I’ve done a lot of blogging about the disconnect between AI perception and enterprise reality. They’re based on the fact that what enterprises tell me about their AI business successes doesn’t align with the popular view of AI, which is mostly as a kind of at-your-shoulder friendly generalist expert. Whether you accept this is up to you, but even if you agree with enterprises (and me) on this point, it doesn’t fully address the question of whether all the money being spent on big AI data centers by players like Amazon, Google, Meta, and Microsoft will pay back.
Some publications are taking this seriously now; Axios thinks that the AI boom is just a justification for these companies to make big new data center investments that will pay off down the line through other applications. However, Google gave a pretty optimistic picture of AI success at the 2025 Goldman Sachs Communicopia + Technology Conference, and Oracle hit the earnings ball out of the park in their latest quarter, citing AI as a major driver. This combination surely raises a question I didn’t address yesterday in my look at enterprise AI projects and labor impacts. I said that enterprises could justify their AI investment (and likely a large one) with the level of job reduction they’re already reporting, which is far less than the stories on the job impact of AI suggest. I also said that my enterprise comments came largely from the IT side, and that they focused on internal AI projects rather than AI-as-a-service, which line organizations can expense, usually without any IT involvement. Is there nothing we can say about the “public” model of AI?
Well, there’s not as much we can say with conviction, but there are some things we can say. With the caveat in place, my qualifier that my own sources aren’t confident they can assess the service model of AI, let’s see what we can pick out.
My IT contacts have only limited direct involvement in the public model of AI. Only about ten percent mention using Microsoft’s Office AI tools, and fewer comment on using any of the Google tools. Those that do suggest that the usage they see in their own operations is limited to higher-paid workers rather than office staff. Technical people, marketing people, the sales force, have highly compensated people who use these tools, and companies seem to accept this usage. None of the enterprises said they had seen efforts to limit it.
This situation is changing a bit, though. Enterprises tell me that “public AI” services, like sales and support chatbots, are now seen as viable cloud applications because the data they access is in fact already public because of its usage. Anyone who buys a product can use support resources, and anyone who purports to do so can use sales information. Competitors have long haunted their rival’s websites. Increasingly, public AI is seen the same way as a website would be, with cloud augmentation almost a given.
One area where IT has very direct use of the chatbot/copilot model is in coding assistance. There, while enterprises say that AI coding tools have improved, there is still a fair amount of skepticism. For example, one enterprise said “We treat AI coding tools as a way to help junior staff; the senior people seem to think they’re less useful.” Another said “You can spend as much time in code review finding issues AI creates as you save in coding assistance. And the review people are more senior.” Most think these issues will be resolved, though, and nearly all think AI will eventually be used in coding assistance tasks by nearly everyone…with stress on the “eventually’.
Some comments on AI in writing and presentation authoring follow a similar line. AI can definitely help writing marketing material, and in particular the generic part that doesn’t get into the details of a company’s own products or positioning. One enterprise CIO said “We put together a lot of marketing documents, presentations, and respond to a lot of RFPs. I wouldn’t say that anyone trusts AI to do any of this on its own, but it does provide a good starting point if you’re careful and precise in what you ask for.”
But, and it’s a big “but”, none of those who commented on chatbot/copilot AI services said it had actually reduced jobs, and very few suggested that it could meet a formal assessment of ROI. Can it generate a better result? Yes, if used correctly. A quicker one? Same answer. Could we monetize the difference? Not so far, but we’re comfortable letting key people use it when they think it’s helping them.
I’ve tried to assess the value of these public chatbot/copilot services myself, using Google Gemini tools since I don’t use Microsoft Office. I also, like many, use Google for search. I’ve had some great results and some less so.
With regard to direct or via-search questions, my results have been decidedly mixed. For what I’d consider basic search requests, I think the AI summaries generated have been almost always accurate and often useful. For more detailed questions that involve statistical data from government sources, the results have rarely been helpful and often totally wrong. I have to admit that often these errors are from prompts/queries for which the “normal” search responses were also not helpful.
What worked better was Gemini’s “Deep Research”, which will generate a report based on your prompt, one that includes citations. I found this could be helpful in situations where I wasn’t already sure of where I’d go for the information, and there were few errors in the results. When Deep Research AI really took a flier, it seemed due to a bad prompt that lent itself to being misinterpreted. I wouldn’t use the result in a client document, but I would take advantage of the references. Others don’t have my reservations about use; enterprise contacts who commented on the use of Deep Research were happy to submit the outcome in about three-quarters of cases.
The best Google AI tool, in my view, is NotebookLM. This is a tool that lets you do a variety of things with a set of sources, which can be documents in PDF or Google Docs/Slides form, audio, video, or website URLs. When you’ve selected your sources (a single document works, up to three or four seem OK), you have a number of choices to present analysis. One is the “Mind Map”, which is a diagrammatic view of the themes and relationships of the sources. You can also get an audio or video overview.
Audio overview is my favorite. While you have four formats available for the overview (a conversation between AI virtual people, a quick audio summary, a critique, and a debate between AI entities), I like the conversational default, called “deep dive”. This generates a highly realistic two-person discussion of the source material. If you use it on a single document, which I’ve tried often and posted two examples for, it offers a discussion of the document that’s very different from simply reading it. If you use multiple documents, it analyzes the material and offers a discussion of the views the sources take. This discussion could be a useful teaching tool and could even be used to create marketing/sales material in audio form.
There is value here, then. The question is whether there’s enough value to justify the investment the online giants are making. I think, based on what I’m hearing, that it probably is, but it’s not nearly as clear that it would justify continued AI investment, and it’s not equally proved for all the major players. Microsoft and Google seem to have a clear path forward to justify current investment, and even to sustain it. Amazon needs to boost its own AI credibility, but since enterprises are pursuing cloud-hosted AI that often leverages open-source models, it surely has a shot. Oracle likewise. Meta? They may be the most problematic in terms of being able to make AI pay off continuously.
The current AI boom, though, isn’t likely to be sustained. This doesn’t mean that AI is failing, but that the dynamic may be shifting, more toward the enterprise AI missions that involve self-hosting, and more to cloud-hosting of a bring-your-own-model form of AI. However, a shift in dynamic would likely trigger changes in how the giants frame and price their services, because none of the giants wants to give up on something that’s been a PR boom for them, at the very least, so we’ll have to wait a bit to see how this all plays out.
