Hugging Face's enterprise hug; Asana's agent play

Welcome to Runtime! Today: Hugging Face releases an open-source tool for deploying AI models, Asana helps customers build AI agents in its project-management tool, and the quote of the week.

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

Ship it

Everybody needs HUGS: There aren't too many enterprise tech companies named after an emoji, but Hugging Face is one of them. The AI model-hosting company unveiled a new open-source service this week that was designed to help companies get those models into production.

HUGS (Hugging Face for Generative AI Services) "are optimized, zero-configuration inference microservices designed to simplify and accelerate the development of AI applications with open models," the company said. Designed to allow users to deploy their models on AWS, Google Cloud, or Digital Ocean, HUGS is an alternative to Nvidia's Inference Microservices, according to Infoworld.

Like a rock: IBM's Granite AI models hit the 3.0 milestone this week, and the company claimed they'll be able to outperform comparable models from Meta and Mistral. There are several different sized options available under the Granite 3.0 moniker, and anyone interested in using them will be able to run them on Amazon and Hugging Face in addition to IBM's watsonX service.

But the most notable thing about the Granite series of AI models is their true commitment to open-source principles, unlike Meta's decision to use that term for its Llama models while placing restrictions on how they can be used. Granite is available under the Apache 2.0 license, which "changes the notion of how quickly businesses can adopt AI when you have a permissive license that enables contribution, enables community and ultimately, enables wide distribution," said IBM's Rob Thomas, senior vice president and chief commercial officer, according to VentureBeat.

Agents are the new tasks: Asana's collaborative planning software is used inside many companies to track the progress of various projects or organize workflows. And like every other company in enterprise software over the last few months, Asana jumped on the agentic AI bandwagon this week with the beta release of its AI Studio.

AI Studio "lets customers design workflows by simply giving AI instructions in plain language," Asana CEO Dustin Moskovitz said in a press release. In an example provided by the company. Morningstar used the new tool to build a workflow that gets activated when a new project request is sent to the IT department and automatically evaluates the request, asks follow-up questions, and decides how important that request is against other pending projects.

Inflection point: Not many companies have had a year like Inflection AI, which watched its co-founder and several key engineers walk out the door in May to join Microsoft and build a whole new AI team. Undeterred, this week Inflection released its own take on agentic AI called Agentic Workflows, which will be part of its enterprise product.

The new service seems designed to help companies upgrade their RPA systems from UiPath, which is partnering with Inflection. "Ensuring your AI can be trusted to do work correctly — both in terms of factual correctness and what is aligned with your company policies and culture — is the first crucial step toward empowering AI agents in your workflows," it said in a blog post.

Hacking the code: Given that AI coding assistants have become one of the stickier products introduced in the generative AI era, swarms of large and small companies are racing to duplicate GitHub Copilot. Augment launched earlier this year with $227 million in funding to chase that goal, and this week it introduced its flagship product, Augment Code.

Augment Code was designed as a more team-oriented coding assistant compared to GitHub Copilot, co-founder and CEO Scott Dietzen told The New Stack. "As you get to these larger teams and more complex code bases, the current AIs treat every developer as if they’re the only one in the code base, and that’s not at all close to reality," he said.

Stat of the week

How many AI projects actually make their way from proof of concept to production? Despite all the hubbub around generative AI, that number actually declined from 2021 to 2024 from 55.5% to 47.4%, and the return on those investments fell by a similar amount according to new data from Appen.

Quote of the week

"There is no sane way to read 300 or 400 resumes and truly give them care.” — Jonathan Eunice, director of platform and security at 3Play Media, highlighting the challenges that hiring managers face when sourcing tech candidates in the AI and return-to-office eras.

The Runtime roundup

Microsoft CEO Satya Nadella asked its board of directors to lower his compensation after a series of security debacles over the last year, so they docked him $5.5 million in cash compensation while raising his overall compensation by 63% to $79.1 million, clearly showing that everyone learned a lesson here.

The Change Healthcare breach exposed the personal information — including in some cases detailed medical histories — of nearly 100 million people, UnitedHealth confirmed Thursday.

Thanks for reading — see you Tuesday!