The Harness Layer Wins
Davis@flatirons_davis
+1.7%
since Oct 8
The idea
I believe AI models become cheap and commoditized, so value shifts to the harness layer built on top of them, and that agents acting at scale will upend an ads model built on human screen time. This basket anchors in Google and Meta, leans into video and audio where advertisers can trust a human is watching, and backs doubted ad tech names with room to grow if assistants open ad slots to outside buyers. The other half owns sticky industry software with hard-to-leave customers, proven tech builders in traditional businesses, and a smaller slice of harness infrastructure.
What it owns
- Software companies that own years of industry data and customers locked in by switching costs have the two things a strong harness needs to succeed. · Veeva Systems Inc.It runs the core software drug companies use for sales and clinical trials, so leaving it is painful, giving it the sticky customers and industry data a strong harness needs., Guidewire Software, Inc.It runs the policy and claims systems for many insurers, which are slow and costly to swap, and sits on years of insurance data that AI tools can build on., Tyler Technologies, Inc.It runs software for local governments, customers that rarely switch and can't hire talent to build their own harness, so they would likely buy one from the company they already use., Thomson Reuters CorporationIt owns Westlaw and widely used tax and accounting tools, the kind of content plus customer base that makes a strong harness even if startups build slicker products., RELX PLCIt owns LexisNexis plus big scientific and risk data sets that customers have relied on for decades and that AI tools need to work well.25%
- Ordinary businesses that have already shown they can build software and hire engineers for it can build their own harnesses instead of buying them. · JPMorgan Chase & Co.It is one of the few regular businesses that has already shown it can build software and hire engineers at scale, which separates the winners among ordinary companies., Capital One Financial CorporationIt rebuilt itself on the cloud and runs like a tech company that happens to lend, the kind of proven builder that can make its own harness., The Progressive CorporationIt has priced car insurance off its own driving data and software for decades, a regular business that already proved it can build tech and hire for it.15%
- In the near term, ad money is likely to flow toward video and audio, where advertisers can be fairly sure a real person is watching or listening. · Spotify Technology S.A.It sells audio ads to listeners wearing headphones, capturing attention that is verifiably human rather than bot-generated, making it a natural destination for ad budgets shifting toward authenticated audiences., Roku, Inc.It controls the home screen on many TVs and sells ads there, so it gets paid as TV ad money moves to streaming., Netflix, Inc.Its ad tier sells TV time to people actually watching, which fits the call that near-term ad money flows to video where a human is present., Magnite, Inc.It is the main independent seller of streaming TV ad slots, working for publishers rather than for Google, so it gains if video ad money grows and stays open to outside buyers.14%
- Cheaper ad buying platforms stand to grow sharply if AI assistants open their ad slots to outside buyers rather than keeping ads closed to their own networks. · The Trade Desk, Inc.It buys ads for advertisers across the open market, giving it reason to build an ads product for agents, and down sharply this year, so it has room to run if assistants open their ad slots to outside buyers., Criteo S.A.OpenAI picked it as its first ad tech partner for ChatGPT ads, with strength in shopping ads built on retailers' data, close to where agents will buy things.13%
- Companies that own both an assistant and an ad auction can likely run a meaningful share of whatever ads aimed at agents turn out to be. · Alphabet Inc.It owns an assistant people talk to and the biggest ad auction, so it can build ads aimed at agents into Gemini and search, and YouTube puts it in the video bet where advertisers can be surer a person is watching., Meta Platforms, Inc.It sells ads in feeds people scroll themselves, the attention agents have the hardest time replacing, and has its own open-weight models plus deep data on what people want., Amazon.com, Inc.Its ads sit right where people buy, so if agents start shopping for people, the checkout becomes the ad slot and it already owns the store.13%
- The data, monitoring and bot-checking tools that harnesses run on become more critical infrastructure as autonomous agents operate at scale. · Datadog, Inc.It is a monitoring tool companies already use, and agents running at scale will need the same kind of watching and debugging., Cloudflare, Inc.It sits in front of a big share of websites and already decides which bots get in, putting it on the identity and proof-of-intent problem once agents browse., MongoDB, Inc.It is a database developers already reach for, and harnesses need somewhere to keep memory, context and app data.10%
- Cash10%