Oracle NetSuite: Trust AI the way you'd trust a new hire

The IT industry has a paradox all its own: things move very fast and very slowly at the same time. Artificial intelligence is the textbook case. Talked about since Alan Turing, a disappointment for most of the decades since, and then, seemingly overnight, the only thing enterprise software (and everybody else, everywhere) wants to discuss.

That's the view from Brian Chess, Senior Vice President, AI, Product and Technology at Oracle NetSuite. "I took a machine learning class back in grad school in the '90s, so yes, it's been around," he told TechDay. "People started thinking about this stuff in the 1950s."

Indeed, search YouTube, and you will find a Computer Concepts video from the 1980s, discussing AI and machine learning. The striking point is that the concepts don't differ from the ones being applied and commercialised today.

Chess joined NetSuite as an early developer in 1999, left to found Fortify Software (sold to HP), and returned in 2012 with what he called "A much better appreciation for what a growing business needs." After all, at Fortify he saw exactly that, up close and personal.

NetSuite has long tinkered with machine learning, including supply chain predictions. "The problem that we had with those early features is they took a lot of effort to develop, because you needed to know a lot of math, and then you also needed a lot of very practical engineering about how you are going to muster and manage this data," he said.

That background meant productisation, packaging, and above all, monetisation was difficult or even prohibitive.

What changed was ease of use. "AI graduated from being an interesting topic to being an important topic when large language models started moving out of their hallucination phase," Chess explained. "What's happened in the last year is, it has transitioned from being an important topic to being the only topic."

From next to now

Last year NetSuite Next was a promise. "My word for last year was next. My word this year is now. It has arrived," Chess said.

Customers can use NetSuite's own AI or hook up Claude, ChatGPT or Gemini via an MCP server. With that AI in there, NetSuite said users can simply ask their ERP to do the things they need done. If all works as it should, conversational interfaces immediately crush the learning curve, while also solving the 'where is it' problem that has always plagued complex organisations.

Efficiency gains like this can't be overlooked: "One way or the other, people should be thinking about how they're going to adopt, because if they don't, they're going to get left behind," Chess noted.

If that sounds like a sales pitch, it really isn't. Agentic AI is now also available for personal use, building on the prescient introduction of AI via consumer channels back in 2023. Once the benefits of squashing personal administrative or information-sourcing tasks is seen, the scales tend to drop rather suddenly from one's eyes.

Analysts agree, up to a point. Gartner predicted 40% of enterprise applications would include task-specific AI agents by the end of 2026, up from less than 5% in 2025. Yet its 2026 CIO survey found only 17% of organisations have actually deployed agents. Fast and slow, in other words; our pick is that as more scales drop from more eyes, an air of inevitability will descend, and people everywhere will appreicate that they can get on board, or get left behind.

That's tangentially confirmed by an observation from Chess: He said NetSuite's AI Connector Service, which lets outside models plug into the ERP via MCP, is moving. "It's the fastest adopted feature I've ever seen at NetSuite," Chess said. "First of all, it's relatively easy to do, and secondly, people see the benefit."

And ChatGPT going to market via consumers back in 2023 may have had a smidge to do with that.

The new hire problem

Trust, of course, is the catch. "Would I allow a personal AI system like OpenClaw to reserve an Uber for me? That doesn't sound so bad. How about log into my bank? No, I don't think so," Chess mused. The trouble, he said, is that "there are no good controls when connecting to the bank that say 'read is okay, write is not okay'.

"A place where enterprise systems may be ahead for once [as in ahead of consumer systems] is in giving people control," he continued.

NetSuite's existing roles and permissions apply to any AI, its own or a customer's. "Why do we think our guarantee is so good? Because it's the same system our customers have used to control what their employees do, all along."

Which brings an unexpected twist. "For the most part in my career, I've been told to avoid anthropomorphisation [of computers, applications, etc]. But now, it actually seems like it might be an advantage," Chess said. "When the new hire starts, should they have access to the payroll system? Probably not, unless they're on the payroll team." The same applies to agentic AI; same controls, same levels of trust, same levels of access and ability to take action.

In other words, start read-only and let trust build. "The very first thing we see people asking the AI is usually a question they know the answer to," he added, noting that this is eminently sensible. After all, how else would one spot an incorrect answer, or assess the accuracy of the service.

The caution is warranted. A Forrester Consulting study for Boomi of 409 IT decision-makers found 86% of organisations have moved AI agents beyond pilots, yet just 34% trust what those agents do. Gartner expects more than 40% of agentic AI projects to be cancelled by the end of 2027, partly over inadequate risk controls.

Whether machine or human, trust is always earned; it also should be a given. The security crowd like saying 'trust, but verify', and for good reason.

We've spent years teaching AI. Now it teaches us (and a pricing puzzle)

Freewheeling a little in the interview, the conversation drifted towards some of the peculiarities and idiosyncrasies of the emergence and popularisation of AI. For years, people had to teach AI; the best example of that is probably CAPTCHA, which turned everyday internet users into free data labelers.

Chess said the Ask Oracle the conversational interface switches it back. "You can have a conversation and then say, 'the analysis we just walked through, I want you to do that every day,' and without realising it, you just made an agent." And the student becomes the master: "AI is a fantastic teacher. It translates what you say you want into the language of [the system], and then also shows you what it did. You just learned a little about how that system works."

Finally, what of implementation partners used to lengthy ERP projects, now that deployments promise to take weeks or days? "I would love to have insight into how that is going to play out. I don't know where it's going to go," Chess conceded. Partners might cut prices, or argue: "We delivered the same value to you. We just came in a compressed time frame. So why should you pay less for that?"

Henry Hazlitt answered that one in 'The Curse of Machinery': when machines make things cheaper, more people buy them. Cheaper implementations should mean more NetSuite customers, not fewer, because the cost of an implementation will move from timeframes of months or years, down to days or weeks. And that means those who wanted NetSuite, or any other enterprise solution but couldn't cut the cheque, may find it within budget after all.