Goldman Sachs Global Institute co-head: remaking our world for machine intelligence

The article argues that rapid AI/LLM “agent” progress will drive new physical and software infrastructure—e.g., drone delivery workflows (Zipline) and “headless” API-first software—shifting how businesses design retail, logistics, and online commerce. It highlights potential efficiency gains (e.g., replacing human-centric interfaces with agent-optimized APIs/JSON schemas) and knock-on effects like more dynamic pricing and auction-like procurement at scale, but frames adoption as nascent with major coexistence/uncertainty.

Analysis

The market consequence is less about “AI hype” and more about where economic rent migrates. If agents become the primary users, value shifts from polished interfaces to machine-readable rails, telemetry, and durable data access; that compresses pricing power for human-centric workflow software and favors platforms that sit on the fulfillment or transaction layer. In that framework, BOX and HUBS are more exposed than the market may appreciate because their product differentiation narrows once an agent can traverse multiple tools without caring about UX.

The more interesting winners are the operators with physical density and orchestration leverage. WMT can absorb machine-native workflows into stores, inventory, and last-mile staging, turning capex into throughput rather than just foot traffic; UBER benefits if agents increasingly optimize rides, delivery, and dispatch in real time because it is a control layer, not a destination. DASH is more ambiguous: agents could expand order volume, but they also make customer loyalty more price-sensitive and restaurant take-rates harder to defend, which is a margin-risk story, not a growth story.

This is a multi-horizon theme, but the tradable window is still limited. Over the next 1-3 months, the main catalyst is sentiment rotation and product announcements; the hard falsifier is whether enterprises actually route material spend through agent-native workflows or merely pilot them. Over 6-18 months, the thesis breaks if authentication, payments, governance, and bot-defense friction keep agent traffic from becoming economically meaningful at scale. Consensus may be too quick to extrapolate the long-run architecture change while underestimating how much of current software and commerce economics are still protected by human friction.

Key Decisions for Investors

• Long UBER / short HUBS as a 3-6 month relative-value pair: UBER monetizes orchestration and physical-world dispatch, while HUBS is more vulnerable to interface commoditization and pricing pressure; target 10-15% relative upside, stop if HUBS re-accelerates net revenue retention or UBER take-rate compresses.
• Long WMT / short DASH for 6-12 months: WMT can re-engineer store and logistics infrastructure for agent-driven fulfillment, while DASH faces a higher risk of auction-style price competition and weaker customer stickiness; prefer adding on market weakness rather than chasing a rally.
• Sell rallies in BOX and HUBS into AI-agent enthusiasm unless management can prove durable seat pricing or usage-based monetization; the risk/reward skews against names whose interface layer becomes optional.
• No fresh long in AMZN, GOOGL, or NET solely on this thesis; they already embed substantial AI optionality, so wait for evidence in usage, take-rate, or security monetization before paying up.