16 October, 2024
Meta won't say whether it trains AI on smart glasses photos
Meta’s AI-powered Ray-Bans have a discreet camera on the front, for taking photos not just when you ask them to, but also when their AI features trigger it with certain keywords such as “look.” That means the smart glasses collect a ton of photos, both deliberately taken and otherwise. But the company won’t commit to keeping these images private.
We asked Meta if it plans to train AI models on the images from Ray-Ban Meta’s users, as it does on images from public social media accounts. The company wouldn’t say.
“We’re not publicly discussing that,” said Anuj Kumar, a senior director working on AI wearables at Meta, in a video interview with TechCrunch on Monday.
“That’s not something we typically share externally,” said Meta spokesperson Mimi Huggins, who was also on the video call. When TechCrunch asked for clarification on whether Meta is training on these images, Huggins responded, “we’re not saying either way.”
Part of the reason this is especially concerning is because of the Ray-Ban Meta’s new AI feature, which will take lots of these passive photos. Last week, TechCrunch reported that Meta plans to launch a new real-time video feature for Ray-Ban Meta. When activated by certain keywords, the smart glasses will stream a series of images (essentially, live video) into a multimodal AI model, allowing it to answer questions about your surroundings in a low-latency, natural way.
That’s a lot of images, and they’re photos a Ray-Ban Meta user might not consciously be aware that they’re taking. Say you asked the smart glasses to scan the contents of your closet to help you pick out an outfit. The glasses are effectively taking dozens of photos of your room and everything in it, and uploading them all to an AI model in the cloud.
What happens to those photos after that? Meta won’t say.
Wearing the Ray-Ban Meta glasses also means you’re wearing a camera on your face. As we found out with Google Glass, that’s not something other people are universally comfortable with, to put it lightly. So you’d think it’s a no-brainer for the company that’s doing it to say, “Hey! All your photos and videos from your face cameras will be totally private, and siloed to your face camera.”
But that’s not what Meta is doing here.
Meta has already declared that it is training its AI models on every American’s public Instagram and Facebook posts. The company has decided all of that is “publicly available data,” and we might just have to accept that. It and other tech companies have adopted a highly expansive definition of what is publicly available for them to train AI on, and what isn’t.
However, surely the world you look at through its smart glasses is not “publicly available.” While we can’t say for sure that Meta is training AI models on your Ray-Ban Meta camera footage, the company simply wouldn’t say for sure that it isn’t.
Other AI model providers have more clear-cut rules about training on user data. Anthropic says it never trains on a customer’s inputs into, or outputs from, one of their AI models. OpenAI also says it never trains on user inputs or outputs through its API.
We’ve reached out to Meta for further clarification here, and will update the story if they get back to us.
UK data watchdog confirms it's investigating MoneyGram data breach
The U.K.’s data protection regulator has confirmed it’s investigating MoneyGram after receiving a data breach report from the U.S.-based money transfer giant.
The U.K.’s Information Commissioner’s Office (ICO), which requires that organizations report data breaches within 72 hours of discovering the incident, confirmed to TechCrunch on Friday that the watchdog had received a report from MoneyGram following a cybersecurity incident at the company.
“We have received a report from MoneyGram and will be making enquiries,” ICO spokesperson Lucy Milburn told TechCrunch.
The extent of a potential data breach at MoneyGram remains unknown. MoneyGram, the world’s second-largest money transfer provider, serves over 50 million people in more than 200 countries and territories each year.
MoneyGram has remained largely silent about the cybersecurity incident beyond a handful of updates posted to its X account.
The company’s website, which is now back up and running after almost a week offline, contains no mention of the cybersecurity incident, and MoneyGram has not responded to TechCrunch’s multiple requests for comment.
MoneyGram first confirmed the cybersecurity incident on Monday following three days of operational downtime, saying it “identified a cybersecurity issue affecting certain of our [sic] systems” and had taken some systems offline in an effort to contain the incident.
The outage forced both the company’s website and app offline, leaving customers unable to make in-person or online payments. The outage also affected MoneyGram partners, including the Bank of Jamaica and the U.K.’s Post Office.
The latest update from MoneyGram, posted on X on Thursday, says that customers can now “send and receive money through both our digital platforms and agent partners,” but adds that the company is still working to fulfill pending transactions.
The company says that its app is now live and available. When TechCrunch checked on Friday, MoneyGram’s app remained offline.
India's Star Health says it's investigating after hacker posts stolen medical data
Star Health and Allied Insurance, one of India’s biggest health insurance firms, is investigating a cybersecurity incident that allegedly leaked sensitive data associated with its customers, including their medical records.
The Chennai-headquartered insurance giant told TechCrunch that a “forensic investigation” is ongoing after data allegedly stolen from the company was shared online.
A hacker group recently created chatbots on Telegram to leak alleged personal data of Star Health’s policyholders, including their full names, phone numbers, and home addresses, as well as medical reports and insurance claims. The data also appeared to include copies of ID cards and individuals’ tax details.
Reuters first reported the Telegram chatbots leaking the alleged Star Health customer data. Star Health says it has provided coverage to 170 million individuals to date.
The hacker group created a website to share the data with the links to the Telegram bots. The site, which TechCrunch has seen but is not linking to as it appears to contain sensitive personal information, also included a video allegedly showing screenshots and conversations between Star Health CISO Amarjeet Khanuja and the hacker group.
Star Health declined to comment when reached by TechCrunch with several questions about the incident.
“Given the circumstances, it would be premature for a listed entity to release a statement without completing a thorough investigation,” Star Health spokesperson Diana Monteiro said in an email.
Earlier on Thursday, Star Health said in a public notice in the Chennai edition of The Hindu newspaper, which TechCrunch has seen, that it was suing Telegram for hosting the chatbots. The insurer also named Cloudflare in its lawsuit for its role in hosting the hacker group’s website on its service.
As a result, the court issued interim injunctions to Telegram and Cloudflare to restrict them from allowing their platforms to be used by the hacker group to share Star Health’s branding in any form.
TechCrunch was able to verify that the hacker group’s website was inaccessible from certain internet providers in India, though the site was accessible from others at press time. Even when the website was blocked, it was redirecting to a web address hosted on a Cloudflare domain.
When asked by TechCrunch if it was aware of the internet blocks, India’s CERT-In said in a brief boilerplate statement that it is “already in process of taking appropriate action with the concerned authority.”
The insurer, which has more than 14,000 hospitals in its network and over 850 branch offices across India, has processed over $3.6 billion claims so far. It provides health, personal accident, and overseas and travel insurance.
A spokesperson for Telegram did not provide comment, when reached by TechCrunch on Thursday. Cloudflare did not respond to a request for comment.
Updated with response from CERT-In.
Northvolt lays off 1,600 workers, but it’s not the end for Europe’s battery champion
When is raising $14 billion not enough? When you’re a battery startup.
Northvolt, Europe’s attempt at building a competitor to Asia’s battery manufacturing powerhouses, announced on Monday that it had halted work on a factory expansion and laid off 1,600 employees, or about 20% of the workforce.
The company was planning to expand its Ett factory in northern Sweden to scale production to 30 gigawatt-hours annually. The expansion would have supplied cathode active material (CAM), a key component required to make completed cells. On September 9, the company also axed another CAM production site in Sweden. Without those factories, Northvolt will almost certainly have to buy it elsewhere, likely from overseas.
The cost cutting is the result of lower-than-expected demand growth, Northvolt said, as automakers trim their forecasts for electric vehicle production. Execution problems are probably also to blame. In June, the company was unable to fulfill an order for BMW on time, leading the German automaker to cancel the €2 billion contract. Northvolt did not immediately reply to TechCrunch’s request for comment, though it’s hard to see how that didn’t influence the company’s cost-cutting measures.
Ultimately, Northvolt faces two challenges.
For one, all battery startups face significant execution risk. Though batteries appear simple from the outside, the chemistry inside is fiendishly complex. It isn’t easy to develop materials that can store energy safely at high densities, that can be recharged at increasingly higher rates, and that can survive for more than a decade inside an automobile. Producing them at a massive scale only compounds the challenge. Just ask GM and LG what happens when you don’t get it right.
Northvolt has additional hurdles to surmount. It’s essentially building a copy of what Asian countries like China and South Korea already possess: a mature, scaled battery-manufacturing sector. Both China and South Korea have been working on it for decades, with consistent government support along the way. By comparison, Northvolt is only eight years old, and it only recently received substantial assistance from the EU and other governments.
The U.S. tried something similar nearly 20 years ago with A123 Systems. The startup pioneered production of lithium-iron-phosphate batteries, which stored less energy than other chemistries but were more durable and safer to charge. It started by selling to power tool manufacturers and then began courting automakers, who even in the late 2000s were projected to buy the sort of volumes that would support large-scale domestic manufacturing.
A123 was in the running to make battery packs for the Chevrolet Volt, but after losing out to LG, its only customer ended up being the first iteration of Fisker, which was also making a plug-in hybrid. After one of those cars caught fire during Consumer Reports’ testing, A123’s fate was all but sealed.
What those high-profile stumbles don’t reveal were the other obstacles A123 faced, most of which revolved around standing up a battery supply chain where there was none. Northvolt has been a bit more successful, in part because there is some political appetite to make it happen. But the Swedish company’s announcements about curtailing CAM production show it’s still not easy to accomplish.
The second challenge that Northvolt faces is that automakers, its key partners, haven’t been able to decide where they stand on EVs. After years spent talking up the transition to all-EV lineups, they’ve since backed off the most aggressive targets. Most automakers’ early forecasts proved overly optimistic, and they appear to have underestimated the amount they’d need to invest to produce successful products. In the face of weaker-than-expected tailwinds, they have plunged into developing hybrids and plug-in hybrids, which require far fewer batteries.
To succeed in early markets requires all players to have conviction. Automakers, parts manufacturers, and investors all need to have bought into an EV future. If any one of them blinks, they all suffer. Northvolt is feeling that pain today.
Does it spell the end of battery manufacturing in Europe or North America, where Northvolt has plans to expand? Hardly. Demand for EVs is still strong and growing. And because batteries are heavy and expensive to ship, it makes sense to produce them near EV factories. Strong incentives courtesy of the Inflation Reduction Act and the European Green Deal help tip the scales further. That doesn’t mean Northvolt can be complacent — it still has to prove it can execute. But by the time that gets sorted, it’s likely the market will be ready for it.
Internet surveillance firm Sandvine says it's leaving 56 'non-democratic' countries
Sandvine, the makers of surveillance-ware that allowed authoritarian countries to censor the internet and spy on their citizens, announced that it is leaving dozens of “non-democratic” countries as part of a major overhaul of the company.
The company, which was founded in Canada, published a statement on Thursday, claiming that it now wants to be “a technology solution leader for democracies.” As part of this new strategy, Sandvine said it has already left 32 countries and is in the process of leaving another 24 countries.
Sandvine did not name the 56 countries, apart from Egypt, where Sandvine promised to leave by the end of March 2025. For the remaining countries — including non-government customers in Egypt — the “end-of-service” date will be the end of 2025.
This change in the company’s direction comes after years of investigations by Bloomberg, which reported that Sandvine had sold its internet surveillance products to authoritarian regimes, including Belarus, Egypt, Eritrea, the United Arab Emirates, and Uzbekistan.
Sandvine said that it based its decision to withdraw from the dozens of countries on a review of its operations based on The Economist Intelligence Unit’s 2023 Democracy Index, which categorizes countries based on their “regime type.” The company also said it made this decision “in consultation with the U.S. Department of Commerce, the U.S. Department of State, and other key members of the U.S. government.”
Sandvine did not respond to a request for comment, asking the company to provide a full list of the countries that it has already left, and that it is planning on leaving.
Contact Us
Do you have more information about Sandvine, or other surveillance tech makers? From a non-work device, you can contact Lorenzo Franceschi-Bicchierai securely on Signal at +1 917 257 1382, or via Telegram and Keybase @lorenzofb, or email. You also can contact TechCrunch via SecureDrop.
Earlier this year, the U.S. Department of Commerce put Sandvine on a blocklist — technically known as the Entity List — accusing the company of selling its products to the Egyptian government, which used Sandvine “in mass web-monitoring and censorship to block news as well as target political actors and human rights activists.”
In the last few years, digital rights research group Citizen Lab published reports about Sandvine, which also exposed the use of the company’s technology in Turkey and Syria, where Sandvine’s equipment was allegedly used to redirect hundreds of users to spyware.
Sandvine’s overhaul appears to show that the pressure applied by the U.S. government actions against the company was effective, according to experts.
“For a long time, we knew about the harm but didn’t know what might effectively pump the brakes on out-of-control surveillance tech proliferation,” John Scott-Railton, a senior researcher at Citizen Lab, told TechCrunch. “Sandvine’s tailspin shows that the U.S. model, which includes sanctions, can have a direct positive impact.”
Ron Deibert, the director of Citizen Lab, agreed, telling TechCrunch that the Sandvine case “shows what can happen when you have careful evidence-based research, investigative journalism, and public interest advocacy combined with targeted and meaningful government regulations.”
In the last couple of years, the U.S. government has targeted other companies that sell surveillance technology. In 2021, the U.S. Commerce Department put NSO Group on the Commerce Department’s blocklist, effectively barring U.S. companies from doing business with the Israeli spyware maker, which sells its mobile spyware Pegasus. In 2023, the U.S. government put Intellexa, a consortium that makes the spyware Predator, on the same economic blocklist.
This year, the U.S. Treasury sanctioned Intellexa’s founder Tal Dilian and one of his business associates. These sanctions, which specifically targeted Dilian, rather than his companies, have caused other spyware makers to worry about getting in the U.S. government’s crosshairs themselves.
Why Lyft's CEO says 'it would be insane' not to go all in on bikeshare
In the summer of 2023, Lyft was contemplating the sale of its micromobility business after receiving strong interest from prospective buyers. Today, the ride-hail company is doubling down on its docked scooter and bikeshare operations, positioning itself as the ideal partner for cities looking to enhance their urban transportation networks.
Since acquiring shared bike and station provider PBSC Urban Solutions two years ago, Lyft has invested in building and deploying better quality e-bikes and docked e-scooters, as well as a new type of docking station that is solar-powered, modular and able to charge both bikes and scooters.
As CEO David Risher reviewed Lyft’s micromobility assets, he asked himself: “Do we sell the whole thing, or do we operate it?”
“E-bikes in particular are growing so fast globally. It would be insane not to take it on ourselves,” Risher told TechCrunch, noting Lyft’s e-bike rides increased 65% year-over-year through August to make up half of total rides. “So we said, let’s do this ourselves and bring it to the real standard of excellence that we have for our whole business.”
For Lyft, that means streamlining and integrating the two sides of its micromobility business – PBSC, which sells bikeshare-as-a-service to local operators and cities in 49 markets globally, and the eight Lyft-owned and operated bike and scooter share programs, like Citi Bike in New York City and Divvy in Chicago. The combined division will be renamed Lyft Urban Solutions, and Michael Brous, Lyft’s former head of operations, will head it up.
Risher noted the renewed focus on micromobility will also include a restructuring of its team and finances. A spokesperson for Lyft told TechCrunch less than 1% of staff on the tech side will be cut as the company shifts resources from R&D to sales, operations and deployment.
“We estimate that the ongoing benefit to the business of this restructuring is going to be on the order of $20 million a year, and it will be a net contributor to the business,” Risher said.
It’s still a small portion of the business, which in the second quarter saw revenue of $1.4 billion. But any amount helps to stay in the green – Lyft recorded an operating loss of $27 million in the quarter, much narrower than the year-ago operating loss of $159 million, and actually earned $5 million in net income, versus a net loss of $114 million in the year-ago quarter.
Lyft’s goal is to slowly update the hardware and software platforms in each city to eventually get to the point of standardization. And Lyft wants to do it in a way that focuses on a station-based, grid-connected approach. That means avoiding dockless scooters and bikes.
“We’re really focused on the docked version because it frankly helps cities organize themselves better,” Risher said. “There was a time where it was a free-for-all, let’s just put a bunch of junk on the street, but we thought we should do it in a way that works for riders but also works for cities.”
Docked bike and scooter-share also has the added benefit of giving companies like Lyft access to longer-term city partnerships, as stations end up becoming part of a city’s infrastructure.
Lyft’s dockless footprint is already small – the company operates scooters in Washington D.C. alongside its Capital Bikeshare service, as well as bikes and scooters in Denver. Lyft said in a blog post Wednesday that it would discontinue both of those services, and explore other options with Denver.
But, so as not to miss out on the dockless scooter craze, Lyft will continue to partner with Spin and, as of recently, Bird (which owns Spin) so riders can book directly through the Lyft app.
Lyft does not break out its micromobility business in quarterly or annual earnings reports, and the company would not share what percentage of its total revenue comes from shared micromobility. That said, the total number of rides Lyft recorded across micromobility and ride-hail hit 709 million in 2023, and about 56.7 million of those, or around 8%, were bike and scooter-share rides.
Fisker failed because it wasn't ready to be a car company
Two years ago, an employee at Fisker Inc. told me that the most pressing concern inside the EV startup was not whether its Ocean SUV would get built. Fisker was outsourcing the manufacturing of its first EV to highly respected automotive supplier Magna, after all. The startup’s November 2022 start-of-production target was aggressive, but not impossible for a company like Magna, which builds vehicles for the likes of BMW.
Instead, this person said, employees were increasingly worried that Fisker wouldn’t be ready to handle all the problems that come after a company puts a car on the road. They were worried the focus was all on building the car and not on the company.
The conversation stuck with me because Fisker founder and CEO Henrik Fisker had an automotive startup fail a decade ago for, arguably, this reason. That company, Fisker Automotive, got a hybrid sports car into the hands of a few thousand customers. But the company buckled soon after as it faced complaints about quality, the failure of its battery supplier, and a hurricane that literally sunk a ship full of vehicles.
The employee’s warning that the new Fisker was heading down a similar path was striking and ultimately prescient. Fisker filed for Chapter 11 bankruptcy protection this week after spending only just one year shipping its SUV to customers around the world. In large part, its undoing is directly tied to its inability to address the worries that employee raised in 2022.
This person wasn’t alone. Dozens of others who worked at Fisker have echoed this sentiment to me in conversations since, nearly all of them on the condition of anonymity because they feared losing their jobs or retaliation from the company. Those conversations informed stories I reported on — the Ocean’s quality and service problems, Fisker’s internal chaos, and decisions from Henrik Fisker and his co-founder, wife, CFO and COO, Geeta Gupta-Fisker, that dragged the company down.
Most all of them told me about how the lack of preparedness ran deep and permeated almost every division of the company, as I’ve previously reported for TechCrunch and Bloomberg News.
The software powering the Ocean SUV was underbaked. It contributed to the delay of the launch of the SUV, and it even kneecapped the very first delivery in May 2023, which Fisker had to turn around and troubleshoot shortly after handing it over. A similar thing happened when the company made its first deliveries in the U.S. in June 2023, when one of its board members’ SUVs lost power shortly after taking delivery.
The company shipped far fewer Ocean SUVs than it originally projected. Even after it lowered its target for 2023 multiple times, it still struggled to hit its internal sales goals. Sales employees have recounted stories of calling potential customers repeatedly in hopes of selling vehicles because so few new leads were coming in. Others wound up pitching in to sell cars even if they worked in completely different departments.
Many customers who did take delivery of their Ocean ran into problems like sudden power loss, trouble with the braking system, glitchy key fobs, problematic door handles that could temporarily lock them in or out of the car, and buggy software. (The National Highway Traffic Safety Administration has opened four investigations into the Ocean.)
Fisker struggled with the quality of some of its suppliers, and employees have said it did not build out a proper buffer of spare parts. This put extra pressure on the people in charge of trying to fix the cars as they ran into problems, and ultimately led to the company plucking parts from not only Magna’s production line in Austria, but also from Henrik Fisker’s own car. (Fisker has denied these claims.)
This whole time, lower- and midlevel employees went to great lengths to do what they could to help out the slow-growing customer base. One owner told me an employee took a phone call on their personal cell phone while at a funeral. Other employees relayed stories of workers doing company business while at the hospital. Many worked long days, nights and weekends — to the point where at least one hourly employee has filed a prospective class action over this very issue.
The company itself admitted on multiple occasions that it did not have enough staff to handle the influx of customer service requests. This was another place where workers from other departments pitched in. Some are even still fielding customer calls today, despite having left Fisker weeks or months ago.
Fisker struggled at the mundane-yet-serious work of being a public company, too. It lost track of around $16 million in customer payments at one point, thanks to messy internal accounting practices. It suffered multiple delays in its required reporting to the Securities and Exchange Commission. One of those delays allowed one of the company’s largest lenders to eventually take the reins in the final months.
Despite all this, Fisker is still touting its speed to market as an accomplishment as it begins the bankruptcy process. “Fisker has made incredible progress since our founding, bringing the Ocean SUV to market twice as fast as expected in the auto industry,” a nameless spokesperson said in a press release about the Chapter 11 filing.
This ephemeral corporate representative goes on to say that Fisker “faced various market and macroeconomic headwinds that have impacted our ability to operate efficiently.” While that is certainly true to an extent, there is otherwise no introspection about the myriad issues that got the company to this moment in time.
Perhaps that will surface in the Chapter 11 proceedings, where the company looks to settle its debts (of which it claims to owe between $100 million and $500 million) and offload or otherwise restructure its assets (totaling between $500 million and $1 billion).
What happens next will depend on how those proceedings go. Fisker always took an “asset-light” approach, likening itself to how Apple leveraged Foxconn to help build the iPhone into a global phenomenon. The problem with being asset-light is that it naturally means there is less to borrow against or sell when things break bad.
Magna has stopped production of the Ocean and expects a $400 million revenue loss this year as a result. It’s unclear how much progress Fisker made on its future products, the sub-$30,000 Pear EV and the Alaska pickup. The engineering firm that was co-developing these vehicles with Fisker recently sued the startup, calling the projects into question.
Fisker said in its press release that it will continue “reduced operations,” including “preserving customer programs, and compensating needed vendors on a go-forward basis.” In other words, it will continue to manage a bare-bones operation in case there is a willing buyer of the assets it’s putting up for sale in the Chapter 11 case.
A decade ago, the bankrupt Fisker Automotive did find a buyer. It ultimately morphed into a startup known as Karma Automotive, which is still nominally around today. There have been similar outcomes lately. Three other EV startups that recently filed for bankruptcy — Lordstown Motors, Arrival and Electric Last Mile Solutions — were able to sell off assets to peer companies in the space.
But the ultimate fate of this startup, and its assets, won’t change the fundamental problem: Fisker wasn’t ready to grapple with bringing a flawed car to market.
Emergence thinks it can crack the AI agent code
Yet another generative AI venture has raised a bundle of money. And, like the others before it, it’s promising the moon.
Emergence, whose co-founders include Satya Nitta, the former head of global AI solutions at IBM’s research division, on Monday emerged from stealth with $97.2 million in funding from Learn Capital plus credit lines totaling more than $100 million. Emergence claims to be building an “agent-based” system that can perform many of the tasks typically handled by knowledge workers, in part by routing these tasks to first- and third-party generative AI models like OpenAI’s GPT-4o.
“At Emergence, we are working on multiple aspects of the evolving field of generative AI agents,” Nitta, Emergence’s CEO, told TechCrunch. “In our R&D labs, we are advancing the science of agentic systems and tackling this from a ‘first principles’ perspective. This includes critical AI tasks such as planning and reasoning as well as self-improvement in agents.”
Nitta says that the idea for Emergence came shortly after he co-founded Merlyn Mind, which builds education-oriented virtual assistants. He realized that some of the same technologies developed at Merlyn could be applied to automate workstation software and web apps.
So Nitta joined forces with fellow ex-IBMers Ravi Kokku and Sharad Sundararajan, who also co-founded Merlyn Mind, to launch Emergence, with the goal of “advancing the science and development of AI agents,” in Nitta’s words.
“Current generative AI models, while powerful in language understanding, still lag in advanced planning and reasoning capabilities necessary for more complex automation tasks which are the provenance of agents,” Nitta said. “This is what Emergence specializes in.”
Emergence has a very aspirational roadmap that includes a project called Agent-E, which seeks to automate tasks like filling out forms, searching for products across online marketplaces and navigating streaming services like Netflix. An early form of Agent-E is already available, trained on a mix of synthetic and human-annotated data. But Emergence’s first finished product is what Nitta describes as an “orchestrator” agent.
This orchestrator, released on Monday, doesn’t perform every task itself. Rather, it functions as a kind of automatic model switcher for workflow automations. Factoring in things like the capabilities of and the cost to use a model (if it’s third party), the orchestrator considers the task to be performed — for example, writing an email — then chooses a model from a developer-curated list to complete that task.
Emergence’s orchestrator can also create workflows by “stitching” models together to perform tasks, after which it attempts to “verify” that workflows have executed successfully.
“Developers can add appropriate guardrails, use multiple models for their workflows and applications, and seamlessly switch to the latest open source or generalist model on demand without having to worry about issues such as cost, prompt migration or availability,” Nitta said.
Emergence’s orchestrator seems quite similar in concept to AI startup Martian’s model router, which takes in a prompt intended for an AI model and automatically routes it to different models depending on things like uptime and features. Another startup, Credal, provides a more basic model-routing solution driven by hard-coded rules.
Nitta doesn’t deny the similarities. But he not-so-subtly suggests that Emergence’s model-routing tech is more reliable and sophisticated than others; he also notes that it offers additional configuration features like a manual model selector, API management and a cost overview dashboard.
“Our orchestrator agent is built with a deep understanding of scalability, robustness and availability that enterprise systems need and is backed by decades of experience that our team possesses in building some of the most scaled AI deployments in the world,” he said.
Emergence intends to monetize the orchestrator with a hosted, available-through-an-API premium version in the coming weeks. But that’s only a slice of the company’s grand plan to build a platform that, among other things, processes claims and documents, manages IT systems, and integrates with customer relationship management systems like Salesforce and Zendesk to triage customer inquiries.
Toward this end, Emergence says it’s formed strategic partnerships with Samsung and touch display company Newline Interactive — both of which are existing Merlyn Mind customers, in what seems unlikely to be a coincidence — to integrate Emergence’s tech into future products.
Which specific products and when can we expect to see them? Samsung’s WAD interactive displays and Newline’s Q and Q Pro series displays, Nitta said, but he didn’t have an answer to the second question, implying that it’s very early days.
There’s no denying that AI agents are buzzy right now. Generative AI powerhouses OpenAI and Anthropic are developing task-performing agentic products, as are Big Tech companies, including Google and Amazon.
But it’s not obvious where Emergence’s differentiation lies, besides the sizable amount of cash out of the starting gate.
TechCrunch recently covered another AI agent startup, Orby, with a similar sales pitch: AI agents trained to work across a range of desktop software. Adept, too, was developing tech along these lines, but despite raising more than $415 million reportedly now finds itself on the brink of a bailout from either Microsoft or Meta.
Emergence is positioning itself as more R&D-heavy than most: the “OpenAI of agents,” if you will, with a research lab dedicated to investigating how agents might plan, reason and self-improve. And it’s drawing from an impressive talent pool; many of its researchers and software engineers hail from Google, Meta, Microsoft, Amazon and the Allen Institute for AI.
Nitta says that Emergence’s guiding light will be prioritizing openly available work while building paid services on top of its research, a playbook borrowed from the software-as-a-service sector. Tens of thousands of people are already using early versions of Emergence’s services, he claims.
“Our conviction is that our work becomes foundational to how multiple enterprise workflows get automated in the future,” Nitta said.
Color me skeptical, but I’m not convinced that Emergence’s 50-person team can outgun the rest of the players in the generative AI space — nor that it’ll solve the overarching technical challenges plaguing generative AI, like hallucinations and the mammoth cost of developing models. Cognition Labs’ Devin, one of the best-performing agents for building and deploying software, only manages to get around a 14% success rate on a benchmark test measuring the ability to resolve issues on GitHub. There’s clearly a lot of work to be done to reach the point where agents can juggle complex processes without oversight.
Emergence has the capital to try — for now. But it might not in the future as VCs — and businesses — express increased skepticism in generative AI tech’s path to ROI.
Nitta, projecting the confidence of someone whose startup just raised $100 million, asserted that Emergence is well-positioned for success.
“Emergence is resilient due to its focus on solving fundamental AI infrastructure problems that have a clear and immediate ROI for enterprises,” he said. “Our open-core business model, combined with premium services, ensures a steady revenue stream while fostering a growing community of developers and early adopters.”
We’ll see soon enough.