the Now & the Next

OpenAI launched a dedicated Deployment Company — a $4 billion entity backed by 19 global investment firms, consultancies, and system integrators — whose explicit mission is to embed AI systems inside enterprises so completely they function as workforce, not tooling. A new job category appeared alongside it: the “forward-deployed engineer,” a human whose entire role is to integrate AI into client operations at depth. The same week, Conagra expanded Microsoft Copilot enterprise-wide for real-time supply chain insights. Levi’s announced a super-agent system with Microsoft that orchestrates specialized sub-agents across the full enterprise from a private cloud. Uber disclosed it runs 60,000 AI agent tasks per week with 1,500 active agents monthly — and that over 90% of its 5,000 engineers use AI tooling daily. The deployment era has a corporate structure, a funding round, and a job title.

U.S. bank regulators paused cybersecurity examinations of major financial institutions — including JPMorgan Chase, Goldman Sachs, and Morgan Stanley — while those banks establish dedicated internal teams to work with Anthropic’s Claude Mythos model, an AI system explicitly designed to identify and exploit security vulnerabilities. OpenAI launched Daybreak the same week, a competing AI cybersecurity platform that automates vulnerability detection and patch validation for enterprises and government agencies. President Trump signed an executive order on AI cybersecurity, directing federal agencies to evaluate AI systems before deployment. The U.S. Marine Corps mandated that all personnel complete foundational AI training by year-end. The adversarial AI arms race is now operating inside regulated institutions — and regulators are stepping back to let it run.

In April 2026, Meta, Oracle, Amazon, PayPal, Block, and Atlassian cut 33,361 technology jobs — 26% of all tech sector cuts that month. The same companies announced record AI data center investments. Amazon, Microsoft, and Oracle are simultaneously reducing workforce and pouring capital into AI infrastructure and semiconductor chips — reallocating payroll to compute. California Governor Gavin Newsom signed an executive order to prepare the state’s workforce for AI disruption, launching a public feedback platform called Engaged California. SHRM published a major analysis asking the question executives are avoiding: are “AI layoffs” genuine transformation, or are companies using AI as a scapegoat for cuts they were planning anyway? The answer requires accounting that nobody has done.

At the Trump-Xi Geneva summit, the United States approved the sale of Nvidia’s H200 AI chips to roughly ten major Chinese technology companies — including Alibaba, Tencent, and ByteDance — as part of broader diplomatic negotiations. The same week, the U.S. government announced an $8.9 billion investment in Intel to secure domestic chip manufacturing capacity, explicitly for Apple’s production needs. CSIS published analysis showing that Section 232 tariffs on steel, aluminum, and copper are adding significant construction cost to AI data center buildouts across the country. SpaceX filed for IPO recasting itself as an AI infrastructure giant, referencing a proposed “Terafab” chip manufacturing initiative with Tesla and Intel. Semiconductor access has entered the same geopolitical grammar as oil in the 1970s.

OpenAI, Google, and Microsoft are shifting their business models from selling large language models as standalone products to embedding AI infrastructure directly within customer organizations, billed by token consumption. Anthropic beat OpenAI in business AI adoption for the first time — driven almost entirely by Claude Code, its agentic coding tool, which became the fastest-growing product in the company’s history. Open-weight AI models are gaining ground on proprietary LLMs, with IT decision-makers choosing smaller, customizable models over locked-in enterprise licenses. SAP announced a broad rollout of agentic AI tools for CFOs covering cash management, tax, financial planning, and billing. VentureBeat reported that the real enterprise battle is no longer about which model is best — it is about which company controls the “agent control plane” — the governance layer that manages AI agent fleets across an organization.

Google DeepMind and Edison Scientific — a spinout from FutureHouse — launched Robin, an AI scientist platform that proposes and runs its own research experiments using OpenAI’s o4-mini and Anthropic’s Claude 3.7 as the underlying models. Bristol Myers Squibb deployed Claude enterprise-wide, specifically to accelerate drug R&D and reduce the lag between clinical data locks and regulatory filings — a step that can currently add months to drug development timelines. OpenAI launched OpenAI for Healthcare, a HIPAA-compliant clinical workflow suite. A landmark study published in Science demonstrated that an OpenAI o1-preview model outperformed emergency room physicians on clinical reasoning tasks using real ER records. The productivity impact of AI is landing first on the most credentialed, most specialized, most protected workforces — not the ones everyone predicted.

David’s Bridal joined Shopify’s Agentic Storefronts initiative, enabling product browsing by silhouette on ChatGPT and collection viewing on Microsoft Copilot — products designed to be discovered and selected by AI agents acting on behalf of human shoppers. Amazon restructured its affiliate program in ways publishers are calling alarming, shifting referral incentives toward AI-driven product recommendations that bypass traditional content intermediaries. The Economist is experimenting with simplified, agent-readable content formats specifically for ChatGPT, Gemini, and Claude. Larroudé’s CEO built an AI system that automatically integrates the company’s Shopify platform with factory data, inventory, site code, and marketing performance — a lean shoe brand now operating at operational speeds only AI can sustain. The storefront is bifurcating: one for humans, one for the agents browsing on their behalf.