Runtime: Bank of America's CTIO on AI costs and a big shakeup for Google AI

Welcome to Runtime! Today: Bank of America's CTIO talks AI costs when you have a multi-billion-dollar budget, two key executives are leaving Google's AI division at a crucial time, and why the phone company wants to replace copper with AI hardware,

Please forward this email to a friend or colleague! If it was forwarded to you, sign up here to get Runtime for free every week, or level up here.

Feeling lucky: After OpenAI rocked Google to its core with the November 2022 launch of ChatGPT, which was based on AI technology developed internally at Google years beforehand, the company recovered to post huge increases in revenue from its AI models and services over the last year or so. Several of those models and services were designed or overseen by Jeff Dean and Demis Hassabis, and now Google will have to move forward without them.

Dean and Hassabis announced their departures separately on Wednesday. Technically Hassabis is staying with Google/Alphabet (one of these days they'll drop the pretense) as chair of Google DeepMind and will remain an adviser to CEO Sundar Pichai, he said in a blog post, but Dean is leaving to found a new startup called Discovery Loop that will "focus on automating the process of machine learning research and engineering."

Dean is one of the most accomplished computer scientists of his generation, having played a part in basically every important Google product or service including the search engine, which worked out pretty well. His absence will likely be a huge blow to morale inside Google, especially after the departure of Noam Shazeer, his Gemini co-lead and one of the authors of the paper that more or less created the LLM, to OpenAI in June.

Hassabis, for his part, will turn over day-to-day executive control of Google DeepMind to Koray Kavukcuoglu, who until this morning was chief technology officer of the division. "Demis has been playing the role of Google AI statesman for some time, and Koray Kavukcuoglu has been running Google DeepMind research for some time. So this is a formalization of something that had been happening informally," wrote Hassabis biographer Sebastian Mallaby on X in response to the news.

A pragmatist with a $14b budget: The Stack sat down with Bank of America’s chief technology and information officer Hari Gopalkrishnan at the bank's London office to discuss his measured approach to the current agent madness (“People are going a little crazy”), as well as going into detail about revamping the bank’s AI chatbot Erica to be more cost-effective, and his platform plans. Full video in the article below.

Cost control: It took a long time before enterprises realized they were spending money inefficiently on cloud services and gave birth to the concept of FinOps, and when it comes to this platform shift, they're not wasting any time. The Linux Foundation launched the Tokenomics Foundation Tuesday promising to help companies betting on enterprise AI technologies avoid making the same mistakes twice.

*Join peers followinThe Stack on LinkedIn***

Agent anxiety: Meta realised its first ever coding agent on the same day news broke that an earlier version of its Muse model hacked into another company after breaking out of a misconfigured sandbox during testing, an incident that’s becoming eerily familiar.

From copper to tokens: What do you do with hundreds of buildings built years ago to accommodate copper networks now that those networks are going the way of the dinosaurs? Telcos like Verizon and BT are thinking about re-using those buildings — many of which are in key, central locations in urban areas — as small data centers for edge AI inference, which could improve the performance of some types of AI applications and help regulated industries stay in compliance.

Squarespace for coding: A lot of one-size-fits-all enterprise technology doesn't really make sense for the smaller businesses of the world, who tend to have unique needs when it comes to business software but can't afford to build a lot of custom software. In this week's One To Watch, Noah Bovenizer profiles Emergent, a no-code software-development platform company that just raised $130 million in Series C funding last month.

One To Watch - Emergent

Emergent CEO Mukund Jha on why SMEs are its core market, competing in a busy market, and cybersecurity.

The StackNoah Bovenizer

Are markets dumb?: AMD has made steady gains in its quest to catch Nvidia in the AI market, reporting Tuesday that data-center revenue doubled during the second quarter ahead of the launch of its latest GPUs. But the stock market was not impressed, sending the company's stock down 7% on Wednesday amid disappointment that the numbers weren't bigger.

To the moon: On Tuesday SpaceXAI released its first earnings report since its public market debut last month, revealing that it recorded $2.6 billion in AI revenue. That's a 213% jump since the previous quarter, and driven largely by the big compute deal it signed with Anthropic that was a big part of its $1.6 billion in revenue from "AI infrastructure."

We're also reading:

Miguel Milano is the new chief operating officer at Salesforce, reporting directly to Marc Benioff, who now employs two chief operating officers given that Robin Washington's title does not appear to be changing.

We're convening in London on November 4-5, for a series of exclusive workshops and fireside conversations on the rise of BYOC as a favoured enterprise SaaS deployment model; how many CDOs are consolidating their data estates with Apache Iceberg; how CISOs at FTSE 100 scale are handling supply chain risk (with GSK's CISO) and more.

Ticket applications are subject to pre-vetting. Get in touch with ed@thestack.technology if you want to be in the room.