Robert Ta argues enterprise AI adoption requires solving four problems: integrating with existing systems, delivering accurate reporting, converting legacy data, and managing organizational change. He proposes embedding a forward-deployed engineer at each customer to bridge vendor and enterprise. ROI remains uneven across companies, with some gaining productivity and others unlocking entirely new capabilities.
They're really admitting in these moves, OpenAI and Anthropic, that APIs are not enough for enterprise. And we've talked about this before, which is in enterprise software, it comes down to a few things. There are 4 problems to solve in getting new enterprise software up and running. It's going to be integrations to existing systems that your company is dependent on. It's going to be reporting to get the right data to the right decision makers so that they have data-driven decisions. There's going to be data conversions from old system to new system. Insert whatever systems you want, X and Y there. And the last one is change management. And I actually think that's where the forward deployed engineer role does best is it's a champion that goes in on the inside, builds trust quickly. You kind of have this sort of person that could do product and do engineering and help with change management as a trusted champion embedded. And then they could bring those learnings from the field directly from the enterprise companies they're trying to install capabilities in back to Anthropic, back to OpenAI. So the learning loop should be quite high, uh, in terms of learning velocity. Well, what do you think about that? The whole, like, I don't know, my, my read on it is APIs are not enough. What does that tell you about kind of state of enterprise AI adoption?
Yeah, the reports that are coming out, they're kind of, um, it's a jagged surface area in terms of like who is getting, uh, return from AI, AI, what does it look like? We know that developers are getting a lot of value individually, and whether or not that's trickling up to like a top-line return on investment number for the enterprise is kind of a mixed bag. And I think there's two kinds of applications. Like, there's enterprises that are looking for AI to like make things more productive. Then there's enterprises looking for AI to like let them do new things in new ways that are unlocked by AI. And I think the, the proof points more in the latter, that there are firms finding ROI. But again, it's quite scattered in the playbook. Isn't like, you solid yet. So we're kind of looking at a situation where there's massive investment in data centers, and now, like, there's kind of, like, massive investment all the way down the supply chain to deliver these tokens to people that are using them on the front line, right? And all along that, there's still kind of, like, an unproven ROI on token mathematics that is universal. It's quite jagged across where it's applied. And so, I mean, it's a bet on the future, and it fast follows the largest infrastructure in that investment in human history, which is these data centers. And so this kind of follow-on investment just makes, makes sense. It's like kind of pot odds for, uh, the, the bets that the enterprise is making as a whole around AI.
