Azeem has a solid post about the bifurcation in AI companies. Some use tons of capital (the AI Labs) and some use little (the Apps and Agents).
But how defensible are they?
Arguably the cash invested alone makes for a small set of AI Labs. The big ones are spending massively to train the next best model, and the infrastructure needed to do the training and inference. Plus lots of stellar engineers that need to invent ever better solutions that code autocomplete won’t solve for the next, next step. Or maybe not (DeepSeek, Manus AI.)
But probably rather defensible. OpenAI has been showing how. Stay a bit better, keep linking more pieces together, be full stack (are you renting the model or GPUs or data?), be the easiest way to get everything through one API (coding, pictures, agents, reasoning, voice, video.)
In Azeem’s top right are companies that are cheaper to get going but their advantage is stuff like: user relationships, brand, private or user-related data, and integrations into workflows.
If it works, OK. But are these really defensible as they seem? The next best agent will migrate your account in a flash, or even layer on top of your old account.
Say you want to attack finance and accounting. The UI and workflows, the bank records and invoices…historically it would be a pain to switch. Hence the old world has an SMB monopoly with Quickbooks and an enterprise monopoly with Netsuite.
But an agent can download every report, convert from PDF to local data structure, clone all the reports you want, and never ask you to use a UI again. You just ask for the forecast or the board deck. The finance agent makes all that lock in go away. Maybe except for brand.
And THAT agent has little such lock in itself. So why not switch immediately.
That’s the current state of consumer user AI chats.
It reminds me of the early search engines that all sucked equally – OpenText, Yahoo, Excite, Altavista, AskJeeves…
Eventually one of them was so much better, and built a self-reinforcing revenue and distribution wheel (AdWords and Adsense feed Chrome and iOS integrations which feed more user interaction data to improve search…) that they were the monopolist.
Data and Distribution, the lock-ins that made Google win. It probably can’t be brand loyalty.
You must be logged in to post a comment.