Agentic Autonomy Worldview
Clay Norris@claynorris
−3.3%
since Oct 2
The idea
I believe AI agents are about to become a dominant source of traffic and economic activity, with compute staying scarce and orchestration layers capturing durable value. This basket tilts toward the underpriced plumbing agents will run on, identity, payments, APIs, and edge infrastructure, plus companies rethinking the cost of expertise like healthcare and legal services, while holding compute and model orchestration as steadier ballast.
What it owns
- If agents become the dominant form of internet traffic, the infrastructure companies routing that traffic and verifying agent identity will collect tolls on every transaction and interaction. · CloudflareIt runs the edge network agents will drive traffic through, and is building tools like bot verification and pay-per-crawl specifically for agent interactions with the open web., OktaIt provides verifiable identity infrastructure, and machine identity could become as large a market as human login if agents need neutral-layer verification to act on behalf of people., VisaIt processes payments and is building agentic commerce credentials, so it collects tolls on every transaction agents execute on their owners' behalf., TwilioIts communication APIs are the surfaces agents will call instead of clicking buttons, making it infrastructure for agent-driven product redesign., ShopifyIt is wiring its merchants into agent checkout protocols, so if agents become the primary way people buy, its merchant base is already reachable by them., FastlyIt is a smaller, cheaper way to own the edge compute layer agent traffic runs through, with upside if that layer gets repriced as agent volume grows.27%
- AI can deliver healthcare, legal and education services at a fraction of traditional expert cost, and new entrants and existing platforms are building that capability at scale. · Hims & Hers HealthIt delivers subscription healthcare with AI-assisted clinicians at a fraction of traditional doctor visit costs, proving the cheap-expertise model works at scale in medicine., DuolingoIt replaces human tutors costing fifty dollars per hour with AI instruction at near-zero marginal cost, applying cheap-expertise disruption to education if AI features keep lifting paid subscriber growth instead of commoditizing the product., LemonadeIt replaced traditional adjusters and underwriters with AI from inception, proving the cheap-expertise model works in insurance if AI can actually price and assess risk accurately., IntappIt sells AI workflow software to law and professional services firms where expertise commands the highest price and AI can compress that cost most dramatically., Teladoc HealthIt operates at scale in telehealth and can use AI triage and diagnosis to cut its largest cost, clinician time, proving cheap expertise works if AI makes visits cheaper to deliver.19%
- Advanced AI chip manufacturing and the infrastructure to power and cool data centers remain constrained, and companies controlling those layers benefit as long as demand for compute outpaces supply. · NvidiaIt manufactures the advanced chips powering AI training and inference, and demand for those chips outpaces supply as long as compute remains the bottleneck., Taiwan SemiconductorIt manufactures nearly every advanced AI chip at the cutting edge, making it the chokepoint under the compute-scarcity hypothesis if leading-edge production stays concentrated and scarce., BroadcomIt manufactures custom AI chips for hyperscalers trying to escape compute scarcity, so it wins regardless of whether that escape attempt succeeds or fails., VertivIt supplies the power and cooling infrastructure for data centers, and compute scarcity means more data centers need to be built and maintained.18%
- As AI models become increasingly interchangeable, the platforms that route work across multiple models and maintain direct relationships with customers capture disproportionate value. · MicrosoftCopilot sits inside the productivity tools hundreds of millions of workers already use, making Microsoft the default orchestration layer for enterprise agents., PalantirIts AIP platform is a control layer for deploying agents inside enterprises and governments with guardrails, exactly the orchestration layer that captures durable value., ServiceNowIt already routes workflows for large companies, making it the natural place to plug in agents to actually execute the work they orchestrate., SalesforceIt is pushing Agentforce through the largest customer-data footprint in enterprise software, betting that orchestration layered over distribution wins in the agent era.17%
- Agents perform only as well as the data and memory they can access, so the platforms storing and serving that context efficiently capture value from AI workloads. · MongoDBIts database with built-in vector search is positioned as the memory and context store for AI applications that agents depend on to reason effectively., SnowflakeIt stores the data enterprises will need their agents to reason over, and its AI query products keep that data valuable to access and reason on rather than obsolete.8%
- Cash10%