5 carefully selected reads across AI, business, and investing.
The enterprise landscape is undergoing a structural shift as AI agents transition from auxiliary tools to primary orchestrators of work. Legacy systems of record are increasingly relegated to backend infrastructure, while real-time, agentic workflows capture the user interface. This pivot is mirrored by a broader industry move toward specialized hardware and metered programmatic access as AI labs refine their business models.
Enterprise software is shifting from being a static database of record to an input for autonomous agentic workflows. Similar to how the news feed eclipsed the friend graph in social media, AI orchestration layers now treat traditional CRMs as backend infrastructure. Employees are increasingly managing their day through agent collections that automate research, communication, and data entry.
Source: a16z NewsThinky is debuting 'Interaction Models' designed to move beyond turn-based AI toward simultaneous full-duplex communication. By using a micro-turn architecture that processes data in 200ms chunks, the system mimics natural human conversation. This approach aims to reduce response latency to 0.4 seconds, significantly improving the utility of AI in physical and social real-world environments.
Source: AI SupremacyAnthropic has shifted its pricing model, linking subscription plans to API credits for programmatic usage. This move effectively ends the era of heavy subsidies for third-party AI harnesses like OpenClaw. As labs solidify their internal agent ecosystems, the market is seeing a widening divide between proprietary platform tools and independent developer-focused challengers like Codex.
Source: Latent SpaceStitch is addressing the 'infrastructure debt' of legacy core banking systems by building an API-first operating system for financial institutions. By providing a clean foundation, the platform enables banks in emerging markets to bypass outdated tech stacks and adopt AI-native features like automated loan origination. The firm saw 10x customer growth in 2025 as regional financial ecosystems modernize.
Source: a16z NewsHardware requirements for unsupervised self-driving vehicles are scaling rapidly, with startup Tensor targeting 8,000 TOPS of compute per vehicle using Nvidia Thor GPUs. This compute density presents a significant cost hurdle for individual car ownership. While companies pursue consumer-grade autonomous sales, existing legacy hardware in consumer vehicles remains largely insufficient for full autonomy.
Source: Understanding AI