Owning the Checking and Memory
Milan@freshperspected
+1.0%
since Oct 16
The idea
AI has made producing software and text cheap, but not knowing if it's right. I'm betting value shifts to whoever holds a company's own records, the real-world data models can't fabricate, and the tools that check AI output, so the take centers on companies owning business context and trusted data, with smaller slices in AI monitoring and independent work platforms. I deliberately avoid AI model and chip makers and pure database storage, spreading broadly with some cash held back for a ten-year view.
What it owns
- These companies hold the detailed records of how businesses actually operate, their IT systems, customer relationships, finances and people, which is the specific context an AI agent needs to be useful within a particular organization. · ServiceNowIt holds the detailed record of how a company's IT, HR and service operations actually run, which is the company-specific context agents need to be useful inside an organization., SalesforceIt owns years of customer history including deals, calls and service tickets, which is the memory an agent needs to sell or support customers effectively within a single company., WorkdayIt keeps the people and finance records companies cannot afford to get wrong, and it recently announced tools to govern what AI agents are allowed to access within the system., Veeva SystemsIt holds the clinical, regulatory and sales records drug companies are legally required to keep correct, combining both company context and the need for verification., Tyler TechnologiesIt runs court, tax and property records for local governments that need records which are accurate and auditable, and slow-moving government buyers tend to keep such systems., Guidewire SoftwareIt is the core policy and claims system for many property and casualty insurers, holding each insurer's operating history and business context.38%
- They own trusted data drawn from the real world that AI models cannot produce on their own, from historical claims and property records to credit ratings and market data that markets and lenders rely on as a reference point. · Verisk AnalyticsIt pools decades of claims and property risk data from across the insurance industry, which is the kind of real-world data that AI models cannot generate., S&P GlobalIts credit ratings and indexes are a stamp of trust that financial markets rely on, built from data and judgment that models cannot simply generate., Fair IsaacIt operates the trust standard US lenders use to decide who gets credit, which is exactly the kind of scarce trust this take values., Intercontinental ExchangeIt owns exchanges, market data and mortgage records that other systems check themselves against, serving as the reference point for traders and lenders., CoStar GroupIt gathers and verifies property data through people on the ground, providing contact with reality that distinguishes it from scraped or generated information., Thomson ReutersIt owns verified legal content and builds AI tools on top of it, giving lawyers a trusted source to check against as AI produces more legal drafts.32%
- As careers fragment into multiple clients and fractional roles, solo workers and small teams need back-office tools to manage finances and operations, plus marketplaces where they can find clients and connect with other specialists. · IntuitIt runs the back-office tools and financial records for solo and small businesses, capturing their financial history in a central place., UpworkIt is where companies hire skilled people for discrete projects, matching the picture of small teams directing agents and pulling in experts as needed., FiverrIt is a marketplace for freelance work aimed at solo and small-team workers, where people offer their taste and judgment to complement AI output.10%
- As AI systems produce more code and decisions, someone has to verify whether they're working correctly, and these platforms watch software systems running in production to catch when agents malfunction or behave unexpectedly. · DatadogIt watches software running in production, which increasingly includes AI systems whose behavior and output need to be monitored for correctness., DynatraceIt tracks whether large, complex software systems are working and why they break, which becomes harder to see as agents write more code.9%
- As digital content becomes cheaper and more abundant, spending shifts toward live experiences shared physically with other people, where presence and direct interaction create value that recordings cannot replicate. · Live Nation EntertainmentIt sells the thing this take says gets scarcer as digital content gets cheaper: being physically present with other people at live events.3%
- Cash8%