DX Today AI Daily Brief - Sunday, May 17, 2026

DX Today AI Daily Brief - Sunday, May 17, 2026

Today on the brief: Figure AI's humanoid trio Bob, Frank, and Gary completed more than fifty hours of fully autonomous package sorting with the new Helix-02 model. Runway has hit a 5.3 billion dollar valuation and is now openly chasing Google's Veo and Gemini Omni. Mira Murati's Thinking Machines unveiled a new architecture called interaction models. OpenAI extended Codex to work from anywhere across devices. A small Mac project named Osaurus is bridging local and cloud AI in a single open-source harness. Sprouts.ai raised a 9 million dollar pre-Series A to scale autonomous B2B revenue agents. PwC committed to rolling Claude out to 30,000 US consultants in a major Anthropic win. Notion turned its workspace into a hub for third-party AI agents. Martha Stewart co-launched home concierge startup Hint with a 10 million dollar seed. Amazon cut more roles in Seller Partner Services amid its accelerating AI shift. Microsoft introduced a Legal Agent inside Word, taking aim at legal tech incumbents. And a new multi-university study warns that just 10 minutes with an AI assistant can measurably blunt unassisted reasoning performance.

DX Today - Your daily AI news briefing, delivered every morning.

Send us fan mail

#AINews #ArtificialIntelligence #HumanoidRobots #EnterpriseAI #DXToday

Transcript

SPEAKER_03 0:00

It's Sunday, May 17th, 2026. You're listening to the DX Today AI Daily Brief. Today, three humanoid robots run a 50-hour autonomous sort. Mira Muriti's Thinking Machines unveils a new class of AI system, and Runway sets its sights on Google's video crown. Let's get into it.

SPEAKER_00 0:23

Figure AI staged a striking demonstration this weekend, putting three humanoid robots, nicknamed Bob, Frank, and Gary by livestream viewers, through more than 50 hours of continuous, fully autonomous package sorting. Chief Executive Brett Adcock said the machines moved nearly 28,000 parcels without any teleoperation or human intervention. The robots run figures in-house Helix Zero Two Neural Network, which processes vision, language, and motor control entirely on board. When a robot got stuck, the system triggered its own reset and resumed work. Adcock framed the run as proof that humanoids are crossing from controlled demos into shift-length repeatable labor inside real warehouses.

SPEAKER_03 1:09

From robots to a creative pivot.

SPEAKER_05 1:11

Runway, the New York-based generative video company that helped popularize AI filmmaking, is now valued at$5.3 billion and adding$40 million in annualized revenue this quarter alone. In a Friday profile, TechCrunch reports, the firm is no longer pitching itself purely as a creative tool for indie auteurs. Co-founders Christobal Valenzuela and Anastasis Germanidas say Runway is moving up the stack into general-purpose video generation, and they are explicit about the target. They want to overtake Google's VEO and the new Gemini Omni System. Hollywood studios, ad agencies, and game makers are now its core enterprise base.

SPEAKER_02 2:01

Mira Marathi makes her move. Thinking Machines, the secretive lab founded by former OpenAI Chief Technology Officer Mira Marathi, broke its silence this week with the unveiling of what it calls interaction models. Unlike conventional chat systems that take turns processing text, the new architecture continuously listens, watches, and speaks at the same time. The company says its models interpret audio, video, and language streams in parallel and respond in real time, opening the door to AI collaborators that feel less like apps and more like coworkers. Limited research previews are planned for later this year. Murati's team has now confirmed it is chasing a fundamentally different shape of intelligence than today's Frontier chatbots. OpenAI moves codecs everywhere.

SPEAKER_01 2:55

OpenAI rolled out a major update to its developer agent on Friday, branded work with Codex from Anywhere. The release extends Codex beyond the desktop integrated environments where it first launched, letting engineers hand off tasks from a phone, a browser tab, or a remote terminal, and pick up the same session minutes later on another device. OpenAI says Codex can now plan multi-file changes across an entire repository, run tests, and post pull requests on its own, with checkpoints that humans can approve along the way. The company is positioning the launch as a direct challenge to GitHub Copilot's agentic features, deep inside Microsoft's own developer stack.

SPEAKER_03 3:38

Now to the desktop AI revolution.

SPEAKER_04 3:41

A small Bay Area project called Osaurus is drawing outsized attention this morning after a tech crunch deep dive. OSAurus is a free open source harness for the Mac that lets users run local AI models like Llama and Mistral on their own hardware, while seamlessly routing harder queries to cloud services such as OpenAI, Anthropic, or Google. Files, memory, and tools all stay on the user's machine, but answers are stitched together transparently from whichever backend works best for the prompt. The launch signals a growing appetite for hybrid, privacy-aware AI setups that do not lock users inside any single vendor's walled garden.

SPEAKER_03 4:23

Funding flows into revenue agents.

SPEAKER_00 4:26

Sprouts.ai, a Singapore and United States-based startup building autonomous revenue agents for business-to-business sales teams, closed a$9 million pre-Series A on Friday. The round was co-led by True Global Ventures and Axel and brings total funding to$14 million. The company's agents ingest a customer's own first-party data, then prospect, qualify, and follow up with leads without a human in the loop for the routine work. Sprouts says paying customers already include Hewlett-Packard, RazorPay, Hiradius, and Udemy. The deal is the latest sign that investors are betting heavily on AI agents that do specific jobs, not generalist assistance.

SPEAKER_03 5:09

Now to Big Four consulting.

SPEAKER_05 5:11

PWC and Anthropic dramatically deepened their partnership on Thursday, with the consultancy committing to roll out Claude to all 30,000 of its United States employees. The expansion makes PWC one of the largest enterprise deployments of Anthropic's model to date. Executives on the call cited internal pilots showing up to 70% faster delivery on client engagements when consultants used Claude for research, drafting, and code review. Anthropic, fresh off a$30 billion run rate milestone, is using the win to underscore its enterprise traction against open AI. For PDWC, the bet is that AI literate consultants now command both higher fees and faster turnarounds.

SPEAKER_02 6:07

Notion opens its workspace. Notion unveiled a sweeping developer platform on Wednesday designed to turn its productivity workspace into a hub for AI agents. The release introduces three new building blocks. Workers, which let teams deploy custom code beside their notes, an external agent application programming interface, which lets third-party agents from companies like Anthropic or OpenAI act directly on Notion data. And Database Sync, which keeps Notion tables aligned with outside systems in real time. Chief Executive Ivan Zhao said the company is no longer competing only with documents and wikis. Notion now wants to be the operating system where humans and agents share the same memory of a project, from the workspace to the home.

SPEAKER_01 6:56

Martha Stewart is back in the news with an unexpected AI venture. Hint, a home management startup that Stewart co-founded with technologists Yi, Han Ma, and Kyle Rush, announced a$10 million seed round on Wednesday led by Slow Ventures. The product is pitched as a household concierge that uses computer vision and language models to track appliances, anticipate maintenance, and warn owners about leaks, drafts, and aging equipment before they fail. Stewart told Fortune the idea grew out of decades of running large estates. The bet is that consumers will pay for ambient AI that quietly handles the unglamorous backstage work of running a home.

SPEAKER_04 7:38

Meanwhile, in Seattle, Amazon's restructuring around artificial intelligence accelerated this week with another round of cuts inside its seller partner services division. The fresh layoffs follow more than 30,000 corporate role eliminations across the company in the past six months. Amazon framed the latest reductions as a streamlining effort, saying affected staff will receive separation pay, transitional health care, and placement support. Internally, leaders have been candid that agentic workflows and large language models are replacing layers of routine seller support work. The company plans to spend roughly$200 billion in capital expenditure this year, with a large share flowing into AI data centers and custom silicon.

SPEAKER_03 8:26

Microsoft enters the contract room.

SPEAKER_00 8:29

Microsoft introduced a legal agent built directly into Word on Thursday, a tool aimed squarely at the high-cost work of contract review. The agent reads agreements as they are drafted, flags obligations, deadlines, and risk language, suggests red lines, and walks negotiators through structured playbooks preloaded by their own firms. Microsoft says the system was designed with input from in-house counsel at several Fortune 100 companies and will roll out to Microsoft 365 enterprise customers in the coming weeks. The move places Microsoft on a direct collision course with established legal tech players like Harvey, Spellbook, and Ironclad's recently launched AI negotiator.

SPEAKER_03 9:12

And a sobering study to close.

SPEAKER_05 9:14

A multi-university research study published this week is raising fresh questions about how generative AI is reshaping thinking. Researchers asked participants to solve novel reasoning problems, first with access to an AI assistant for 10 minutes, and then without it. Performance on the unassisted round dropped measurably even after the brief AI session, suggesting that short exposure can blunt independent problem-solving skills. The authors stop short of declaring a long-term cognitive cost, but they urge schools and employers to pair AI tools with practice protocols that keep humans actively reasoning, not just approving. It is an early warning sign in a year defined by relentless AI adoption.

SPEAKER_03 10:08

That's your briefing for Sunday, May 17th, 2026. For DX today, stay curious.