10 carefully selected reads across AI, business, and investing.
The AI industry is shifting from an era of unchecked spending and broad experimentation toward a focus on operational ROI and infrastructure constraints. Enterprises are moving away from 'tokenmaxxing' to scrutinize agentic costs, while the race for power infrastructure and specialized hardware highlights the physical limitations of scaling. Meanwhile, the narrative is hardening around AI as an augmentation tool for human expertise rather than a wholesale replacement.
Enterprises like Salesforce and Uber are transitioning from unchecked spending on AI coding agents to a more disciplined, ROI-focused approach. Initial token budgets have proven to be significant underestimates, forcing companies to move beyond early hype toward sustainable engineering economics.
Source: NewcomerSnowflake's 1Q26 results demonstrate that Cortex Code is acting as a critical revenue accelerator by absorbing non-data workloads. The product's adoption is outpacing historical benchmarks, signaling strong demand for integrated infrastructure-based AI solutions.
Source: FundaAIData from B2B support platform Pylon reveals that AI is functioning primarily as an invisible triage agent rather than an end-to-end replacement. By handling noise and routing complex tickets to humans with context attached, AI tools are increasing efficiency and reducing human workload by a third.
Source: a16z NewsDigitalBridge's $1 billion acquisition of ArcLight underscores the critical convergence of AI data center development and power infrastructure. As electricity access becomes the primary bottleneck for AI, major infrastructure firms are bringing power expertise in-house to secure a competitive advantage.
Source: Data Center RichnessCognition’s recent $1 billion Series D round highlights the growing momentum behind fully autonomous, asynchronous developer agents. While early AI tools focused on in-the-loop assistance, the market is shifting toward agents capable of executing full spec-to-PR workflows in the background.
Source: Latent SpaceInitial testing of Anthropic’s Opus 4.8 shows strong performance in greenfield prototyping and fast execution but reveals persistent struggles with edge cases in existing codebases. The model remains a powerful tool, yet still faces challenges with deep-context maintenance and hallucinations.
Source: Lenny's NewsletterAnthropic’s recent tokenizer update functions as a silent 30–50% price increase, a strategy that highlights the need for companies to move toward model-agnostic workflow architectures. This hidden cost adjustment poses a significant challenge for enterprise budget planning based on previous token consumption rates.
Source: DIGITAL STORMOpenAI's recent achievement in disproving a long-standing geometry conjecture is a significant milestone but not a radical AGI breakthrough. The model effectively utilized existing mathematical techniques, confirming a future where AI and humans collaborate to grind through complex proofs.
Source: Understanding AIMarvell's recent earnings report highlights the sustained strength of the AI interconnect business despite aggressive market expectations. The company is seeing significant growth in its 1.6T DSP offerings, reinforcing its position as a vital player in the AI supply chain.
Source: FundaAIFor banks, AI models are now considered 'table stakes' rather than a source of competitive advantage. Durable moats will only emerge from deep integration into the total customer journey and the development of proprietary, hard-to-replicate operational workflows.
Source: Rich Turrin