Pulse News

**CIOs Face a Glut of AI Sellers**

By Jon Victor

Mar 13, 2025, 10:41am PDT

Convincing professionals, software engineers, and other application developers to open their wallets for new artificial intelligence that speeds up their work seems easier than ever. But selling AI to large enterprises is a different ballgame. That was painfully clear this week at the first annual HumanX conference in Las Vegas, which brought together founders, venture capitalists, and executives from major businesses like Visa and Moderna.

Many founders told me they had traveled to Vegas in the hopes of promoting new AI products but were disappointed by the customer turnout. That might be the enterprise AI market in a nutshell. At least 8,600 "AI" startups raised funding in the last year and a half, according to PitchBook, many of them focused on selling to businesses. The sheer number of companies offering “agentic" solutions for tasks ranging from tech support to marketing is enough to make a chief information officer's head spin.

Meanwhile, many big businesses are moving carefully with new AI products, given concerns around cost and performance. The AI industry's promises about the technology have outpaced what it can actually deliver, said Roberta Schwartz, executive vice president and chief innovation officer at hospital system Houston Methodist. The conversation has moved quickly from simple uses for AI, like transcription, to claims the software can automate wide swathes of white-collar work.

"I feel like it's gone from zero to 60 in a nanosecond," she said. The good news for organizations like Houston Methodist is that the flood of AI apps hitting the market means that prices are coming down significantly, she said. Her company has already started using AI products by Ambience and Microsoft-owned Nuance to transcribe patient conversations and help doctors write up case notes. And it is using a service called Lena that uses AI to converse with patients before or after appointments to help them with insurance-related questions.

Executives also have been studying other transcription tools, including one for nurses, as well as how AI could help with analyzing images for radiologists and cardiologists, she said. But it takes time to evaluate the impact of any potential new product on patients and clinicians, and Houston Methodist set up an AI "council" that has to sign off first, Schwartz said. Many other large companies have similar committees for approving new AI tools, meaning they likely won't move at the pace many AI sellers hope.

The glut of AI options doesn't mean that individuals aren't getting major benefits from tools like ChatGPT. At a talk I moderated Tuesday about AI automation in the enterprise, multiple panelists mentioned that they regularly were using ChatGPT's new Deep Research feature, part of a $20 per month subscription, to produce exhaustively detailed summaries of topics they were researching. Brice Challamel, Moderna's head of AI products and innovation, told me he used the feature last week to suggest changes to a document about the use of AI in the pharmaceutical industry, before Moderna and other pharma companies submit the document to the Food and Drug Administration. "I bring one perspective, and the AI helps you bring another perspective," Challamel said.

**A121 Labs Eyes Challenges of Using Many Models**

Businesses trying to automate tasks with AI "agents" are having a hard time figuring out which AI model is best for each purpose, according to Ori Goshen, co-CEO of AI21 Labs, a developer of models. OpenAI's GPT 4.5, Anthropic's Claude, Google Gemini, and DeepSeek are just some of the hundreds of options AI buyers consider.

To fix that, the Israeli startup this week launched a product that runs simulations to determine which of the models will work best for specific tasks that buyers are interested in. For example, the system can run millions of simulated customer support conversations for a company and score how different models perform, Goshen said. Once a company uses AI models to automate those conversations in the real world, AI21's product aims to select the right model to handle each query.