in companies like OpenAI, Databricks, Stripe, and ElevenLabs. RVII's fundraising target hasn't been set yet, though RVI fell several hundred million short of its $1 billion goal. The funds aim to democratize venture investing by allowing non-accredited retail investors to buy shares in a portfolio of private startups through regular brokerage accounts, with daily liquidity and no carried interest. CEO Vlad Tenev envisions retail investors eventually participating in seed and Series A rounds alongside traditional venture firms.
Why it matters
This is a genuinely significant development in the democratization of venture capital, though it comes with substantial risks that the article only briefly acknowledges. The fact that RVI has doubled in two months likely reflects speculative AI hype more than fundamental value creation, which should concern anyone watching retail investors pile into illiquid private company stakes wrapped in a liquid wrapper. The no-carry, no-accreditation model is compelling and addresses a real inequity, but…
The Verge AI
OpenAI just released its answer to Claude Mythos
OpenAI has launched Daybreak, an AI-powered cybersecurity initiative designed to detect and patch software vulnerabilities before attackers can exploit them. Daybreak combines OpenAI's Codex Security AI agent (launched in March) with specialized cyber models including GPT-5.5-Cyber and GPT-5.5 with Trusted Access for Cyber. The system creates threat models based on an organization's code, identifies possible attack paths, validates likely vulnerabilities, and automates detection of high-risk ones. The launch comes about a month after Anthropic announced Claude Mythos and Project Glasswing, a competing security-focused AI initiative that Anthropic deemed too dangerous for public release but which was reportedly accessed by unauthorized parties. Like Glasswing, Daybreak is not built on a single AI model but integrates multiple models and security partners. OpenAI says it is working with i…
Why it matters
This represents a significant escalation in the AI cybersecurity arms race between major AI companies. The framing as an 'answer to Claude Mythos' highlights how competitive dynamics are now driving the development of increasingly powerful security-focused AI tools. There are both promising and concerning aspects here. On the positive side, AI-powered vulnerability detection could meaningfully improve software security by finding flaws faster than human security researchers can. However, the ra…
TechCrunch AI
GM just laid off hundreds of IT workers to hire those with stronger AI skills
General Motors has laid off approximately 600 salaried IT employees — over 10% of its IT department — in what the company describes as a deliberate skills swap. Rather than simply reducing headcount, GM is actively hiring replacements with AI-focused expertise, including AI-native development, data engineering, agent and model development, prompt engineering, and cloud-based engineering. This follows 18 months of white-collar layoffs at GM, including 1,000 software workers cut in August 2024. The restructuring has been shaped by chief product officer Sterling Anderson, hired in May 2025, who consolidated GM's technology businesses and prompted the departure of three senior executives. GM has since hired new AI-focused leaders, including former Apple AI lead Behrad Toghi and former Cruise head of AI Rashed Haq. TechCrunch frames this as a signal of what enterprise AI adoption looks like…
Why it matters
This story represents a significant and somewhat sobering milestone in the AI transformation of traditional industries. GM's move is notable because it's not a cost-cutting exercise disguised as modernization — they're explicitly replacing one set of skills with another, which is more honest than most corporate restructurings but no less painful for the affected workers. The specific roles they're hiring for — agent development, model engineering, AI-native workflows — suggest GM is serious abo…
Ars Technica AI
Linux bitten by second severe vulnerability in as many weeks
A severe Linux vulnerability called 'Dirty Frag' has been disclosed, allowing low-privilege users and containers to gain root access on Linux servers. The exploit chains two kernel vulnerabilities (CVE-2026-43284 and CVE-2026-43500) related to page cache handling in networking and memory-fragment components. The exploit is deterministic, works reliably across virtually all Linux distributions, and causes no crashes. Proof-of-concept exploit code was leaked online, effectively making it a zero-day, and Microsoft has observed hackers experimenting with it in the wild. This comes just weeks after another similar vulnerability called 'Copy Fail.' Patches are being released by distributions including Debian, AlmaLinux, and Fedora, and organizations are urged to apply them immediately.
Why it matters
This is a genuinely alarming development for Linux security, particularly for shared hosting environments, cloud providers, and container orchestration platforms where privilege escalation from untrusted users is a critical threat model. The fact that this is the second severe page-cache exploitation vulnerability in two weeks, following the same bug family as Dirty Pipe from 2022, suggests a systemic weakness in how the Linux kernel handles page cache references that needs deeper architectural…
The Verge AI
Here’s what Mira Murati’s AI company is up to
Thinking Machines, the AI company founded by former OpenAI CTO Mira Murati, announced it is developing 'interaction models' designed to enable real-time, multimodal collaboration between humans and AI. Unlike current models that process input in a single thread and freeze perception while generating responses, interaction models continuously take in audio, video, and text, thinking and responding in real time. The company shared demos including real-time speech translation, detecting animal mentions in stories, and posture correction. Thinking Machines argues this approach solves a 'bandwidth bottleneck' in human-AI collaboration by meeting humans where they are rather than forcing them to adapt to AI interfaces. The models are not yet publicly available, with plans for a limited research access program.
Why it matters
This is an interesting conceptual framing, though the actual differentiation from existing multimodal real-time AI systems (like GPT-4o's voice mode or Google's Project Astra) remains unclear from the announcement alone. The 'interaction model' terminology is clever branding, but the demos shown—real-time translation, audio monitoring, posture detection—are capabilities others have already demonstrated. The key question is whether Thinking Machines has a genuinely novel architectural approach t…
From X/Twitter
- Santiago breaks down how to learn Claude Code from scratch: start with the agent loop, then permissions, then auto mode.
- A 65-line markdown file starred by 120,000 developers reportedly cut Claude's coding mistake rate from 41% to as low as 3%.
- Software engineering is shifting from a deterministic contract to a probabilistic one — agent fleets grinding 24/7 while you sleep, you wake up to triage PRs.
- Matt Pocock makes the case that AI didn't make code cheap — it made bad code lethal, and fundamentals matter more than ever.
- Thariq's thesis: HTML gives agents a richer canvas than Markdown — diagrams, styling, SVGs, interactive elements, and better human-in-the-loop workflows.
- Garry Tan says when people ask how he prompts AI, the answer is he doesn't — the skills are the prompts.
From Reddit/HN/YC
- [Hacker News] Bloomberg makes the case that $70 should be the most worrying number for LNG markets right now.
- [Hacker News] Linux hit by a second severe vulnerability in as many weeks — not a great stretch for kernel security.
- [Hacker News] New research catalogs LLM hallucinations in the wild, mapping how and when models confabulate outside the lab.
- [Hacker News] Peter Wang reflects on his first 20 years of Python — from early adopter to Anaconda CEO.
- [Hacker News] Somebody got 60fps video running on a CGA with a technique called GlyphBlaster. Retro wizardry.
- [Hacker News] The Great Trigonometrical Survey — the 70-year project that measured India and discovered Everest's height.