AWS hopes agents are the new cloud

Welcome to Runtime! Today: How AWS is trying to take advantage of a generational lead in cloud computing in the agentic AI era, Google Cloud snaps up a piece of business AWS would have very much liked for itself, and the latest enterprise moves.

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

Disruptors get disrupted

AWS is finally ready to acknowledge that the launch of ChatGPT caught it flat-footed in late 2022 and 2023: "We didn’t have some of the whiz-bangy things that you could get out there quickly," CEO Matt Garman told Fortune this week during the company's New York Summit. His predecessor argued over and over again while trying to buy time that the generative AI era was just getting started, rendering any first-mover advantage moot in, as AWS executives like to say, "the fullness of time."

Nearly a year into an industry-wide push to make agentic AI the new center of enterprise tech, it's hard to say that AWS is late to the party given that most businesses are still struggling to articulate how agents make sense for their companies, outside of a few easy wins like coding and customer service. But assuming they figure that out, AWS — like everyone else in enterprise tech — hopes the next generation of business software will be built on its platform, and this week it rolled out new services with that goal in mind.

AgentCore is a new addition to Amazon Bedrock that was designed to help companies solve the "last mile" problem that AWS's Swami Sivasubramanian identified last year at re:Invent 2024.
It consists of several tools that can make the deployment process easier, such as a serverless runtime to execute the agent, memory to help agents learn from events in the past, and observability to make sure everything is running smoothly.
"With agents comes a shift to service as a software. This is a tectonic change in how software is built, deployed and operated,” Sivasubramanian said during the event, according to VentureBeat.
But agents will still have to work with everything that came before them, and AgentCore also includes identity-management technology for making sure agents can securely access corporate assets and play nicely with existing APIs.

AWS has one key advantage compared to the dozens of enterprise software companies trying to own the agentic era: A huge number of businesses are using its cloud platform in some fashion or another already. But agentic AI is a new challenge.

A decade ago AWS was the place where startups and new business ventures were built, but today it is trying very hard to move its existing customer base into a new era, which is what older enterprise tech companies like Oracle, IBM, and HPE have been trying to do for a long time.
While it's never been easier to mix and match services from different cloud providers in an application, it's not exactly easy, and there are a lot of companies that would be content to build their agentic AI apps in the same place they developed and currently run their traditional apps.
And now that there are several different options when it comes to competitive foundation models (unlike the scene in late 2022) and the OpenAI-Microsoft relationship has never been weaker, there's a chance AWS could strike a deal to put future versions of OpenAI's models in Bedrock.

Platform shifts are tricky, and they tend to produce winners that were born of the new era free from the baggage of the previous one. One day after its New York event, AWS laid off "hundreds" of people on Thursday as it tries to fashion a new set of building blocks for enterprise software.

Thanks to the immaturity of early enterprise generative AI tools and antiquated data-management strategies inside a lot of enterprises, AWS bought itself some time to catch up in the generative AI race.
But this shift represents perhaps the greatest threat to its perch atop the enterprise world since it grabbed that spot around a decade ago.

Outside looking in

While it chases AI business among existing customers, AWS remains shut out of one of the biggest money-makers of the generative AI era: OpenAI's compute budget. After Microsoft agreed to allow OpenAI to find computing capacity outside Azure's data centers last year, the GPT maker has now signed deals with CoreWeave, Oracle, and now Google Cloud.

On Wednesday OpenAI confirmed reports from earlier this year that it will tap Google Cloud for excess computing capacity as it trains future versions of its GPT models. The company will run ChatGPT Enterprise on Google Cloud's infrastructure in several regions around the world, which has got to be a little weird for the folks at Google given that it would love to call those enterprise customers its own.

But with billions in the bank and quite a bit more in store should OpenAI find a way to convince Microsoft to modify their agreement and allow OpenAI to become a for-profit corporation, that kind of business is hard to turn down. However, assuming Project Stargate falls short of the lofty expectations laid out in January, at some point OpenAI may have no choice but to turn to the cloud infrastructure leader.

Enterprise moves

Dilipkumar Khandelwal is the new chief customer officer at Celonis, joining the process mining company after technology leadership roles at SAP and Deutsche Bank.

Paul Smith is the new chief commercial officer at Anthropic, the first person at the foundation model company to hold that role following sales leadership roles at Microsoft, Salesforce, and ServiceNow.

Scott Holden is the new chief marketing officer at Vanta, joining the compliance company following a similar role at Brex.

The Runtime roundup

ScaleAI laid off 14% of its staff after several companies cut ties with the data-labeling company following CEO Alexandr Wang's decision to join Meta.

ServiceNow's bid to acquire Moveworks will face additional regulatory scrutiny as it tries to move into the HR software market, according to Bloomberg.

Thanks for reading — see you Saturday!