Big Business is still skeptical about AI agents

Welcome to Runtime! Today: Almost a year after agentic AI became every vendor's North Star, business remains slow, OpenAI clarifies its roadmap, and the latest enterprise moves.

(Was this email forwarded to you? Sign up here to get Runtime each week.)

Trust the process?

Steve Jobs once said (probably), "People don't know what they want until you show it to them." After endless demonstrations of AI agents from dozens of vendors across almost a year's worth of launch events and road shows, businesses have seen and heard a lot about AI agents, and they're still not sure if they want them.

The Wall Street Journal conducted a survey of attendees at its CIO Network Summit this week, and 82% of respondents said they were either experimenting with AI agents or not even entertaining the idea. As has been the case for several months, attendees said the accuracy of the results produced by the current generation of AI agents is simply not good enough for them to consider putting agents into production.

• It's also a trust issue; 29% of survey respondents said they don't think there are enough security measures in place to protect their data.
• There are lots of companies that want to believe in the promise that AI agents can cut their costs and allow their remaining employees to focus on higher-level tasks that deliver more value.
• But right now, they are voting with their budgets: The Next Platform noted that Cisco's results this week underscore how slowly the generative AI revolution is rolling out across big business.

In response to that pushback, vendors at the CIO Network Summit essentially said: "trust us." That's a tough sell to CIOs who have seen a few different technology cycles in their careers and, for existential reasons, are skeptical of spending money on technology that doesn't work.

• In keeping with years of enterprise software history, there's a both carrot and a stick in the agentic AI pitch: Adopting the technology we're selling will make your company better, and not adopting the technology will make your company fall behind.
• “Accept that it is imperfect,” said Bret Taylor, co-founder and CEO of Sierra, at the summit. “Rather than say, ‘Will AI do something wrong’, say, ‘When it does something wrong, what are the operational mitigations that we’ve put in place to deal with it?’”
• Companies looking at cloud computing 15 years ago had many of the same concerns about unreliability and security, and most of those concerns were ironed out over time.
• But in an era where IT budgets have tightened, those "operational mitigations" don't magically appear out of nowhere, and a reasonable question in response to Taylor's statement might be, "isn't that your job?"

Enterprise technology shifts always move slower than vendors would like, but this time around it feels like they are under pressure to deliver results given the enormous costs required to build the technology behind agentic AI. If that's true, acknowledging customer concerns and working directly with them to find a way forward might be more productive than scolding them for not jumping on the bandwagon.

• According to the WSJ, 75% of those surveyed said they have only gotten "a small amount of value" from their investments in generative AI technology over the last two years.
• And those people have been under a lot of pressure to invest in generative AI from upper management and boards who saw what happened to companies who ignored mobile computing, the cloud, or even the internet itself and think generative AI is that next shift.
• “They’re so scared to be left out of a revolution in progress that they buy an AI thing of some kind, and they try to figure out how to drive value with it,” said Palantir CIO Jim Siders, at the event.
• At some point in the not-too-distant future, if businesses haven't figured out how to drive value with generative AI, those same managers and board members are going to tell them to stop spending.

There can be only one

OpenAI CEO Sam Altman acknowledged Wednesday that the company's foundation-model product lineup and roadmap got way too complicated over the last year, with multiple GPT models existing next to multiple chain-of-thought "reasoning" models. The company now plans to release its o3 reasoning model — which had been expected to arrive any day now — with GPT-5, now due at some point later in 2025.

"We hate the model picker as much as you do," Altman said in a post on X. GPT 4.5, code-named "Orion," will be the next update and "the last non-chain-of-thought model we ship," he said, while GPT-5 will incorporate the o3 model.

GPT-4.5 will arrive in "weeks," Altman said in a follow-up post, and GPT-5 is now expected later this year. Orion was expected to be the GPT-5 release, but The Wall Street Journal reported in December that OpenAI was running into problems shipping Orion, which implies the company decided to avoid trying to justify it as a major upgrade to the almost two-years-old GPT-4 model.

Enterprise moves

Don Johnson is the new CEO of Docker, joining the company after building Oracle's cloud infrastructure business until 2023 and playing a key engineering role in the early days of AWS.

Jagroop Bal is the new chief financial officer of Druva, following similar roles at VMware and HPE.

Ashling Kearns is the new global GTM chief operating officer (that's a new one for me) at Celonis, an expansion of her current role as senior vice president of marketing at the process mining company.

The Runtime roundup

Arm plans to release a data-center CPU later this year with Meta signed up as the first customer, according to The Financial Times, which could put it in competition with its own potential customers.

Fastly's shares fell more than 20% Thursday after it reported disappointing earnings on Wednesday, another signal that the CDN market is going through some things.

The "four nines" standard for cloud services allows for a little over an hour of downtime per year, but if operators and customers were willing to tolerate a few days of downtime each year, the U.S. power grid could add as much as 100GW of capacity — multiple Project Stargates —without having to build new generation plants, according to new research out of Duke.

Thanks for reading — see you Saturday!