Musely, a direct-to-consumer telemedicine platform, has secured over $360 million in non-dilutive capital from General Catalyst’s Customer Value Fund (CVF).

The company specializes in compounded treatments for skin, hair, and menopause care. Musely co-founder and CEO Jack Jia told TechCrunch that when CVF investors reached out to him last year, he wasn’t looking to raise capital.

That’s because Musely, which was founded in 2014 as a wellness community before pivoting to prescription skincare in 2019, has been cash flow positive for years, he said. Jia didn’t want to reduce his ownership in the company by selling off a chunk of it to VCs. They frequently approached him about a potential round and he consistently turned them down, he said.

But unlike traditional venture capital, CVF wasn’t looking to take an equity stake, nor was it offering a loan that would carry interest rate charges. Instead, CVF’s alternative financing is similar to a tiny revenue-share agreement: Companies with predictable revenue streams borrow capital, and then repay the funds along with a fixed, capped percentage of revenue it generates from the use of General Catalyst’s fund.

Although Jia was initially skeptical, he quickly realized CVF’s terms were more favorable than a standard bank loan and far less costly than a dilutive equity round.

“When I mathematically modeled it, I found this absolutely compelling,” he said.

While Musely has been growing its revenue on average 50% year-over-year and has served over 1.2 million patients, acquiring new customers for DTC brands like Musely can be very costly, Jia explained. “When you become a billion-dollar revenue company, you need another billion in order to grow to the next billion,” he said. “That’s why most of the DTC companies, if you look at the capital burn, it is huge.”

The funding from CVF solves this problem, providing Musely with a capital war chest to support its customer growth. The funding will support sales, marketing, and other customer acquisition efforts.

Musely joins a CVF portfolio that includes Grammarly, Lemonade, and Ro. The fund maintains its own distinct limited partners, and the capital it invests was not included in General Catalyst’s last $8 billion fundraise.

Unlike many of its peers, Musely has been remarkably capital-efficient. After raising $20 million from DCM and other investors in 2014, the company has not raised a single dollar of equity capital since, according to Jia. Musely allows patients to access prescription products through asynchronous consultations with board-certified dermatologists and OB-GYNs.

Enterprise AI teams are hitting a wall — not because their models can’t reason, but because the workflows underneath them were never built for agents. Tasks fail, handoffs break, and the problem compounds as organizations push agents deeper into back-office systems. A new architectural layer is emerging to address it: workflow execution control planes that impose deterministic structure on processes agents are expected to run.

One of the companies bringing this to the forefront is Salesforce, with a new workflow platform that turns back-office workflows into a set of tasks for specialized agents to complete. Users can upload their processes or use one of the set Blueprints provided by Salesforce, and Agentforce Operations will break it down for agents. 

Salesforce senior vice president of Product, Sanjna Parulekar, told VentureBeat in an interview that the problem is that many enterprise workflows are not built for agents. “What we’ve observed with customers is that a lot of times, the brokenness in a process is probably in your product requirements document,” Parulekar said. “So when that’s uploaded into a product, it doesn’t quite work. We can optimize it and cut out some things and replace it with an agent.”

Without this control panel layer, enterprises could risk deploying agents that increase cost rather than fix their workflow problems.

Making the workflow work for agents, not just humans

Enterprises deploying agents are learning a costly lesson: Their workflows were designed around human judgment gaps, not machine execution. Processes that evolved through years of workarounds — loosely defined steps, implicit decisions, coordination that depends on individuals knowing what to do next — break when agents are asked to follow them literally.

Even with all of an enterprise’s context at its fingertips, AI systems will have difficulty completing tasks if it is not clear what it’s supposed to do. 

Parulekar said her team found that focusing on what makes the process tick and breaking it down into more explicit steps and workflows makes the system more deterministic. Then, when platforms like Agentforce Operations introduce agents, those agents already know their specific tasks.  

“It forces companies to rethink their processes and introduces observability into the mix because of the session tracing model in the system,” she said.

Parulekar said human checks can be built into the system, so the process is more transparent.

What makes this approach different from other workflow automation offerings is that it doesn’t rely on agents to decide what to do next; the system does. Unlike more traditional automation tools that route tasks and agents on probabilistic decision-making, this enforces execution on a more pre-defined, deterministic structure.

The problem it introduces

Codifying a workflow doesn’t fix a broken one. If a process has flawed steps, encoding it for agents locks in the problem at scale. And once workflows are distributed across agents, the challenge shifts from execution to governance: who owns the process, who validates it, and how it evolves when business conditions change.

It puts the onus on teams to take a hard look at what works for them and what doesn’t.

Organizations need to consider that, along with the execution control plane offered by platforms like Agentforce Operations, someone should be made responsible for task completion and success. 

Brandon Metcalf, founder and CEO of workforce orchestration company Asymbl, told VentureBeat in a separate interview that the key to both humans and agents following a workflow is a shared goal. 

“You have to understand the goal or the agent or human won’t complete the task successfully,” Metcalf said. “Someone has to manage that outcome that has to be delivered. It can be a person or an agent.”

The bottleneck has moved. As Metcalf framed it, the question is no longer whether agents can reason through a task, it’s whether the workflow underneath them is coherent enough to execute. For enterprises that built their processes around human judgment and institutional memory, that’s a harder fix than swapping in a smarter model.

Servers operated by Ubuntu and its parent company Canonical were knocked offline on Thursday morning and have remained down ever since, a situation that’s preventing the OS provider from communicating normally following the botched disclosure of a major vulnerability.

Attempts to connect to most Ubuntu and Canonical webpages and download OS updates from Ubuntu servers have consistently failed over the past 24 hours. Updates from mirror sites, however, have continued to work normally. A Canonical status page said: “Canonical’s web infrastructure is under a sustained, cross-border attack and we are working to address it.” Other than that, Ubuntu and Canonical officials have maintained radio silence since the outage began.

A decades-long scourge

A group sympathetic to the Iranian government has taken credit for the outage. According to posts on Telegram and other social media, the group is responsible for a DDoS attack using Beam, an operation that claims to test the ability of servers to operate under heavy loads but, like other “stressors,” are, in fact, fronts for services miscreants pay for to take down third-party sites. In recent days, the same pro-Iran group has taken credit for DDoSes on eBay.

Last year, Spotify removed more than 75m spam tracks from its platform.

The world’s biggest music streaming platform, Spotify, does not ban AI-generated music, but does admit that it finds it hard to detect it.

In its latest attempt to tackle growing AI spam on the platform, Spotify is introducing a vetted artist verification badge to help users identify human-made music from AI-generated ones.

Fraudulent music distribution is especially an issue for the platform, whose total artist payouts have grown from $1bn in 2014 to $10bn in 2024. However, artist payout per stream has reduced on average since 2021, further incentivising spam music to increase earnings.

Artists who receive this verification badge are understood to show consistent listener activity and engagement, compliance with Spotify’s policies and signal a human artist behind the account. The company said that it would also look for off-platform presence such as concert dates and social media accounts.

It added that at launch, “profiles that appear to primarily represent AI-generated or AI-persona artists are not eligible for verification”.

The new badges will begin rolling out in the coming weeks, Spotify said. These will appear next to artist names in search, represented with a light green checkmark icon.

“In today’s music landscape, the concept of artist authenticity is complex and quickly evolving, and we’ll continue to develop our approach over time,” Spotify said in a blogpost on 30 April.

“At launch, we have ensured that more than 99pc of the artists Spotify listeners actively search for will be verified, representing hundreds of thousands of artists – the majority independent – spanning genres, career stages and geographies.”

The company already has a number of other features, including ‘expanded song credits’, ‘about the song’ sections and AI credits, which help listeners find more information about the artists they listen to.

The new standards for verification, according to the company, would be paired with human reviews to identify “real artists” behaving in good faith, Spotify said. It also has an AI impersonation policy, as well as mechanisms to “better” stop fraudulent music distribution.

Last year, Spotify said that it removed more than 75m spam tracks from its platform. It acknowledged that AI is used by bad actors and content farms to create deepfakes and spam to deceive artists, pushing “slop” into the ecosystem. Spam tactics also include mass uploads, duplicates, SEO hacks and artificially short track abuse.

The verification badge comes after publisher Sony Music requested the removal of 135,000 songs by fraudsters impersonating its artists on streaming services.

Meanwhile, direct-to-fan music platform Bandcamp took a more aggressive approach in January by outright banning songs “generated wholly or in substantial part by AI”.

“Any use of AI tools to impersonate other artists or styles is strictly prohibited,” the company said in a post on Reddit. Spotify, however, allows artist impersonations as long as consent is provided.

Last October, Spotify’s founder and CEO Daniel Ek stepped down from his role and became the company’s executive chair in January.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Orreco uses AI, computer vision and biomarker data to optimise athlete performance, predict injury risk and accelerate recovery, according to the company.

Galway-based athletic bio-analytics company Orreco is to participate in the third edition of Major League Soccer (MLS)’s Innovation Lab, the North American professional football league’s initiative striving to advance the next generation of sports technology.

Orreco uses AI, computer vision and biomarker data to optimise athlete performance, predict injury risk and accelerate recovery, and is “trusted by elite teams and athletes across major leagues worldwide”, according to the company.

The programme cites key areas of focus in attempting to evolve technology around football as fan engagement, media technology and on-field development. MLS said it has chosen five AI-focused companies for the lab to “deploy solutions designed to address real challenges across the soccer ecosystem, creating a direct pathway from innovation to implementation”.

“It’s a privilege to be chosen for this year’s MLS Innovation Lab cohort and to play a role in advancing both MLS and the global game,” said Dr Brian Moore, co-founder and CEO of Orreco.

The company said its team includes 20 PhDs and that the team has published more than 400 peer-reviewed scientific papers. It works with elite athletes “to optimise performance, accelerate recovery and extend playing careers” in sports such as football, basketball, ice hockey, American football, golf and motor racing.

The company said it was selected for the initiative – which will allow the participants to test and scale their technologies in real-world MLS game environments – “following a rigorous evaluation process which vetted hundreds of companies from around the world”.

The other participants chosen for the upcoming cohort are Springbok Analytics, which uses AI to transform scan data into detailed 3D muscle analytics; Fit:Match, a player identification and assessment platform that uses mobile devices, AI and computer vision to capture detailed body data; Advanced Image Robotics, which aims to make broadcast-grade sports video production more accessible using AI-powered robotic camera systems; and WMT AI Ticketing, which claims to optimise pricing and distribution in real time by analysing fan behaviour, demand and market dynamics.

Chris Schlosser, MLS senior vice-president of emerging ventures, said: “With each new cohort, MLS Innovation Lab continues to surface technologies that have the potential to meaningfully impact how the game is played, experienced and consumed.

“This group represents a strong mix of real-world AI technologies that can have immediate impact across athlete performance, television production and fan-focused innovation.”

Orreco was founded in 2010 by sports scientist Moore and haematologist Dr Andrew Hodgson. The company utilises applied physiology, biostatistics and cognitive computing to better inform coaches, medical teams and athletes with actionable insights that help improve performance.

In January, the company acquired British Olympian Jessica Ennis-Hill’s ‘Jennis’ women’s health and performance platform. At the end of 2025, Orreco raised $4m in a funding round that included participation from Enterprise Ireland and television personality and billionaire investor Mark Cuban, as well as athletes and existing investors.

Don’t miss out on the knowledge you need to succeed. Sign up for the Daily Brief, Silicon Republic’s digest of need-to-know sci-tech news.

Just last night, I was scrolling through endless Pinterest boards and online sites to find a dress for my best friend’s engagement party. I was deep into scrolling before deciding to just wear something I already have. So I went to my Photos app on my phone and started scrolling to see what formal outfits I’ve worn in the past for inspiration. 

If you find yourself doing this often (I do it at least once a week), a new Google Photos AI feature might help with this by cataloging the clothes you’re wearing in photos saved to the app. From there, it will organize your clothes into a digital collection, so you can style, mix and match and try on clothes virtually. 

The Google Photos wardrobe feature is rolling out this summer, first to Android and then iOS.

How Google’s AI-driven wardrobe feature works

The wardrobe feature uses AI to scan the photos in your camera roll to create a digital closet based on pieces you’ve worn in the past. With this saved collection, you can filter by category, such as “jewelry” or “tops” to find that one particular item. 

Google also seems to be taking a page from Pinterest with the ability to create digital mood boards. Rather than pulling out your entire closet and trying on 10-plus outfits that you send to your friends to see what they like best, you can use the wardrobe feature to mix and match items into outfit ideas that you can then save to a shareable mood board. You can save these mood boards for different categories or occasions, like “wedding guest” or “work outfits.” 

With the wardrobe feature, you’ll also be able to “try on” clothes virtually to save time getting dressed. You can select clothing items saved in your collection, then click “Try it on” for a preview of how it will look on your body. Something to keep in mind is that the AI doesn’t really know what size clothes are or how they’re cut, so it’s at best a rough approximation of how any particular article will fit on a particular person.

Last year, Google released an AI-powered try-on feature in Search. But that feature was only for clothing you were shopping for and did not already own. The technology works by having an AI image generation model like Nano Banana to generate a guess at what you might look like in those clothes. Google said it won’t use the images you upload for the try-on feature for AI training, use it for other Google services or sell it to third parties. 

CNET’s Abrar Al-Heeti tested the Search try-on feature last year and found it would, in fact, generate bare arms to show off a sleeveless dress. A similar feature on the Samsung Galaxy S26 and Google’s Pixel phones, called “Find the Look,” adds this function to Circle to Search. That means you can take a screenshot or a photo and get an idea of what you might look like wearing it.

Amjad Masad has been building Replit for a decade, but the last 18 months have been something else entirely. The AI coding assistant company went from $2.8 million in revenue in all of 2024 to tracking toward what Masad describes as a billion-dollar annual run rate.

At TechCrunch’s sold-out StrictlyVC event in San Francisco on Thursday night, we covered a lot of ground in a short time, beginning with the question everyone in the industry is asking right now: in a world where rival Cursor is reportedly in talks to be acquired by SpaceX for $60 billion, is Replit also bound to sell? We also got into Replit’s net revenue retention — a measure of how much existing customers expand their spending — which Masad says is reaching as high as 300%, his willingness to take Apple to court over what he called outright lies in its App Store battle with Replit, and the possibility of the company beginning to invest in its own customers.

On the question of independence, Masad was unambiguous. Unlike Cursor, which he said has been operating at negative 23% gross margins, he argued Replit has the economics to make that path viable — even if he stopped short of ruling out a sale entirely.

The following has been edited for length and clarity:

TC: Cursor’s reported SpaceX deal was the talk of the industry last week. What did you make of it?

AM: It’s kind of hard being an independent, smaller AI company that’s building on foundation models, especially if you’re burning a ton of cash. Part of the reporting suggested Cursor has negative 23% margins, and if you’re also wanting to invest in training models, that makes it incredibly hard to stay independent.

For us at Replit, partly because we target a different customer set, we’ve been able to run the business more rationally. We’ve been gross margin positive for over a year. We’re slightly more expensive, but we provide a lot more. Our audience tends to be mostly non-technical users who previously haven’t been able to create any software. We provide an end-to-end platform — from the prompt all the way to a deployed application that can scale. We handle security, databases, database migration. And we’ve been doing this long enough that we’ve built a lot of those primitives into the platform.

Is Replit for sale? I would assume you are talking with potential acquirers all the time; it’s your fiduciary responsibility.

Yeah. We have amazing partners, and they sometimes bring up these topics. But we’re going to try to stay independent. I would love for us to remain an independent company. We’ve been around for 10 years, before it was even accepted that you could make apps just from ideas. We were talking about creating a billion software creators back in 2018 at YC, and people sometimes actually laughed at that dream. Now that dream is possible, and we kicked off this revolution with our agentic coding experience in September 2024. It just feels like we can take it much further.

You work closely with Anthropic, Google, and OpenAI. If you had to rank them — who’s doing it best?

Anthropic is still undefeated on the core agentic loop. They have the best tool calling; the agent can stay coherent much longer. GPT-5 is catching up quickly. Google’s Flash family of models is just amazing on price-performance. If you want something fast and cheap, they’re actually beating open source right now. We use all three, and honestly I wouldn’t discount the newer labs either. Reflection AI is coming out with open-source models we’re hearing great things about. And the Chinese models are impressive — Kimi is as good as an Anthropic-generation model from January, so it’s only about three months behind.

When you’re in a bake-off for an enterprise deal, what wins it for you?

Most of our sales are inbound or organic — very product-led. We’ve acquired customers like Zillow and Meta purely through people adopting the product and then raising their hand to buy an enterprise plan. When it does go top-down and there’s a formal bake-off, we usually win on product. But even in cases where we might be missing a feature, once it hits the C-suite and the IT group, Replit wins on security. A lot of vibe-coding tools will generate a website and connect it to an external database — great products, but it makes security much harder, because the database is open to the public and you need to configure row-level security, which is especially difficult for non-technical builders. Replit being full stack, with the database built into the project and not open to the public — that makes the app inherently more secure.

We also spent 10 years battling crypto scammers and hackers, so our cybersecurity function is as good as a dedicated cybersecurity startup. Every time you deploy an app on Replit, we create an entirely new isolated project on Google Cloud. We inherit Google’s security model.

Can we talk about churn? How long do you hold onto customers if the best prototypes eventually get rebuilt into a company’s existing stack?

Churn is very, very low, and net retention is incredibly high — 300% in some cases. What we actually hear from customers is that when engineers get nervous and try to rebuild an app into their own stack, they often make it worse. Once enterprises get comfortable with the full Replit stack — especially when we set up a single-tenant environment for them — they keep the apps on Replit. Bain & Company, for example, replaced Tableau and Power BI with Replit and Databricks.

There’s a growing concern about AI bloat — non-technical users generate far more code and burn through far more tokens. That’s good for you [given your usage-based fees]. What about your customers?

We don’t have a lot of regrettable spend. Enterprises are very ROI conscious, and they tell us about the returns they’re getting. For the most part they feel the investment is totally worth it — often one, two, three orders of magnitude. If they spend $100,000 a month with Replit, they’re usually generating $2 million, $3 million, $10 million in some kind of return.

Let’s talk about Apple. Another rival, Lovable, just got an app-building app approved by the App Store this week. Replit has been in App Store purgatory, with Apple blocking your updates for months. How much does that hurt you?

It’s not life or death — we could lose the app and it wouldn’t do anything meaningful to our business. But it’s an app people genuinely love. We’ve been on the App Store for four years. Kids in underprivileged communities learn to code on Replit on their Android devices. Executives use it in meetings.

The reason Replit got blocked when others weren’t, we believe, is that Replit makes iOS apps. When we launched that capability in December, there were charts going around showing how many apps were getting into the App Store through us. We think Apple feels threatened by that.

Apple’s stated reason is that you’re downloading new code to the device [after the approval process], which violates their guidelines.

That’s a lie. And we can prove it in court if we have to.

Is that going to happen?

I hope not. I’m a fan of Apple, and I’d love to collaborate and build something great together. We’re happy to send customers to Xcode [Apple’s own development environment]. But you can’t run a marketplace that a billion people have access to and make decisions that are discriminatory or based on whims.

Just wondering if, like Nvidia, OpenAI and others, you’re thinking about investing in your own customers in exchange for equity.

We’ve thought a lot about it, and it is a consideration. I’ve personally invested in a few startups that started on Replit before they made any money. Some of them, like Magic School — a teacher decided to take his time during COVID to learn a little bit of vibe coding and built an AI app for other teachers. He found this problem that in America, we burn out a lot of teachers. He wanted to use AI to reduce the workload. He did that, and he made $20 million in the first year. Other companies that started on Replit, I think, are valued at half a billion dollars. The entrepreneurship happening on Replit right now is genuinely exciting. We integrated with Stripe a few months ago, and the transactions flowing through Replit are growing triple digits month over month. Pretty soon, our customers will be making more revenue than we are.

You can watch our full conversation with Masad below:

Microsoft Word is getting an AI legal agent, which sounds helpful until you remember how badly this has gone before. The new Legal Agent can review contracts, suggest edits, compare versions, and flag risky clauses inside Word. On paper, these features sound quite useful and convenient, however, cases of generative AI tools hallucinating and inventing entire cases, citations and quotes from thin air have dragged some real people in real court trouble before.

What can Microsoft’s Legal Agent do?

Microsoft says Legal Agent is available through Copilot in Word for users in its Frontier program in the U.S. It currently works on Word for Windows desktop. There is no separate app or installation required, though some users may need to restart Word before the agent appears.

Legal Agent is meant for contract and document review. Microsoft says it can check a contract clause by clause against a legal playbook, review a full agreement, compare different versions, flag risks and obligations, and suggest edits with tracked changes. It is also keeps the original formatting, tables, lists, and negotiation history intact.

The company is also trying to avoid the obvious nightmare scenario for its users and itself. The feature has built-in safeguards like providing citations linked to source language, so reviewers can check suggestions before using them, along with clear disclaimers that it does not provide legal advice, may produce inaccurate content, and still requires review by a qualified legal professional before anything is relied on.

Why should lawyers still be nervous?

There is already precedent for AI going rogue in legal settings as two New York lawyers were sanctioned in 2023 and ordered to pay a $5,000 fine after submitting a court filing that included fake cases generated by ChatGPT. Michael Cohen, Donald Trump’s former lawyer, also admitted that he unknowingly gave his attorney fake case citations generated by Google Bard. While Cohen was not sanctioned, the judge still called the episode embarrassing and stressed the need for skepticism when using AI in legal work.

These are not isolated cases as Judges have questioned or disciplined attorneys in multiple instances involving AI-assisted filings, and one French data scientist and lawyer identified hundreds of court documents containing fake citations and nonexistent references over the past year.

The bigger problem is that hallucinations remain unresolved. AI chatbots can still produce answers that sound confident while being partly or completely wrong. In legal work, that is especially dangerous, because a made-up citation or invented case can end up in a filing and create serious consequences.

Microsoft has put many safeguards on Legal Agent to prevent these issues, however, the lesson is already written in court records. AI can speed up legal work, but the responsibility of fact checking still falls on the lawyer.