Software 3.0 Picks and Shovels
Clay Norris@claynorris
−4.8%
since Oct 1
The idea
I believe natural language is becoming the primary programming interface, with agents writing more and more of the code that ships. Since the pure plays are mostly private, this take owns the public picks and shovels: platform owners that control the entry points, infrastructure agents lean on as they run, and tools priced per unit of code shipped, monitored, and secured. It's built broad and long horizon, large platforms as ballast, mid-cap infrastructure as the core, smaller bets on the expanding pool of people coding by describing what they want, with a cash buffer held aside.
What it owns
- Agent-written code multiplies what needs monitoring and securing, so tools priced per unit of code win even as seat-priced tools bleed. · DatadogDatadog bills per unit of software monitored, so a flood of agent-written code directly multiplies its revenue., CrowdStrikeAgent-written code expands attack surface dramatically, and CrowdStrike sells the security layer defending it., DynatraceDynatrace watches complex systems and increasingly fixes them automatically, which matters far more when no human wrote the code., OktaEvery agent acting on someone's behalf needs an identity and permissions, turning machine identity into a paid product line., JFrogEvery build an agent ships still moves through an artifact registry and supply chain check, which JFrog controls and prices on volume., SentinelOneSentinelOne is the smaller, AI-first security bet on the premise that more machine-written code means more surface to secure.26%
- Whoever owns the repos, editors, and models controls where agents enter the stack, and that control is hard to dislodge. · MicrosoftGitHub is where code lives and Copilot is how agents write it, making Microsoft the default front door for agent-native development., GitLabEvery pipeline run and code review in GitLab becomes a billable event, scaling with agent-written code volume., AlphabetAlphabet owns the model layer with Gemini and the cloud infrastructure agents run on, controlling a core part of the stack., AtlassianAtlassian owns Jira, where dev work gets planned and tracked, the coordination layer agents will need to plug into.24%
- Agents need memory, context, edge compute, and APIs to run, and these companies sell exactly that by the unit. · CloudflareCloudflare Workers provides cheap, fast compute at the edge priced per unit of work, the natural runtime for agent execution., MongoDBAgents need memory for context, and MongoDB's document and vector database is positioned as the standard store for that state., ElasticElastic sells the search and vector retrieval that feeds agents the context they need to become more useful., DigitalOceanMost newcomers to development won't sign enterprise cloud contracts; DigitalOcean is the simple cloud that new builders reach for., TwilioIt provides the API infrastructure that turns communications like SMS and voice calls into a few lines of code, positioning it to capture volume growth as AI agents become callers alongside humans.23%
- Every agent-written line of code runs on AI compute, the deepest picks-and-shovels layer of the whole story. · NVIDIAAgents are made of inference, and inference runs overwhelmingly on NVIDIA chips across the industry., BroadcomBroadcom builds the custom AI chips that hyperscalers design to cut their NVIDIA bills, so it wins on compute demand either way.12%
- If the developer count expands massively, the tools non-coders use to describe and ship software capture the new arrivals. · FigmaFigma is where non-coders already describe what they want visually, making design-to-code a natural on-ramp for new builders., Wix.comWix lets people describe what they want and get working software back, aimed at the tens of millions of new builders expected.7%
- Cash8%