Nvidia, Microsoft, Meta, IBM, Palantir, Hugging Face, a16z, and Y Combinator signed an open letter warning against "premature restrictions" on open-weight models as Washington weighs a ban on Chinese ones. OpenAI and Google have both voiced support, though reporting differs on whether either formally signed. The line worth reading twice: "Relying solely on closed models is not inherently safe." (CNBC)

Anthropic shipped Claude Opus 5 at $5/$25 per million tokens, unchanged from Opus 4.8 and roughly half of Fable 5's price. It scored 43.3% on Frontier Bench v0.1 against Fable 5's 33.7%, and it's now the default on Claude Max. If you're running long agent loops, the cost curve moved under you over the weekend. (SiliconANGLE)

Moonshot AI published the full weights for Kimi K3 at 00:00 UTC, making a 2.8-trillion-parameter model the largest open-weight release in history. It runs a 1-million-token context window and reportedly beat Claude Fable 5 in the Frontend Code Arena benchmark. The weights are roughly 1.4TB, so you won't run this on a workstation, but anyone can now download, modify, and self-host a frontier-class model without sending a single token to someone else's API. (Tom's Hardware)

Anduril is in talks to raise at roughly a $100 billion valuation, up from $61 billion just two months ago and more than triple last year's mark. That puts a company founded in 2017 in the same conversation as Northrop Grumman and Lockheed Martin. One structure floated: a two-stage round where investors commit to a second, higher-priced tranche tied to hitting financial benchmarks. (Yahoo Finance)

A single fallen power line outside DC caused more than 3 gigawatts of data centers to drop off the grid within 30 seconds, and it took 11 minutes to stabilize instead of the usual few. That's 3% of total PJM demand, across a grid serving 67 million people. The fixes already exist: sequential disconnect protocols, campus-scale battery UPS, and ERCOT-style "ride through" rules. Worth knowing where your compute actually sits. (TechCrunch)

Three hires. None of them could do the job. Their English was fine.

You've read the resume that checks every box. Right titles, right tools, right number of years. Then the person starts and can't actually do the work. That gap between how someone looks on paper and how they perform is where most hiring money disappears.

A founder we spoke with recently had leaned on a cheap monthly-retainer agency for years. His best engineer was buried in admin, and this was supposed to take it off his plate.

Three hires came and went. One lasted about a year, another six months, and one quit by email after a single day. Language was never the issue. The role needed judgment and a high tolerance for detail, and both kept getting missed, so his engineer spent his days re-checking someone else's work instead of doing his own.

The retainer billed the entire time, including the empty stretches between hires. Every one of those hires was chosen on paper and price. None were chosen on proof.

"Show me they can do the work before I commit."

Good. That's the right instinct, and it's how we think about it too. We don't lead with resumes. We test the work. Before anyone reaches a client, we verify their history and run them through tasks that mirror the real role. What you see is proof, not a promise.

We recently vetted an operations coordinator in Johannesburg who runs an entire hiring-to-onboarding pipeline. She builds the SOPs new team members are trained on, automates her own deadlines so nothing slips, and keeps leadership current with a single daily check-in. Enough to stay informed, never enough to drown in detail.

When she spotted a bottleneck in the sales process, she didn't flag it and wait. She mapped it, researched the fix, and rolled it out.

Her rate is $2,000 a month, which is what that seniority costs in her market. Senior judgment at a junior price, and exactly the profile that ends the churn.

So the incentive is a hire that lasts rather than a retainer that bills whether the seat is filled or not.

When the match is built on demonstrated work instead of a well-written PDF, you stop gambling on first impressions. If you've been burned by candidates who looked perfect and performed poorly, the answer isn't reading more resumes. It's changing what you evaluate.

→ Book an obligation-free call with our recruitment specialists 

🧲Want your newsletter to book sales meetings for you?
Bobb turns your newsletter into your #1 sales channel. It finds qualified leads, writes and sends the emails, and books calls with your most engaged readers, all without cold email.
→ Start free here

🤝 Want to hire better people with less risk?
Go Carpathian sources top-tier global talent from world-class companies, screens them for skill and fit, and hands you candidates who are ready to perform from day one. Check out how they can help you hire smarter, faster, and 70% cheaper than U.S. rates.
→ Speak to a recruitment specialist