Violence against Sam Altman, the 'AI populism' debate, and Block's new AI org chart in practice

A disturbing week for the AI industry's public profile. Multiple attacks on Altman's home spark a real conversation about rhetoric and responsibility. And the agentic 'harness' is becoming the strategic battleground.

A serious and unsettling week. The violence directed at AI executives is the kind of thing that should force the industry to confront its public messaging. The technical news is overshadowed but still worth tracking.

Multiple violent attacks on Sam Altman's home

My take: Including a Molotov cocktail and gunfire. An arrested suspect had an anti-AI manifesto and a list of other AI executives. This is a grim moment for the industry. The follow-on debate about whether AI safety rhetoric or job-displacement messaging contributed is uncomfortable but important. If you communicate publicly about AI, the words you use matter.

75% of executives admit their AI strategy is 'more for show'

My take: From a Ryder study. Combined with KPMG data showing over 50% of large companies have agents deployed, the picture is: real deployment is happening, but executives privately know it's superficial. This is the gap to close. The companies that move from theater to actual capability in 2026 will pull ahead permanently.

'Harness Engineering' becomes the new strategic focus

My take: The 'harness' is the system of tooling, memory, and environment around a model. Anthropic launched Managed Agents specifically to provide enterprise-ready harnesses. Translation for you: the model itself matters less than the surrounding infrastructure. If your AI tooling is just a chat box, you're missing 80% of the value.

Meta's Muse Spark released, China's GLM 5.1 open-sourced

My take: Meta is back in the model conversation with Muse Spark, a multimodal model strong on visual understanding. Chinese lab Z.AI open-sourced a 754B parameter model that claims to beat both GPT-5.4 and Opus 4.6 on coding. The open-weight competition is real. If you have compliance or sovereignty constraints, the open-weight options are now genuinely competitive.

GPU rental prices up 48% over two months

My take: The compute crunch is now a serious cost factor. Power availability through 2026 is already spoken for. If your business is doing serious model training or running heavy inference workloads, lock in capacity contracts now. The spot market is going to keep getting worse before it gets better.

GitHub on pace for 14 billion code commits this year

My take: Up from 1 billion last year. That's a 14x growth driven almost entirely by agentic coding. Two implications: code review processes need to scale dramatically (humans cannot review 14x more code with current tooling), and infrastructure providers (including GitHub itself) are under strain. Expect more outages.