AI companies want their companies to automate white-collar work, but the strong default for Zero Data Retention means you can't train on enterprise roll-outs.
Existing enterprises are obviously heavily disincentivised from giving OpenAI or Anthropic the data they need to automate away their work.
Imo the most underrated AI development of 2026 is therefore the announcements of the "OpenAI Deployment Company" and Anthropic's "Joint Venture". In short:
1. Within 1 week of one another, both OpenAI and Anthropic announced new standalone entities co-owned with PE firms, consultancies and banks.
2. From what I understand, these new companies will basically be entirely dedicated to forward-deploying engineeers into boomer corporations around the world to teach them to use Claude/GPT correctly.
3. Pure speculation from my side, but it strikes me as plausible that we should view this as a deliberate attempt from the frontier labs to get closer to "work data".
4, Maybe this literally means that deployments accompanied by the "deployment company" will be free from ZDR. This might be the case if eg the company that AI is being deployed in is entirely owned by a PE firm with a stake in the "Deployment Company", because that PE firm can then share in the upside of using models trained from that roll-out to automate the rest of their portfolio.
5. Or maybe it just means getting access to more distal proxies of "what works" in enterprise settings. Still, useful.
6. Either way, the question of "who owns data generates by enterprise deployments" seems heavily, heavily underscrutinised to me. Seems like an essential part of leverage for the would-be future "permanent underclass" - misplay this hand at your peril
For most domains, real work is the only environment that useful data can come from.
A big reason why labs are building products, acquiring companies, and forwardly deploying engineers into enterprises is to gather enough work data to train their agents on a wider range of tasks.