The moat question at the end is the right one, and I think it deserves a slightly different frame than the traditional switching-cost analysis.
Carlota Perez's work on technological revolutions suggests that what we're watching in SaaS isn't moat destruction so much as value migration. Every major revolution has a moment where the business models optimized for the installation phase become structurally misaligned with the deployment phase. Per-seat SaaS licensing is an installation-era business model: it assumes the end user is a human who needs a tool. If AI shifts the work from humans using software to agents operating through software, the moat isn't switching cost anymore. It's whether you control the data layer, the workflow logic, or the integration surface that agents need to pass through.
Bloomberg is actually a perfect example of what a deployment-resilient moat looks like. The terminal's value isn't really the analytics (AI can replicate that). It's the network: the counterparty graph, the chat protocol, the fact that the entire industry's communication infrastructure runs through it. That's a moat that gets stronger as AI agents need to interact with the same counterparties humans do.
The companies trading at 10-12x earnings deserve a harder question than "is this cheap enough?" The question is whether their moat is built on human user dependency or on something structural that persists regardless of who, or what, is doing the work.
Glad this resonated. Perez is worth the deep dive if the moat question interests you. “Technological Revolutions and Financial Capital” is the core text.
And agreed on Bloomberg. That’s a good example of the tension: a business model built for how humans consumed financial data, now facing a world where the consumption pattern itself is shifting.
On the broader moat question: my read is that the long view isn’t really about whether today’s installation-phase companies keep their moats. It’s about building a framework that helps you identify the deployment-phase companies in the first place; the ones quietly embedding intelligence into existing workflows, where value accrues inside revenue lines that already exist rather than in headline-grabbing new segments. The moat question changes shape entirely when you ask it from that direction.
I explored the installation vs. deployment distinction in my last post (“The Turning Point Is Not a Moment”), and what I’m working through next is exactly the question you’re raising: what are the actual characteristics that distinguish a deployment-phase company from an installation-phase one? Because the traditional metrics weren’t built for that distinction, and that’s where I think the real analytical gap is.
I like the distiction between installation phase and deployement phse. Saas companies focused on winning installations since that meant a good revenue stream if retntion rates were high . Now the focus will be on usage which iwll be a function of the value the software provides for the customers work processes whether they are being done by humans or AI agents .
The moat question at the end is the right one, and I think it deserves a slightly different frame than the traditional switching-cost analysis.
Carlota Perez's work on technological revolutions suggests that what we're watching in SaaS isn't moat destruction so much as value migration. Every major revolution has a moment where the business models optimized for the installation phase become structurally misaligned with the deployment phase. Per-seat SaaS licensing is an installation-era business model: it assumes the end user is a human who needs a tool. If AI shifts the work from humans using software to agents operating through software, the moat isn't switching cost anymore. It's whether you control the data layer, the workflow logic, or the integration surface that agents need to pass through.
Bloomberg is actually a perfect example of what a deployment-resilient moat looks like. The terminal's value isn't really the analytics (AI can replicate that). It's the network: the counterparty graph, the chat protocol, the fact that the entire industry's communication infrastructure runs through it. That's a moat that gets stronger as AI agents need to interact with the same counterparties humans do.
The companies trading at 10-12x earnings deserve a harder question than "is this cheap enough?" The question is whether their moat is built on human user dependency or on something structural that persists regardless of who, or what, is doing the work.
Thanks for your detialed and considered response. I had not hear of Perez.
I agree with re Bloomberg.
The really big question is indded whether the moat will persist, It is also a very difficult one,
Glad this resonated. Perez is worth the deep dive if the moat question interests you. “Technological Revolutions and Financial Capital” is the core text.
And agreed on Bloomberg. That’s a good example of the tension: a business model built for how humans consumed financial data, now facing a world where the consumption pattern itself is shifting.
On the broader moat question: my read is that the long view isn’t really about whether today’s installation-phase companies keep their moats. It’s about building a framework that helps you identify the deployment-phase companies in the first place; the ones quietly embedding intelligence into existing workflows, where value accrues inside revenue lines that already exist rather than in headline-grabbing new segments. The moat question changes shape entirely when you ask it from that direction.
I explored the installation vs. deployment distinction in my last post (“The Turning Point Is Not a Moment”), and what I’m working through next is exactly the question you’re raising: what are the actual characteristics that distinguish a deployment-phase company from an installation-phase one? Because the traditional metrics weren’t built for that distinction, and that’s where I think the real analytical gap is.
I like the distiction between installation phase and deployement phse. Saas companies focused on winning installations since that meant a good revenue stream if retntion rates were high . Now the focus will be on usage which iwll be a function of the value the software provides for the customers work processes whether they are being done by humans or AI agents .