This week, public software and data businesses have seen their market caps fall dramatically, as investors realized AI-native agentic workflows threaten incumbent software companies. I think this take is wrong: AI isn’t coming equally for all businesses; it will disrupt companies that don’t have strong network effects built into their business models, but benefit those who do.

A few thoughts on defensibility right now:

1. Most enterprise software companies rely on “stickiness” as their moat, which is a very weak moat. From ERPs to CRM companies to vertical SAAS, the dominant thinking over the past decade has been to move fast, get “customer lock-in”, and customers won’t migrate off you because the switching costs are too great (workflow stickiness, data stickiness, training stickiness, compliance stickiness, etc.). The issue with this logic is, while high switching costs prevent companies migrating for a small improvement in ROI, when there are step-functions in ROI, migration is worth it. AI is one of those step-function moments for many industries, so anyone with “switching cost” or “stickiness” as their primary moat is vulnerable.
2. Stronger forms of moats exist, but are rare in enterprise tech. In the consumer world, most of the biggest tech businesses have strong moats — either a network business model (eg. Facebook) or a marketplace business model (eg. Amazon) — that are extremely difficult for a new company to displace. These are much more rare in B2B — where the market has traditionally rewarded more end-to-end workflow ownership than defensibility. But they do exist; for example, as I wrote in this piece, in the data world, “data currencies” and “data marketplaces” have relatively strong network effects, but good curation and dashboards don’t. As a result, some companies are much more subject to disruption from AI-native approaches.
3. The companies that have strong moats will be beneficiaries of AI. So far, the market has punished all enterprise tech in its first reaction. But my bet is that while the companies with weak moats will be automated away over time, the companies where there are strong intrinsic network effects will perform even better. In the immediate term, AI will reduce their costs (while the network effect protects their revenue); but in the long-term, strong moats that provide durable distribution advantage will be one of the most valuable assets in the AI boom.
4. This is all true for startups too. I’m fairly skeptical of a large number of AI-native businesses that have had explosive growth, but limited defensibility (eg., does anything in the AI coding space actually have defensibility?). I’m even skeptical that the foundational models have real defensibility; “switching costs” as a model gets to know a person/enterprise/workflow seem to be the main argument for how the models will not be commoditized over time. To endure, startups either need to build with a strong network effect from day 1, or quickly have a “phase 2” with a moat after they quickly establish market share — or the forces of creative destruction will come for them too.

In the long-term, efficient markets will win, and things that don’t have moats will see their margin disappear. This has taken a while for enterprise software that sees “switching costs” as their main moat, and is finally about to happen. But it’s not a blow to the whole enterprise software industry, as the companies with real moats will thrive in this era.

‍