TLDRocket
3 August 2026
The day's dominant thread is the maturation of agentic AI—systems that autonomously execute multi-step tasks—revealing a fundamental split in how enterprises will deploy these tools. Palantir's Alex Karp crystallized one side of the debate, accusing frontier AI labs of using customer data to build competing businesses while his company's $1.9 billion quarterly revenue (up 93% year-over-year) reflects enterprise appetite for keeping AI infrastructure in-house. AWS and Superblocks announced a multi-year partnership positioning AI within private clouds rather than relying on external labs, while DeepSeek's V4-Flash—a smaller, open-weight model outperforming its own flagship on agent tasks—underscores how post-training can replace raw scale, giving organizations control over deployment. Alibaba's Qwen3.8-Max meanwhile executed 16 days of autonomous coding across 265 GitHub commits, extending the range of what these systems can sustain unsupervised, though the 2.4 trillion parameter model remains impractical for most to self-host. Meanwhile, Formula 1 compressed data engineering workflows from weeks to 40 minutes using agentic AI on AWS, and Microsoft Research released Orchard, an open-source framework that achieves near-parity with vastly larger systems through smarter training. The subtext: enterprises now see autonomous agents as infrastructure to control, not services to consume. This matters because it reshapes the power dynamic between cloud providers, frontier labs, and customers. The EU's AI Act transparency rules taking effect August 2nd adds a regulatory layer—companies must now explicitly label AI interactions rather than discretely embedding them. Beneath the practical wins, though, lurks a mundane problem Niklas Gruhn named: people becoming "meat proxies," passively relaying AI output without processing it. True value requires critical engagement, not copy-paste.
Read the full briefing →
