The AI Daily Brief: Artificial Intelligence News and Analysis - Are AI Acquihires Screwing Up Startups?

Starting point is 00:00:00 Today on the AI Daily Brief, are AI aquihers ruining the startup space? Before than on the headlines, we're going to have to wait a little bit longer to get OpenAI's new open model. AI Daily Brief is a daily podcast and video about the most important news and discussions in AI. All right, friends, quick announcements before we dive in. First of all, thank you to the sponsors of today's show, Blitzy, plum, and Superintelligent. To get an ad-free version of the show, go to patreon.com. slash AI Daily Brief. And if you were interested in sponsoring the show, shoot me a note at NLW at Breakdown.network. We've got a lot to discuss today, so let's dive in.

Starting point is 00:00:43 Welcome back to the AI Daily Brief Headlines edition, all the daily AI news you need in around five minutes. For the open model, enjoyers out there, I begin today with some bad news. Open AI has again delayed the release of their open model after finding some issues. Now you'll remember that back in June, Sam Altman took to Twitter to write, we're going to take a little bit more time with our open weight's model, i.e. expect it later this summer, but not June. Our research team did something unexpected and quite amazing, and we think it will be very much worth the weight, but needs a bit longer. Still, it appeared that it had been getting closer until Alman tweeted, we plan to launch our open weight model next week. We are delaying it. We need time to run additional

Starting point is 00:01:19 safety tests and review high-risk areas. We're not sure yet how long it will take us. While we trust the community will build great things with this model, once weights are out, they can't be pulled back. This is new for us and we want to get it right. Sorry to be the bearer of news, we are working super hard. Now, speculation began immediately on whether this delay was an alignment problem or performance issues. Open AI researcher Aden Clark commented, we're delaying the open weights model. Capability-wise, we think the model is phenomenal, but our bar for an open source model is high, and we think we need some more time to make sure we're releasing a model we're proud of along every access. This one can't be deprecated. Now, while some took this explanation on face value,

Starting point is 00:01:57 in other words, accepting that an open weights model did have a higher threshold when it came to safety issues, others thought there was something else going on. TOR taxes responded to Allman's post saying, I too think K2 is very good, Sam. What they're referring to is the release of a new benchmark topping open model out of China. A lab called Moonshot AI released a new model called Kimmy K2 that claims to beat open AI's models across multiple dimensions. On the coding benchmark, sweep bench verified, Kimmy K2 achieved better results than GPT41 and falling just short of clot four opus. On the AIME 2025 math benchmark, Kimi K2 achieved better results than GPT41, Claude4 Opus, and Gemini 2.5 Flash. Now, as always, the comparisons are a little cherry-picked, but some people reported some really positive results with the thing in practice. Pietro

Starting point is 00:02:44 Sharano writes, Kimi K2 is so good at tool calling in agentic loops, can call multiple tools in parallel and reliably, and knows when to stop, which is another important property. It's the first model I feel comfortable using in production since Claude 3.5 Sonnet. Menlo's Didi Das writes, China's Kimmy K2 is having its mini-deepseek moment. It is now number 14 on OpenRouter today ahead of Grok 4 and GPT41. This is a non-reasoning model which scores the highest on major EQ and creative writing benchmarks. Best model smells since Sonnet 3.5.

Starting point is 00:03:15 The model is designed in a similar manner to Deepseek v3, using a mixture of experts architecture to achieve a trillion total parameters. Still, when push comes to shove, the most craft. incredible insider reports that we've seen just don't make the connection. Uchenjin writes, rumors that Open AI delayed their open source model because of Kimi are fun, but from what I hear, the model is much smaller than Kimmy K2, super powerful, but due to some frankly absurd reason I can't say,

Starting point is 00:03:39 they realized a big issue just before release, so they probably have to retrain now. So I'm not sure what that big issue is, but for now we will just have to wait and see. Moving over to Embodied AI, Hugging Face racked up a half million dollars in first day pre-orders for their latest open-source robot. On Wednesday, Hugging Face started taking orders for the Reachery Mini Robot, one of two models they unveiled in May. This model is the small desktop model about the size of a teddy bear. The other yet to be released model is a full-sized humanoid made from readily

Starting point is 00:04:07 available components. Reachy Mini comes with microphones, speakers, and cameras, but lacks arms and legs, so the diminutive robot is mostly about prototyping interactive AI experiences. The bulk of the demos just show it moving its head to a beat, but that's compelling enough for many. TechCrunch actually referred to it as the Seinfeld of AI hardware, saying that the bots might do nothing in particular, but they're still captivating. I think ultimately we are just at the very, very beginning of the embodied AI in robotics era, and it makes sense to me that a developer-focused prototype like this actually might get some traction with the enthusiast and tinkerer community. Lastly today, Zuck's hiring spree continues as meta acquires voice startup Play AI. This company was

Starting point is 00:04:47 formerly known as PlayHT. Bloomberg disclosed that Meta has completed the deal. to buy out the small startup at an undisclosed valuation. The deal is solely focused on talent acquisition with an internal memo stating that the entire Play AI team will join Meta this week. The acquisition reinforces that voice is a big part of whatever Zuckerberg is cooking up. One of the earliest signings for Meta's superintelligence team was Sesame AI's machine learning lead, Yohan Schwalwick. You'll recall that Sesame blew up a little while ago for really figuring out life like AI voices, adding natural sounding pauses and voice ticks to make the experience less robotic. The Play AI team will report to Showick, and presumably form a subgroup within the new superintelligence division.

Starting point is 00:05:24 The memo heralding the team's arrival said that PlayAIs, quote, work in creating natural voices along with a platform for easy voice creation, is a great match for our work and roadmap across AI characters, meta AI, wearables, and audio content creation. Still, that is not the aqua hire that we are most interested in this week, but for that, we have to turn to the main episode. This episode is brought to you by Blitzy, the Enterprise Autonomous Software Development Platform with Infinite Code Context.

Starting point is 00:05:50 Blitzy is used alongside your favorite coding copilot as your batch software development platform for the enterprise seeking dramatic development acceleration on large-scale codebases. While traditional copilots help with line-by-line completions, Blitzy works ahead of the IDE by first documenting your entire codebase, then deploying over 3,000 coordinated AI agents in parallel to batch build millions of lines of high-quality code. The scale difference is staggering. Copilots might give you a few hundred lines of code in seconds, but Blitzie can generate up to 3 million lines of thoroughly vetted code.

Starting point is 00:06:21 If your enterprise is looking to accelerate software development, contact us at blitzie.com to book a custom demo or press get started to begin using the product right away. Today's episode is brought to you by Plum. You put in the hours, testing the prompts, refining JSON, and wrangling nodes on the canvas. Now it's time to get paid for it. Plum is the only platform designed for technical creators

Starting point is 00:06:43 who want to productize their AI workflows. With Plum, you can do. build, share, and monetize your flows without giving away your prompts or configuration. When you're ready to make improvements, you can push updates to your subscribers with a single click. Launch your first paid workflow at useplum.com. That's Plum with a B and start scaling your impact. Today's episode is brought to you by Super Intelligence, specifically Agent Readiness Audits. Everyone is trying to figure out what agent use cases are going to be most impactful for their business, and the agent readiness audit is the fastest and best way to do that.

Starting point is 00:07:17 We use voice agents to interview your leadership and team and process all of that information to provide an agent readiness score, a set of insights around that score, and a set of highly actionable recommendations on both organizational gaps and high-value agent use cases that you should pursue. Once you've figured out the right use cases, you can use our marketplace to find the right vendors and partners, and what it all adds up to is a faster, better agent strategy.

Starting point is 00:07:41 Check it out at B-Super.aI, or email agents at B-super.a.I. to learn more. Welcome back to the AI Daily Brief. Today we are talking about the aqua-hire of windsurf by Google. And while at first glance, this might strike you as a sort of insidery baseball story, right? Something that is about the machinations of Silicon Valley and how these sort of deals come together. This is the type of insider baseball story that has incredibly long tale of implications. Part of the story tells us where we are in the realm of vibe coding.

Starting point is 00:08:13 and agentic coding more broadly, there may be implications for OpenAI and specifically OpenAI's deal with Microsoft. More broadly, however, this is part of a trend that is upending the very fabric of how startups work in ways that very honestly could make employees think twice about working for young companies. So let's talk about WinSurf, what they are, what happened, how it all went down, and try to understand what it all means for all of us. Now, for those who are not familiar, Winserve is one of the leaders of the vibe coding slash AI coding assistants that quickly became the hottest sector in AI this year. If you have bolt and lovable on one end of the spectrum and cursor on the other, this is certainly closer to the cursor end of the market, being an

Starting point is 00:08:54 AI coding assistant in IDE for knowledgeable programmers rather than a vibe coding product for normies and non-technical people who can now speak in code. At the same time, the line between these two ends of the spectrum is getting a lot blurrier, and what's important to understand is that every major AI lab feels very strongly that they need to stand up their own AI coding platform as a matter of urgency. Now, that appeared to be the logic when news broke that OpenAI had agreed to buy Windsor for $3 billion back in May. OpenAI did release their own coding tool called Codex a week later, but acquiring an established product alongside with its user base and talent obviously has a lot more velocity. AI coding is absolutely the biggest land grab for an AI use

Starting point is 00:09:32 case since chatbot's post-chatGBTGBT. Now, it was very clear that OpenAI had put priority in this area. Before settling on the windsurf deal, they had apparently tried to acquire cursor as well. Now, in advance of the acquisition, it's not exactly clear how strong a position windsurf was in. On the one hand, they had reportedly grown to 100 million ARR, but there were also reports that they were trying to raise another round at a $3 billion valuation. And the reality is, is that even an extremely well-capitalized company in this space was going to just be a sales from all sides. You've got bottoms-up insurgencies from the vibe coding platforms. You've got all the other agentic IDEs coming at you. And then, of course, you have all the hyperscalers and foundation model

Starting point is 00:10:11 companies themselves who are in a very sort of frenemy relationship with all of these platforms in that their models power them, but they also want to own those customers directly. Whatever the case, the option of going it alone got a lot harder in early June when Anthropic cut windsurf off from direct access to Claude models. Now, this seemed to be in direct response to the news of OpenAI's acquisition. Said Anthropic co-founder and chief science officer, Jared Kaplan, we're really just trying to enable our customer. who are going to sustainably be working with us in the future. I think it would be odd for us to be selling Claude to OpenAI.

Starting point is 00:10:43 And while Winsurf was able to figure out how to get access for its customers to Claude models via third parties, this almost certainly put cost and operational pressure on the startup. As time dragged on, more and more people noticed that we just hadn't really heard anything about the acquisition since that first wave of rumors. That was until the middle of June, when we started to get reporting that the Winsurf deal had become a major sticking point in Open AIs negotiations. with Microsoft. The tech giant holds the deciding vote on whether OpenAI can convert into a

Starting point is 00:11:12 for-profit public benefit company. They reportedly want to retain the right to access all of OpenAI's IP, arguing that that should extend to WinServe if they were acquired. Indeed, the dynamic is such a significant part of the story that many outlets are taking the view that Microsoft effectively killed the OpenAI WinSurf deal. For their part, WinSurf were themselves also uncomfortable with giving Microsoft full access. And indeed, in retrospect, last week's announcement that Microsoft had partnered with replet on an integrated vibe coding service was a pretty big sign that the deal was heading south. Now, technically, the catalyst for this all falling apart was an exclusivity window around the OpenAI deal expiring, which allowed Google to swoop in. The terms of the new deal are that Google

Starting point is 00:11:52 is hiring Winsurf CEO of Aaron Mohan, co-founder Douglas Chen, and some of Winsurfs R&D employees. They'll pay $2.4 billion to license Winsurf IP but won't make a formal investment in the company. In other words, this is very clearly another wonkily structured aqua-hire deal that circumvents or is at least meant to circumvent antitrust regulations. Google's new employees can start immediately, and Winsurf will continue to operate as an independent company. All of which sounds fine theoretically if weird, right? Maybe there's a question around whether this actually passes regulatory scrutiny, but whatever,

Starting point is 00:12:27 if it works, it works. Except for the fact that there are a whole slew of employees who are not moving over and who it's not clear are actually going to get the benefits of any of their hard work over the past few years. Natasha Mosca Rana's from the information writes, employees with vested shares will receive cash, employees who joined less than 12 months ago are not vested and won't get payouts under the current terms, Winserve negotiated to keep $100 million on its balance sheet, company will shift focus to enterprise customers, remaining company will now be employee owned. Now the real sticking point is how the deal impacts employees. Entrepreneur Dave Pack wrote, a few weeks ago I was thrilled

Starting point is 00:13:01 for a buddy at Windsurf when the Open AI acquisition was announced. I joked in our group chat that he'd be picking up the tab on the next boy's trip. Now with Google's Aqua hire, the news is devastating. Here's what I've gathered. The top 30 AI engineers in leadership are going to Google. Existing employees are getting nothing. Even early team members with significant vested equity are reportedly receiving peanuts. I was DM'd by several who asked to remain anonymous. The company still has a massive cash balance but is gutted. Now Dave goes on to explain why this is such a problem. Early startup employees are are the people who take real risks, leaving stable jobs, accepting lower salaries, buying into the dream that equity might someday mean something. They work more hours, they take on more stress, and yes,

Starting point is 00:13:39 they sign up for the possibility of a big win. When that big win actually happens and people get cut out, it breaks trust in the whole system. A $3 billion plus exit is the dream scenario. This is when things go right. Calling it an aqua hire to dodge regulatory scrutiny while stiffing your team is just greed disguised as compliance. As a founder, this will hurt us dramatically. Hiring was already getting harder. Post-ZERP, the salary gap between startups and big tech did shrink, however, overfunded startups made employees skeptical of equity. I hear more and more early employees negotiating for less equity in favor of cash and have experienced that myself.

Starting point is 00:14:11 They think it's just a nice bonus if things work out and have been trained to think it's worth $0. I don't know the solution here, but there is a ton of money to play with on a $3 billion exit. Even a $250K to $500K bridge for every employee would make a massive difference. There's enough to do that and more. If you're a founder or a VC, this is your wake-up call. If this becomes the norm, startups will be staffed by mercenaries, not missionaries. Equity will mean nothing. The model breaks.

Starting point is 00:14:35 Stop being a short-term thinker and squeezing every last dollar out of this deal. And this really was the tone all around the internet on this. Jordy Hayes of TVPN seemed to confirm what David had been writing about. He said the founders and dozens of engineers are going to Google. This group, along with the preferred shareholders, will be sharing the $2.4 billion headline number. The exact split is unknown, but investors are making some money on the deal, and the founders, plus the select group are making a ton. That brings us to the hundreds of employees that aren't going to Google. From what I've heard,

Starting point is 00:15:01 they're all getting screwed regardless of their vesting status. The consolation prize is they now own 100% of the original company. WinServe's leadership is making the argument that this is a win for those that aren't joining Google. Their claim is that Winserve still has a meaningful amount of revenue and a solid balance sheet. But Winserve will now be facing intense competition from not only its former founders and engineers who are now at Google, but every other company in Kodgen that they were already competing with slash losing to, Cursor, Anthropic, etc. Given the Google now has a license to the core technology, it's safe to assume that windsurf will struggle and on a longer time horizon will be a zero. Jordy continues, the structure appears to be very similar to Google's

Starting point is 00:15:37 deal with character AI. Google effectively acquired Noam Shazir and left character AI employee owned. The difference there is that Google had no desire to compete in AI companionship. They just want a nom and some key people. WindSurf is left in a much worse position. My read is that the windsurf leadership team was desperate to find a way out and facing competition from the labs and cursor structured a deal to benefit themselves. Why they thought they could structure a deal like this and get away with it is anyone's guest. At this point, I imagine all the parties involved are scrambling to try to find a resolution because if things stay as they are, it will be a massive stain on the industry. Now, if that represented the common sentiment, there were some who pleaded for caution.

Starting point is 00:16:15 Jordi's co-founder and co-host at TBPN, John Coogan wrote, My Steelman here is that, one, employees who haven't reached a one-year vesting cliff don't have that strong acclaim around, I built this with sweat and need a liquidity event, and we don't know the tenures have all left behind employees. Two, the real culprit here might be FTC antitrust. This is a hyper-competitive market and Google still feels like they can't just do a normal acquisition. Three, new facts might come out. Now, the vesting cliff that he's talking about for those of you who don't work in startups is that usually when you are at a startup, the standard sort of arrangement with your equity is that it is distributed over a four-year vesting period. That means that you don't

Starting point is 00:16:50 just get whatever percentage of the company you're going to own or whatever set of stock options you're going to own all at once, you're incentivized to stay there for a period of time. The one-year cliff is the standard term at which any of your equity is realized. So if you only stick around for seven or eight months, you don't get anything. What Coogan here is saying is that it could be that a lot of those employees that were, quote, unquote, left behind are people who have only been there for six or seven months, which, while yes, they did work hard, maybe doesn't entitle them to that much of this exit. Now, there's tons of debates to be had there.

Starting point is 00:17:20 I've been fortunate enough to be in positions where CEOs have had some version of partially or fully accelerated vesting as part of an exit like this. So it's not like they're locked into doing nothing even if they haven't hit their cliff. But John is pointing out that the brutality of this may be a little bit different if we dig into the details. Now, Jordy did later update and say, updated belief is that this was almost entirely a comms problem, i.e. Google made founder-to-team messaging difficult, and the team left behind will get a fair outcome in the end after distributions. Alex Cohen gave a version of that, saying, If I was one of the employees stuck holding the bag at Windsurf, I'd simply vote to shut down the shell company

Starting point is 00:17:56 and pay the remaining 200 of us 500K each off the balance sheet. Modern problems require modern solutions. Now, clearly this is part of a larger trend. Like scale AI's 49% stake sold to meta, that really seemed to be all about getting CEO Alexander Wang to lead their superintelligence team, the Google Character AI deal that we talked about, which was $2.7 billion for a couple,

Starting point is 00:18:18 a couple of founders and 30 staff. Then there was the inflection deal, where Microsoft spent $650 million to hire CEO Mustafa Silliman and licensed the models while leaving inflection nominally independent, and Amazon's aqua hiring of adept and co-variant also follows this model. I think it's fair to say that these deals have not left the remaining company in a particularly strong position. Character AI may be the best positioned of them, but they've still struggled to make a lot of advancements or access additional funding. On the other end of the spectrum, it's been widely reported that Scale AI is bleeding customers and might not be able to compete post the meta deal. I would say at this point, most people's bet is that WinSurf follows the negative

Starting point is 00:18:56 model. Gurgolior-Ros writes, sucks for anyone who joined WinSurf in the last 12 months. Also, we can say that WinSurf as a company and product is 99% dead in 12 months. Enterprise customers should move off ASAP. This company will likely go bankrupt in that much time, because it'll be hard and possible to raise more funding. And part of the issue in this area specifically is the growing recognition that in the coding space there's very little moat. Netflix engineer Jake writes, windsurf is absolutely dead in the water, and while cursor hangs on for now,

Starting point is 00:19:24 all value is going back to the models. He shared a post from Hacker News that wrote, in my opinion, other than the Microsoft IP issue, I think the biggest thing that has shifted since this acquisition was first in the works is Claude Code has absolutely exploded. Forking an IDE and all the expense that comes with that feels like a waste of effort,

Starting point is 00:19:40 considering the number of free and open source CLI agentic tools that are out there. Now, what he's referring to is this sort of frenemies dynamic that I was articulating before, where all of these models are powered by Claude, but Claude is still very much competing for those end customers as well. Zooming out just beyond the implications for agentic coding, it really appears that this is a new profile of acquisition. The ECHF of Category VC calls it the Blitzhire acquisition and believes we're going to see a lot more of these deals. To get specific, in a normal

Starting point is 00:20:07 acquisition, either the entire company is bought out or the employees in IP are acquired leading the company to be shut down. These new blitz hires leave a shell company behind that may or may not continue. Many investors laid the blame for this at the former FTC chair, Lena Kahn, and her aggressive pursuit of antitrust enforcement during the last administration. Martin Casado of Andresen Horowitz posted, there's an irony that Lena Kahn's activism has resulted in deals that are far worse for everyone except the people she was targeting. Still, Vili thinks that these blitz hire acquisitions would have happened anyway due to the pace of AI advancement. If Google had done the deal as a normal acquisition, it could have taken up to

Starting point is 00:20:40 a year for FTC approval, even if it was just a rubber stamp. Going this route avoids the need to notify the FTC at all, with Google instead dealing with the risk of a future investigation. Billy wrote, A year in the world of AI is eternity. Speed is everything and speed is the driver for this new blitz higher structure. It circumvents the antitrust review process, allows the acquirer to take over the key employees and IP of the startup and have the employees with a badge in their office the next day. And maybe that's true, but it is not without cost.

Starting point is 00:21:08 Sophie at NetCap Girl writes, Winsurf and other deals like it are proof that tech M&A being broken is a massive net negative for the whole ecosystem. Healthy capital markets are an essential part of the engine that drives innovation. I don't have the solution, but I can't imagine things can continue like this. John Ludig from Founders Fund sees a bigger change at work. He wrote that, just like the top 1% of startups drive the majority of VC returns, he believes that power law is now applicable to individual engineers as well. While big tech can spend tens of billions on GPUs, those investments mean nothing without the correct talent in place. Therefore, Lettig believes that the capital canon is shifting

Starting point is 00:21:42 towards talent. He suggested, hypercapitalist AI talent wars will rewrite employment contracts and investment norms, concentrate returns, and raise the bar for mission and capital required to create great new companies. Letting foresees a complete repricing of labor and a restructuring of employment agreements to look more like sports or entertainment contracts. Now, there is a whole additional dimension of this story, which is what it means for Open AI. The company has obviously been assailed by the hypercapitalist AI talent wars with meta coming at them, complications with they're deal with Microsoft, losing out on this acquisition. But that, I think, is a subject for another show. Ultimately, what we have here is an example of just the hyper-competitive dynamics

Starting point is 00:22:22 and the increased stakes of AI changing long-held norms. How it all shakes out ultimately remains to be seen. But what's very clear is that even when it comes to the startups that are building AI, the expectations of yesterday will not be the expectations of tomorrow. For now, though, that's going to do it for today's AI Daily Brief. Appreciate you listening or watching as always. time. Peace.

The AI Daily Brief: Artificial Intelligence News and Analysis - Are AI Acquihires Screwing Up Startups?

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.