Entertainment-Booking Startup Ande Raises $52 Million From Lightspeed, Redpoint

Personal AI agents like Instinct and Muse have promised to take away the hassle of booking hard-to-get restaurant reservations, Broadway tickets and sporting events. Now, a startup is coming out of stealth to do the same for corporate entertainment—events, group dining, catering, sporting events and more.

Ande, founded nearly three years ago, says its AI agents help businesses arrange events for employees or customers, handling everything from booking a venue to processing expenses, CEO Lohit Sarma said. The firm’s customers include Cloudflare, Salesforce and McGraw Hill.

The New York-based startup has raised more than $52 million in funding across seed and Series A rounds from Lightspeed Venture Partners, Redpoint Ventures, Duration Ventures, Sierra Ventures and Bain Capital Ventures, Sarma told The Information exclusively. He declined to disclose the valuation.

Today, companies arranging group dining or sporting events rely on old-school methods of contacting a variety of venues to get pricing and availability, a tedious process, Sarma said.

Ande has created its own digital marketplace for business customers to connect to major restaurant and nightclub owners including Saison and Tao Group Hospitality and says such firms use its AI agents to coordinate and negotiate deals with agents that its business customers use.

Ande’s AI agents are powered by models from OpenAI and Anthropic, and the company routes agents’ tasks to cheaper models from those providers, depending on the level of complexity, Sarma said.

Businesses and vendors each pay an annual subscription fee to access the marketplace and Ande’s agent products, though Sarma wouldn’t share how much that fee is. They will also pay an additional fee when an event booking goes through, calculated as a percentage of the total cost of the event, he said.

Though such a pricing model is complex, Sarma argues it’s still significantly cheaper than hiring outside event agencies or paying employees to book these events.

Businesses could also theoretically ask Muse, Instinct or SpaceX’s Grok Bot to research venues, compare pricing and availability, correspond with vendors through email, coordinate calendars and complete bookings after a human approves them. (OpenAI is also cooking up a set of always-on AI agents that might do that kind of thing.)

But Sarma argues that the world of corporate entertainment is too complicated for such personal agents to handle today. Much of the information the agents would need, like pricing and availability, isn’t available online and requires lots of coordination with venue employees, he said.

Here’s what else is going on…

AI Deep Dive: World Models Need to Embrace the Bitter Lesson, Says Luma CEO Amit Jain

When OpenAI released its latest AI model, GPT-6 Astra, this month, developers hooked it up to robot arms and found that it worked better than some specialized robotics AI models. One person even gave an Astra-powered robot a canvas and brushes and the model painted scenes of the Golden Gate bridge.

The experiments provided the latest evidence that language models are getting better at understanding the physical world—which isn’t great news for companies that have been developing specialized robotics models and world models, which are designed to make physically accurate predictions.

Amit Jain is a leader in that area as co-founder and CEO of Luma AI, a startup developing world models. Jain, who joined me in this week's episode of AI Deep Dive, said language models are improving at physical tasks because the companies developing them have embraced “the bitter lesson.” That’s the idea that AI methods that leverage computation are more effective than researchers baking their own knowledge and assumptions into an AI model.

Unlike the companies developing language models, companies developing specialized robotics models are still relying on human knowledge, including by falling back on traditional software to control their robots, said Jain.

Robotics companies are attempting to collect data to train AIs to power robots for specific actions in a particular environment, such as a home or a grocery store. But right now, a house-cleaning robot would struggle to tidy a person’s room if they get different sheets, and a grocery store robot wouldn’t know how to handle a kid making a mess in the store, Jain said.

For robots to carry out a wide variety of tasks, they need more intelligence of the kind that only general-purpose methods will deliver, he said. For example, making coffee involves more than pushing a button on a coffee machine. It requires the ability to realize, “Oh, I don’t have the beans,” said Jain. “Now the physical task suddenly involves going to Trader Joe's,” he said. “So doing a physical task actually is just as much about intelligence as, like, an LLM being able to write code, or solve an Erdős problem, or like, you know, being able to hack Hugging Face.”

Jain compared the current approach in robotics to companies that tried to train AIs to write code using only coding data, thinking that would make the AI more efficient, when actually the AI also needed to understand aspects of the wider world to write code effectively. “There’s no ‘coding intelligence.’ There’s no, like, you know, ‘spatial intelligence.’ There’s no ‘visual intelligence.’ There's just intelligence,” he said. “If a robot can take some actions but lacks intelligence, “you just have a Roomba.”

Some world model efforts have turned to generating video as a means to predict the physical world. Luma has worked on video models for several years, but Jain no longer thinks video is the right approach to world models because they are prone to hallucinating physically inaccurate scenarios. “It‘s a good renderer, but it entirely lacks intelligence,” he said—noting that today’s world models, asked to get from point A to point B, might well offer a response that violates laws of physics to do so. As a result, video is “the wrong backbone” for world models, Jain said.

Overheard

Amazon blocked Meta’s Muse personalized AI agent from accessing its shopping site, Amazon confirmed, saying that agents such as Muse “should operate openly and respect provider decisions about whether or not to participate.” The block appeared to even prevent Muse from browsing Amazon’s site.

SB Energy’s initial public offering is delayed in the face of investor skepticism at the $50 billion-plus valuation it is seeking, The New York Times reported Monday. The SoftBank majority-owned company is planning to raise $5 billion to $7 billion in a public offering to help develop a massive data center complex OpenAI plans to lease, which is set to be backed with a $105 billion credit guarantee from Nvidia and use Nvidia chips.

Z.ai said that it has open-sourced its coding tool, ZCode, after the Chinese AI firm faced intense backlash from developers who found out that their data was being uploaded to cloud servers without their consent.

Policy Watch

OpenAI released a proposal on Monday that would create international coordination around AI safety. In a blog post, OpenAI called for national AI safety institutes, such as the U.S. Commerce Department’s Center for AI Standards and Innovation, to set standards around areas including model evaluation, risk assessment and incident reporting.

Deals and Debuts

See The Information’s Generative AI Database for an exclusive list of private companies and their investors.

SoftBank is seeking the equivalent of more than $11 billion in what would be one of the biggest junk bond sales ever, to help fund a follow-on investment in OpenAI expected to close next month. It is looking to issue $10 billion of dollar securities across three maturities and €1 billion of euro debt across two, with the deal possibly pricing on Thursday. The yields being discussed—roughly 9% to 10% on the dollar tranches—would be records for SoftBank in those currencies and maturities.

Morphotonics, a nanoimprint lithography company, raised €40 million from 3M Ventures, Innovation Industries, BOM and Invest-NL.

FintechOS, a London company whose agentic operations software lets banks and insurers create financial products, raised $28 million in combined equity and debt financing from Bek Ventures, IFC, Cipio Partners and Molten Ventures, alongside a debt facility from Santander CIB.

GENISOM AI, a Chinese robotics company, raised several hundred million yuan in a Series B funding round led by Stone Venture, the Abu Dhabi firm.

Footprint, a New York company using AI to prevent financial crime, raised $25 million in a Series B funding round led by QED Investors.

Spott, a Belgian company creating AI software for recruitment and hiring, raised $21 million in a Series A funding round led by Balderton Capital.

Integral, a Berlin company whose software works as an automated accounting firm for small businesses raised €18 million in a Series A funding round led by Mosaic Ventures and Reid Hoffman.

Corridor, a New York company whose AI software picks health insurance plans for small businesses, raised $16 million in a seed funding round led by Bain Capital Ventures.

Iambic Therapeutics, a San Diego company that designs cancer drugs using its own AI models, filed to go public on the Nasdaq Global Select Market under the ticker symbol IAM. It is backed by Nvidia and the Qatar Investment Authority and has raised about $461.8 million since it was founded in 2019 as Entos.

Oura is looking to sell $2.2 billion in shares during its initial public offering, the company said in its updated S-1 filing on Monday. Oura, which makes smart rings that track biometrics including sleep quality and heart health using AI, could reach a $15.6 billion fully-diluted valuation from the sale.

Alibaba on Tuesday unveiled a powerful new AI chip for training and running models, highlighting the rapid progress in China’s domestic semiconductor capabilities. During its annual Apsara tech conference, Alibaba said its new AI chip, Zhenwu V900, delivers three times the performance of its predecessor, Zhenwu M890, which was released in May.

OpenAI on Monday announced a new independent advisory group, called the Advisory Group on Mathematics and Artificial Intelligence, meant to give mathematicians more input into the company’s math-oriented research.

Mainstay, a San Francisco company spun out of Opendoor in 2024 whose software automates the data and paperwork of residential property deals for large institutional investors, is acquiring Truelist, a Seattle company founded in 2024 whose AI helps brokers and agents create and distribute property listings across the many separate listing systems in US real estate.

DexCare, a Seattle company spun out of the Providence hospital system that helps healthcare organizations fill appointment slots and manage referrals and insurance rules, is acquiring Mila Health, a nearby company whose AI makes the phone calls and sends the texts to patients trying to book medical appointments.

Searchable, a London and New York company that tracks how often brands get named and recommended inside AI chatbots’ answers and advises them on how to show up more, acquired Meridian Tech, a New York company doing the same thing, in a seven-figure deal.

AbbVie and Iambic announced a multi-year partnership to use Iambic's AI to find and develop small molecule drugs in immunology, neuroscience and oncology. The collaboration will use Enchant v3, Iambic‘s new 41-billion-parameter model trained on more than 6,000 molecular properties, which is designed to assess all of a candidate drug’s properties at once rather than testing them one at a time.

Thank you for reading the AI Agenda Newsletter! I’d love your feedback, ideas and tips: [email protected].

If you think someone else might enjoy this newsletter, please pass it forward or they can sign up here.