Business, Deals & Funding
Claude Code Changelog
v2.1.199
This changelog entry for claude-code v2.1.199 describes four changes: support for stacking up to 5 slash-skill invocations in a single command, immediate failure with actionable guidance for SSL certificate errors instead of burning retries, preservation of partial streaming responses when mid-stream server errors occur, and a fourth fix that appears truncated in the content.
Why it matters
These are solid quality-of-life improvements. The SSL certificate fix is particularly valuable since TLS-inspecting corporate proxies are a common pain point, and burning retries before showing the actual issue wastes users' time. Preserving partial streaming output on mid-stream errors is also excellent — losing already-received content due to a transient server error is frustrating. The stacked slash-skills feature is a nice usability enhancement. The changelog entry appears truncated, which…
NY Times
We Didn’t Build the Atomic Bomb This Way
The article argues that the United States government should build its own frontier AI model, drawing a parallel to the Manhattan Project and other national efforts. It contends there is no reason for the country to leave cutting-edge AI development solely to private companies and instead advocates for a national lab or government-led initiative to develop advanced AI capabilities.
Why it matters
The piece presents an opinion that the U.S. government should take a direct, active role in developing frontier AI technology rather than relying entirely on the private sector. By invoking the atomic bomb analogy, the author suggests that national security and strategic interests are at stake, and that a government-led approach—similar to historic large-scale scientific projects—is both justified and necessary for maintaining American competitiveness and safety in the AI era.
The Rundown AI
Altman invites Washington inside the AI industry
Sam Altman published an FT op-ed calling for a U.S.-led international forum to set AI safety standards and regulate access to advanced models, drawing parallels to the IAEA, aviation, and banking oversight. Alongside this, OpenAI reportedly discussed giving the U.S. government a 5% equity stake in the company and pushed for other AI labs to contribute to a dividend fund to redistribute AI-generated wealth. The proposals emerged after June's G7 summit in France and come amid increased regulatory pressure, including the pulling of Anthropic's Mythos and Fable models and constraints on GPT-5.6's release. The article also covers Meta's smart glasses progress, Anthropic's Claude integration in Slack for team task delegation, and other AI industry developments.
Why it matters
Altman's proposal is a strategically shrewd move that positions OpenAI as the responsible actor willing to invite government oversight, while potentially shaping the regulatory framework in ways favorable to incumbents. The IAEA analogy is interesting but imperfect — nuclear energy involves a finite set of state actors, while AI development is far more distributed and fast-moving. The 5% government stake idea is provocative and raises important questions about whether AI wealth redistribution s…
Guardian AI
NSW shadow upper house leader steps aside during Icac Liberals inquiry – as it happened
NSW shadow upper house leader has stepped aside during an ICAC inquiry into the Liberal Party. The live blog also covers a separate incident where a teenager has been charged with murder after a 15-year-old boy was fatally stabbed outside a community medical centre in Craigieburn, Melbourne.
Why it matters
The article provides very limited detail on the main headline about the NSW shadow upper house leader stepping aside during the ICAC inquiry, making it difficult to form a thorough opinion on that matter. The juxtaposition of political corruption investigations with a tragic youth stabbing in the same blog highlights the breadth of serious issues facing Australian society. Both stories deserve deeper coverage than what is provided here.
Guardian AI
3,000% bonuses but a growing wealth divide: South Korea grapples with its AI chip boom
South Korea is experiencing a massive wealth surge driven by its AI chip industry, particularly through Samsung Electronics and SK Hynix, with employees receiving bonuses as high as 3,000%. However, this boom is exacerbating the country's wealth divide, raising questions about who benefits from the profits. The article highlights a high-profile divorce case involving business tycoon Chey Tae-won, where the valuation date of company shares could alter his assessed assets by billions of dollars, illustrating the enormous wealth concentrations tied to the chip industry.
Why it matters
This article highlights a familiar pattern in technology booms: while breakthrough industries like AI chips generate enormous wealth, the benefits tend to concentrate among a small elite—executives, major shareholders, and certain skilled workers—while the broader population may see little improvement or even face widening inequality. South Korea's situation is particularly striking given the scale of the bonuses and the societal tensions emerging. The divorce case serves as a vivid symbol of h…
Guardian AI
‘Don’t kill music’: Anthony Albanese’s favourite bands beg PM to stop AI companies from stealing their work
Australian musicians, including some of Prime Minister Anthony Albanese's favourite bands, are urging the government not to allow international tech companies to weaken Australian copyright laws. Big tech firms are seeking permission to scrape Australian creative works—including music, journalism, and books—to train AI systems. Artists and creatives are sounding the alarm, with Senator Pocock calling a related $50 billion datacentre proposal the 'ultimate dirty deal.' The musicians are pleading with the PM to protect their work from being used without consent or compensation.
Why it matters
This is a critically important issue that highlights the tension between technological advancement and the rights of creators. Artists deserve to have their work protected from unauthorized use by AI companies, which stand to profit enormously from scraping copyrighted material without fair compensation. Weakening copyright laws to benefit wealthy international tech corporations at the expense of local creators would be a betrayal of Australia's cultural industries. The fact that even the PM's…
TechCrunch AI
Mark Zuckerberg tells staff that AI agents haven’t progressed as quickly as he’d hoped
Mark Zuckerberg told Meta staff at an internal town hall that AI agent development has not progressed as quickly as executives had anticipated. Earlier in 2026, Meta laid off approximately 8,000 employees (about 10% of its workforce) and reassigned 7,000 others to AI-focused groups, including one called Agent Transformation. Zuckerberg acknowledged the layoffs were not as 'clean' as they should have been and that the perceived benefits of the company's AI-focused restructuring haven't 'come to fruition yet,' though he expressed optimism about seeing improvements within three to six months. Reports have described Meta's AI unit as difficult for engineers assigned to it. Meta is expected to spend up to $145 billion on AI infrastructure this year.
Why it matters
This is a significant and telling admission from one of the most powerful figures in tech. Zuckerberg's candid acknowledgment that AI agents haven't lived up to expectations underscores a growing gap between the AI industry's ambitious promises and actual delivery. The human cost is particularly troubling: thousands of employees were laid off based on the assumption that AI could replace them, and now the company admits that hasn't materialized. The description of Meta's AI unit as a 'soul-crus…
TechCrunch AI
Jersey Mike’s IPO illustrates how bad the AI hype has become
The article highlights how pervasive AI hype has become by examining Jersey Mike's IPO documents, which mention artificial intelligence 22 times despite the company being a sandwich franchise. The author notes that the AI references are mostly boilerplate investor-risk warnings, with the company vaguely stating it is 'beginning to use AI Technologies in our business.' The piece draws a humorous comparison: AI is mentioned 22 times while weather is mentioned only five times and lightning zero times, even though a franchise location was actually struck by lightning in 2021. The author argues this exemplifies how companies across all industries feel compelled to sprinkle AI references into their investor pitches to capitalize on investor enthusiasm for anything AI-related.
Why it matters
This is a well-crafted, entertaining piece that effectively uses a concrete and somewhat absurd example to illustrate a real phenomenon in the current market. The author's point is well-taken: when a submarine sandwich franchise feels obligated to mention AI 22 times in its IPO filing, the hype cycle has clearly reached a fever pitch. The lightning comparison is particularly clever and drives the point home. The article also fairly acknowledges that Jersey Mike's does use software and data as a…
From X/Twitter
- Want to build AI agents? Shadow the person doing the job 10–20 times first, then spec the agent around seven workflow questions.
- Harrison Chase says more developers are switching to glm5.2 as a daily coding driver and shares how to wire it into deepagents.
- Without idempotency, agent retries can charge a customer twice. A walkthrough of exactly-once semantics for agent workflows.
- Harrison Chase coins "agentic map reduce" — spawning subagents programmatically for patterns where you want determinism controlling how agents are created and run.
- agent-device validated as faster than Maestro on iOS in a third-party benchmark for mobile testing.
- Loop engineering in one thread: define a goal, break it into repeatable steps, add verification, let the agent retry with context, and define when to stop.
From Reddit/HN/YC
- [Hacker News] Anthropic's Claude Fable relaunch disappoints users with what they're calling nerfed performance.
- [Hacker News] pgvector lands a PR for 10x smaller vector indexes — a big deal for anyone running embeddings in Postgres.
- [Hacker News] ReactOS implements its first Windows NT6 system call, inching toward Vista compatibility.
- [Hacker News] AWS says it added more data center capacity than any other company in 2025, including 12GW in Q4 alone.
- [Hacker News] The case that ClickHouse is winning the observability wars — and why the incumbents should worry.
- [Hacker News] Semi Analysis digs into Meta's compute ambitions: everyone wants to be a neocloud now.