The top news stories in enterprise AI

Is Claude having its ChatGPT moment?

Today’s top story, from reporter Aili McConnon:

For most in the tech world, ChatGPT’s 2022 launch was a watershed moment. Not because large language models (LLMs) were new, but because OpenAI wrapped them in a consumer-friendly interface that made generative AI accessible to a mass audience.

Now, Anthropic’s Claude is hitting a similar inflection point—driven primarily by two products, the newest of which was released last week. Claude Code acts as an autonomous coding assistant, while Claude Cowork brings similar agentic capabilities to non-technical users for general productivity tasks. Since those launches, Claude’s total web audience in December has more than doubled year-to-date, according to market intelligence firm Similarweb, while research firm Sensor Tower found Claude’s daily unique desktop visitors up 12% globally month-over-month.

That rapid adoption underscores a looming question on whether traditional software companies may face an existential threat from AI agents that can automate coding and other knowledge work. But much like ChatGPT’s dramatic debut, the real story is the shift in how software gets made and used.

On the latest episode of the Mixture of Experts podcast, Gabe Goodhart, Chief Architect of AI Open Innovation at IBM, recalled the ChatGPT debut. “The form factor shifted and all of a sudden it crossed that uncanny valley into something that felt like you could trust it,” said Goodhart. “I think that’s what we’re seeing happening with Claude Code here.”

According to IBM Distinguished Engineer Chris Hay, ease of use and reliability are big factors for Claude’s crossover from developers to the mainstream. And with the release of Claude Cowork, users no longer need a dedicated terminal but can access Claude on their desktop, he said.

“Trust in the quality of the output and the safety of the process is what’s combining to hit this moment,” Goodhart said.

That emphasis on safety and trust is also at the foundation of the partnership Anthropic and IBM announced last year, aimed at accelerating enterprise adoption of AI while building security and governance into the software development lifecycle. “Enterprises are looking for AI they can actually trust with their code, their data and their day-to-day operations,” said Mike Krieger, Chief Product Officer at Anthropic, in the announcement. The partnership, he said, will aim to “build the open standards that will make AI agents genuinely useful in business environments.”

Catch the full episode of Mixture of Experts on YouTube, Spotify or Apple Podcasts.

Today’s AI advances have dramatically impacted the tech stack. And it’s already showing up in companies’ operating margins, with a projection of 19% for 2026, up from 13% in 2019, according to Rob Thomas, SVP of Software and Chief Commercial Officer at IBM.

“I’m not sure at any point in our careers we have seen that type of acceleration in such a condensed period of time,” said Thomas during a conversation for IBM’s Technology Summit, which will stream on Tuesday, January 27.

With innovation comes automation, more software use cases and a boost in productivity. And businesses are only seeing the beginning of AI’s impact. With less than 1% of enterprise data being represented in today’s foundation models, according to Thomas, companies face a pivotal question: how do we bring enterprise data into the models?

Furthermore, as enterprise architectures become more agentic, companies will have to contend with two untapped opportunities within enterprise data: unstructured data and real-time data. “We’re at the start of a new era in data, which is going to be more focused on unstructured, real-time data and making enterprise data accessible in large language models and small language models,” he said.

In 2026, enterprises will also need to think about their resilience, beyond security, sharpening focus in areas like risk management and third-party dependencies. “There’s a whole driver around enterprise and geographical resilience,” Thomas said. “And the key to that is sovereignty, security, governance.”

Digital sovereignty is a growing demand from businesses, and the Technology Summit will explore what it takes to deploy AI securely and compliantly, without slowing down on innovation.

— Anabelle Nicoud

That’s the number of confirmed hallucinated citations found by AI detection tool GPTZero in over 50 of the 4,841 academic papers accepted by the AI-focused conference on Neural Information Processing Systems.

— Patrick Lucas Austin

Ahead of the 2026 Grammy Awards, IBM rolled out GRAMMY IQ, an AI-powered fan experience, built with watsonx, that turns the Recording Academy’s music data into interactive quizzes and digital features.

IBM Research, working with Ansys and backed by DARPA, developed a new machine learning tool that can predict heat inside advanced computer chips with near one-degree accuracy, faster than existing methods.

OpenAI’s Chief Global Affairs Officer revealed that the company is on track to unveil its first physical AI device in the second half of 2026.

OpenAI is building age prediction into ChatGPT to flag users under 18.

Liquid AI released LFM2.5-1.2B-Thinking, a compact reasoning model designed to run on a smartphone while matching larger models on math and problem-solving benchmarks.

— Sascha Brodsky

A team of engineers at Columbia University has created a “humanoid robot face” that learned to read lips by watching hours of YouTube videos of human conversations.

The perfectly sensible goal, as research team member Hod Lipson explained in a blog post, is for robots to learn how to move their own mouths properly when speaking, “or they will forever remain uncanny.” Study lead Yuhang Hu explained: “The more the robot watches humans conversing, the better it will get at imitating the nuanced facial gestures we can emotionally connect with.” In other words, scientists, don’t let your robot give humans the ick.

Still, lip-reading is kind of a weird place to land. Sure, we hear about AI excelling at human tasks all the time—writing Beatles songs, picking an outfit that suits your tastes, giving you guidance on ghosting without guilt. But a machine quietly watching human mouths and learning to infer meaning from them feels … more personal. Like it’s picking up on something it wasn’t explicitly invited to notice and pointing out the spinach in our teeth.

This is the part where pop culture usually jumps in to whisper, this is how it starts. We don’t need a full sci-fi montage to know the trope in which AI watches too closely, listens too carefully and ends up knowing more than anyone is comfortable with. The scene in which HAL lip-reads Dave and Frank’s conspiratorial conversation in 2001: A Space Odyssey triggers our primal fear of being judged and punished by an omniscient being—just like some people recoil at pop-up ads that correctly deduce your need for athlete’s foot cream or a good personal injury lawyer.

The unromantic reality, of course, is that this system isn’t “watching” anything in the human sense. It’s training on vast amounts of video data and learning patterns, the same way most AI systems do. If lip-reading robots feel unsettling, it’s probably less because of what the technology is actually doing and more because of how easily we project intent onto machines.

If anything, the watchout here is that even the most sophisticated AI machine still operates on one of the oldest rules of computer programming: GIGO, i.e., garbage in, garbage out. Which is worth bearing in mind if the machine in question is learning by mining the depths of user-generated content.

— Euny Hong

Was this email forwarded to you? You can subscribe here.

You are receiving this email because you signed up for: IBM Think Newsletter

IBM US, 1 North Castle Drive, Armonk, NY 10504, United States