Anthropic’s new research maps three wildly different futures for the AI economy—and the future of knowledge work

Anthropic outlined AI-economic scenarios ranging from internet-like gradual gains to 15% annual GDP growth and severe job displacement if AI automates nearly all knowledge work, but stressed that its model excludes demand shocks, business cycles, financial disruptions, and policy responses. Current adoption data point to a slower outcome: Ramp found business AI adoption rose only 0.4 percentage points in August to 56.1%, while spending on AI tools is falling and productivity gains outside coding agents remain limited. Economists Ben Moll and Alex Imas challenged double-digit growth projections, citing adoption speed, demand, cyber-risk, and automated AI research as major hurdles.

Analysis

The relevant market signal is not model capability but the widening gap between infrastructure investment and enterprise workflow monetization. That gap favors hyperscalers with diversified cash flows and proprietary distribution—GOOG and META—over AI software vendors whose valuations embed rapid seat expansion. Near term, slower non-coding adoption raises the risk that corporate AI budgets are reallocated from fragmented point tools toward bundled offerings inside Google Workspace, Microsoft 365, and Meta’s consumer/advertising ecosystem.

For META, agent launches matter primarily if they improve advertiser conversion, creative throughput, or support costs; consumer engagement alone is unlikely to justify a material earnings re-rate. The more important second-order implication is pricing power: if AI lowers content and creative-production costs, smaller advertisers can bid more aggressively for inventory, supporting auction density, while the same dynamic may increase content-quality, safety, and compute costs. GOOG is relatively insulated because search and cloud can monetize AI through distribution and usage, but its multiple remains exposed if AI answers cannibalize high-margin query volume faster than commercial-intent monetization evolves.

Over 1-3 months, evidence of weak broad deployment should compress revenue multiples for application-layer AI and private-market benchmarks rather than derail capex leaders immediately; enterprise contracts and depreciation schedules delay the earnings impact. Over 6-18 months, the key risk is a capex digestion cycle if incremental AI revenue and productivity savings fail to cover rising inference costs. A major agent-security incident would accelerate governance requirements and favor scaled platforms with compliance tooling, while impairing autonomous-agent adoption.

Consensus still treats slower adoption as bearish for all AI exposure. That is incomplete: gradual diffusion can entrench incumbents because integration, data governance, and distribution—not raw model quality—become the bottlenecks. The falsifier is sustained acceleration in non-coding AI spend and measurable enterprise productivity gains, which would reopen upside for specialized software and validate current infrastructure spending.

Key Decisions for Investors

• Maintain a 3-6 month quality tilt: long GOOG versus a basket of high-multiple application-layer AI software (IGV proxy). Rationale: GOOG can bundle AI into existing distribution and absorb a slower adoption curve; the short leg is vulnerable to deferred seat growth. Reassess if broad enterprise AI spend accelerates for two consecutive monthly reads.
• Treat META as a catalyst watch rather than a standalone agent trade: add only ahead of earnings if management quantifies AI-driven ad conversion, incremental auction demand, or support-cost savings. Target a 2:1 upside/downside setup; exit if AI expense growth materially outpaces ad-revenue acceleration for two reporting periods.
• Do not initiate a directional RAMP position on the spend index alone. Use its monthly AI-spend data as a leading indicator: a further deceleration in adoption breadth supports reducing exposure to pure-play AI application vendors; a reacceleration outside coding is the key falsifier.
• Hedge 6-18 month AI monetization risk with selective downside on semiconductor/infrastructure exposure rather than selling GOOG/META outright. The trigger is evidence that hyperscaler capex guidance remains elevated while disclosed AI revenue, cloud consumption, or productivity benefits fail to improve; this would raise free-cash-flow and multiple-compression risk.