AI’s next moves: Murati’s startup, Google’s play, Airbnb’s doubts

It’s Friday.

In today’s edition:

• Mira Murati’s new AI startup

• Airbnb and HSBC on AI hype

• Google’s “Agentspace” approach

• DeepSeek’s funding puzzle

1: Mira Murati debuts Thinking Machines Lab

Former OpenAI executive Mira Murati has launched Thinking Machines Lab, a new AI startup focused on human-machine collaboration rather than fully autonomous AI. Instead of producing bigger, more powerful AI models that operate on their own, Thinking Machines aims to develop “multimodal” systems that tap into people’s existing expertise and adapt to individual needs.

Murati is joined by several ex-OpenAI colleagues, including Barret Zoph as CTO and John Schulman as chief scientist—forming a roughly 30-person team. Though they haven’t shared details on funding or product timelines, the company wants to be transparent about its research. Their core belief? Too many AI systems today are black boxes that leave users fumbling to customize or fine-tune them. By emphasizing collaboration and improved user interaction, Thinking Machines hopes to close that gap.

How to make any LLM meet your IP liability standards

Navigating IP liability with AI models can be complex, but it doesn’t have to be. In this on-demand webinar, Tabnine breaks down how to ensure any LLM meets your company’s IP and compliance standards—without unnecessary risks.

Learn how to:

• Identify and mitigate IP risks in AI-generated code

• Implement practical safeguards for your team

• Ensure compliance without slowing down development

Watch on demand now

2: Airbnb, HSBC offer an AI reality check

While many CEOs tout AI’s ability to slash costs and boost productivity, not everyone is on board with the hype. Airbnb CEO Brian Chesky recently admitted that despite the explosion of AI coding tools, there hasn’t been a “fundamental step change in productivity” yet. According to Chesky, he’s spoken to other tech CEOs who echo the same lack of tangible gains.

HSBC’s head of generative AI, Edward Achtner, took aim at companies promoting AI “success theater” without real results. Still, HSBC is experimenting with AI-driven customer support and code-writing tools. Goldman Sachs has also reported modest productivity boosts (around 10%) from using GitHub Copilot across its 10,000-plus developers.

Not everyone sees a slowdown, though. Pinterest’s CTO says 15% of its new code is AI-generated each quarter, which they consider meaningful. Even at Airbnb, Chesky says the company will keep moving forward, planning to roll out AI-powered customer support. For now, it’s a mixed landscape—some businesses see clear returns, others remain skeptical that AI is truly transforming their bottom line.

3: Google’s Agentspace push

Google Cloud is gearing up to launch Agentspace, a new AI-powered search chatbot for businesses that goes beyond basic data lookups. Besides letting workers find information across all their apps, Agentspace can also generate “AI agents” to automate tasks like expense reporting. Built partly from NotebookLM—Google’s popular tool for summarizing documents—Agentspace will be a top focus at the company’s annual conference in April, according to insiders and partners.

But some Google Cloud salespeople are concerned. Their compensation plans now include Agentspace sales targets, despite the product still being in development. It’s also up against Glean, a venture-backed leader in enterprise search that boasts over 100 app connectors and advanced data-ranking. All eyes are on how many integrations Google’s new product can launch with—connectors could be the key to convincing businesses that Agentspace is ready for prime time.

4: DeepSeek’s growth spurs funding dilemma

Chinese AI startup DeepSeek has rocketed to prominence with R1, a model that rivals U.S. offerings at a fraction of the cost. Until now, the company avoided outside funding to keep investors from demanding quick commercialization. But surging usage for its AI chatbot—and the need for more chips and servers—has DeepSeek considering a round of external financing. That could mean partnering with giants like Alibaba or China’s state-backed funds, which have already expressed interest.

Still, founder Liang Wenfeng is cautious. Accepting state-owned investment might draw unwanted scrutiny in Washington, especially since U.S. lawmakers are tightening export rules and blacklisting certain Chinese AI firms. Meanwhile, DeepSeek’s low prices for its API services offer minimal revenue, and even its popular new consumer app is still free. To secure more Nvidia chips—restricted in mainland China—DeepSeek may run data centers in places like Malaysia. The big question: Can Liang’s firm sustain the costly pursuit of advanced AI while keeping the U.S. market and regulators at bay?

Editors picks ☕️

Axios: OpenAI finds new Chinese influence campaigns using its tools.

Vox: Is AI really thinking and reasoning — or just pretending to?

Reuters: Wall St Week Ahead Nvidia to offer AI trades reality check.