Software Engineering & Architecture
IBM Think 2026 delivered what the company calls a "Blueprint for the AI Operating Model" — a framework where AI agents are embedded into every layer of the enterprise stack, from IT ops to developer pipelines. IBM's data shows a widening "AI divide" between companies that have restructured around agents and those still treating AI as a productivity bolt-on, and the gap increasingly maps to engineering culture: teams with platform abstractions that expose clean APIs to agents are accelerating; those without are stalling.
Think 2026: IBM Delivers the Blueprint for the AI Operating Model
Global Macro & Markets
Markets are pricing in the Trump-Xi summit planned for May 14–15 in Beijing, with Polymarket showing near-even odds (50.5% "No") on a US-China tariff agreement by May 31. The current arrangement — tariffs reduced to 10% bilaterally through November 2026 under last year's deal — has held, but the summit is being watched as a potential catalyst for extension or escalation. This plays out against a backdrop of record US equity indices and continued Strait of Hormuz disruption keeping oil near $120/barrel.
US x China tariff agreement by May 31? Predictions & Odds
Enterprise AI & AI Strategy
A report released Monday finds 46% of enterprise AI initiatives have failed to meet expectations despite 74% of organizations increasing investment — with only 29% reporting significant ROI from generative AI. The gap between "frontier" companies (deploying 3.5x more AI per employee) and laggards is structural: most organizations are still buying access rather than embedding AI into operational workflows. Separately, Capgemini today announced an investment in OpenAI's Deployment Company, joining 18 other global firms in the new majority-OpenAI-owned enterprise deployment vehicle.
Enterprise AI Is Stalling: 46% of Initiatives Fall Short Despite Rising Investment
AI Tools & Developer Utilities
Anthropic's Code with Claude conference (SF last week, London up next on May 19) shipped three features that reframe what agent infrastructure means: Multiagent Orchestration for parallelizing work across agent fleets, Outcomes for defining measurable success criteria so agents self-correct, and Dreaming — a between-session process where agents review their own prior work and write new memory entries for future sessions. The telling signal: Anthropic released no new models, explicitly framing the bottleneck for production agent systems as infrastructure, not model capability.
Code with Claude 2026: 5 New Agent Features Anthropic Just Shipped