Daily briefing
Monday, 3 August 2026
The throughline today is about control—who owns it, who keeps it, and why it matters. Palantir CEO Alex Karp crystallized the emerging tension by attacking frontier AI labs during an earnings call, accusing them of seeking to capture customer data and intellectual property while Palantir itself posted $1.9 billion in quarterly revenue, up 93 percent year-over-year. His critique speaks to a broader industrial shift: enterprises are tightening their grip on AI infrastructure. AWS and Superblocks announced a partnership keeping customer data on private clouds rather than frontier labs' servers. DeepSeek released V4-Flash under an MIT license, reducing switching costs and letting organizations deploy without dependency. Even the EU moved decisively, with transparency rules now in effect requiring explicit labeling of AI use rather than leaving it to company discretion.
Simultaneously, the capability side is accelerating in unexpected directions. Alibaba's Qwen3.8-Max conducted 16 days of autonomous coding across 265 GitHub commits—real infrastructure proving ground, not benchmark theater. Microsoft Research released Orchard, an open framework for training agentic systems that hit 69.7 percent on SWE-bench with only 3 billion parameters, suggesting efficiency gains are materializing faster than expected. Formula 1 compressed data onboarding from six weeks to 40 minutes using autonomous agents on AWS. Meanwhile, researchers demonstrated a self-replicating AI virus achieving 37 percent attack success, and 1,337 AI lab employees petitioned the US government for tools to deliberately slow development—a rare unified signal that internal competitive pressure prevents unilateral restraint.
But there's friction in the adoption story. Developer trust in AI tools dropped from 40 to 29 percent even as usage rose from 76 to 84 percent. Apple released an improved Siri that felt anticlimactic precisely because AI has moved beyond functional chatbots into multi-step reasoning and autonomous work. The real challenge emerging isn't capability or control but integration: how to build governance, trust, and cultural processes around systems that behave unpredictably. Mend.io published a security framework acknowledging traditional application security fails when agent behavior emerges from model-prompt-tool interactions. The market is splitting into layers: frontier labs generating models, infrastructure providers hosting them, enterprises retaining their data, and middleware companies like Palantir bridging the gap. The winner may not be the fastest model but the company that lets enterprises keep their data while moving fast.
Top stories from this issue
Don't be a meat proxy
Simon Willison's Weblog · 1 month ago ·
16
White House invites AI companies to review its new AI safety framework
SiliconANGLE · 1 month ago ·
33
Tokenomics: Why making AI pay is tricky
BBC · 1 month ago ·
46
After killer quarter, Palantir CEO Alex Karp calls AI industry ‘Marxist’
TechCrunch · 1 month ago ·
4
Netflix co-founder backs $312M round for optical inference appliance maker Olix
SiliconANGLE · 1 month ago ·
36
Evaluating Multimodal Vision Models with Moonshot PerceptionBench Using Robust Data Loading and Automated Judging
MarkTechPost · 1 month ago ·
46
US company’s AI lets Ukraine’s cheap kamikaze drones track targets on their own
Ars Technica · 1 month ago ·
19
Apple is getting this wrong
OpenAI · 1 month ago ·
35
Stynar
Product Hunt · 1 month ago ·
20
The Inference Engineering Masterclass — Philip Kiely & Ali Taha, Baseten
Latent Space · 1 month ago ·
28