Google’s Big AI Test Comes Next Week

Alright friends, quick announcements before we dive in. First of all, thank you to today's sponsors, KPMG, Granola, Bolt, and Section. To get an ad-free version of the show, go to patreon.com slash AI DailyBrief, or you can subscribe and Apple Podcasts. To learn more about sponsoring the show, send us a note at sponsors at aidaybrief.ai. While you are on aidaybrief.ai, check out our careers page.

I am hiring a full-time growth engineer that is someone who is engineering growth, not a developer by training necessarily, although you will be building lots. But it is a very cool way to be a part of this ecosystem. And one final note, Enterprise Claw Cohort 3 is currently enrolling. You can find out about that either from our website or at enterpriseclaw.ai.

Today we kick off with a follow-up of a story that we have been watching this week, which is the public market debut of Cerebrus. TLDR, the company, delivered a massive first day of trading, kicking off a potentially significant IPO season for AI companies. Now, heading into the big day, Cerebrus had upsized their share offering, raised their price, and still, after all of that, ended up pricing the sale above their guided range.

Red hot demand during the private roadshow flowed through into public markets on Thursday, with the opening trade seeing the stock double in price before settling into a sixty eight percent gain at the end of trading. Cerebrus began the day as a$40 billion company, touched$100 billion for a minute, and now has a market cap of$66. Now, predictably, for a frothy IPO, and I think that is a certainly reasonable frame, the launch brought out a fair number of contrarians on the stock.

CNBC's Jim Kramer warned his audience to tread carefully, arguing that the price had detached from fundamentals. While there might be a situation in the future where I can recommend Cerebrus, he said, I just can't even come close to justifying the valuation up here given how much it's already run right out of the gate. For now I say keep your bat on your shoulder and hope the stock gives you a giant pullback because at these levels it's too rich for me.

Meanwhile, General Intelligence's Andrew Piccinelli was one of many declaring, quote, the Cerebrus IPO may be the top. CNBC reported that there were forty five buyers for every seller of the stock, with Packy McCormick tongue in cheek arguing that the same is true for Cerebrus' product, posting Once again, you people don't understand infinites. If inference demand is infinite, Cerebris at$400 is ridiculously cheap. Infinity times their inference market share equals infinity.

Now, obviously, anytime you see this much intense demand for a single issue, it's worthy of some amount of skepticism. However, it sets up a pretty interesting dynamic for the mega IPOs coming down the pike. SpaceX is expected to finalize their paperwork next week, so they could go live by the end of the month. Then we have Anthropic and OpenAI rumored to be lining up their IPOs by the end of the year. Investor Kip Harriage suggested we shouldn't overthink it, writing.

If you're bearish on this market, just as we are entering the IPO boom phase, good luck to you. You're gonna need it. This is one of those moments where anytime you see anyone arguing about the fundamentals, it kind of feels divorced from the reality of the moment, which is that fundamentals don't matter if everyone is bidding AI and everyone right now is bidding AI.

For what it's worth, this is also why I think a lot of the discussion around the OpenAI versus anthropic IPO is just a little preposterous, like there isn't going to be absolutely infinite demand for both of those stocks. Couple more stories staying in markets.

Figma is the latest software company to come back from the dead on the back of strong AI revenue. Figma was one big victim of the SAS Pocalypse narrative, seeing their stock down as much as 50% this year. However, like Atlassian before them, the addition of AI features seems to have put them back on the right course.

During Thursday night's earnings, Figma reported that revenue grew at a forty six percent pace in the past quarter, accelerating from forty percent in the previous quarter. Figma credited their AI features with CFO Pravi Melwani stating, You can't dismiss the significance of new tools.

Figma is one of those companies where as the AI's gotten better, so is our pitch for customers. Now an interesting nugget given the themes that we've been exploring around the end of the AI subsidy era, in early March. Figma introduced a usage cap and started charging for token use above a limit. They said that the change hasn't made a dent in retention, noting that 75% of customers are still using their AI features, either sticking within the cap or paying for additional use.

Whatever the combination of reasons, the market now seems to believe in the SAS recovery, sending the stock up eight percent in after hours trading. Meanwhile, NVIDIA is very quietly having a major surge as the markets get mega bullish on AI.

The stock is up twenty percent over the past seven days, pushing the world's largest company close to a six trillion dollar valuation. Thursday's session added four point seven percent to an already hot run. Look, things could change fast, but right now the market is very much back on the AI hype train.

Next up, a bit of a weird one, OpenAI might be heading for another messy breakup as things reportedly get rocky with Apple. The information reports that OpenAI is considering legal action for breach of contract in relation to Apple's ChatGPT integration. Now you might remember that this all seemed a little weird right from the start. It was announced as part of Apple intelligence during WWDC 2024, but the whole thing had an air of half-commitment.

Sam Altman was present at the event, but he wasn't called on stage as part of the announcement. And overall the integration turned out to be a bit of an afterthought. The idea was that Apple Intelligence could kick complex requests from Siri over to ChatGPT, but OpenAI's technology was not being integrated as a core part of the product.

Now prior to the conference, people treated it like it would be Altman and OpenAI's coronation, enshrining them as a cornerstone of the Apple ecosystem. Instead, they kind of got the little brother treatment. Now that said, we know exactly how Apple intelligence has gone or not gone in the subsequent years, but apparently OpenAI is now reportedly considering suing Apple for failing to deliver on their side of the contract such as it was.

A source at OpenAI said that the company had been trying to improve their relationship with Apple over recent months, but there's been a lack of effort on Apple's part. That source added that OpenAI would prefer not to sue but wouldn't rule it out unless Apple begins showing more interest in collaborating with OpenAI. Meanwhile, OpenAI did not take part in last year's bake-off to determine who would win the contract to power the new version of Siri, which was ultimately won by Google.

The information is also reporting that Apple is now largely using Claude internally for coding and business work. and is reportedly testing native integrations of Claude and Gemini for iPhone, giving them the same level of system access as ChatGPT. Right now, this all feels like extremely thin sourcing to me, but it's worth noting, especially as we're heading into a week where there's going to be a lot of discourse thanks to Google I.O. around the state of the AI race across all these labs.

Meanwhile, heading into whatever the next phase of that race is, Anthropic appears to be trying to set a price floor for their IPO, with the Financial Times reporting that a new Anthropic round is all but a done deal. They state that terms have been agreed to and Anthropic will be raising$30 billion at a valuation of$900 billion, inching them ahead of OpenAI's last valuation.

A number of traditional venture firms like Sequoia and Altimeter are said to be co-leading the round, with each likely to invest two billion or more, and it appears that Anthropic has no shortage of investors willing to take the rest of the allocation.

Once it closes, this will be not only one of the largest venture rounds in history, but also one of the largest jumps in valuation ever at this scale, close to tripling up from the three hundred and eighty billion dollar valuation during their Series G round in February.

Now, while investors continue to be extremely enthusiastic about Anthropic, at least one big company is heading in a different direction. Microsoft has begun canceling cloud code licenses, shifting their developers across to the GitHub Copilot CLI instead. Microsoft first gave their developers access to Cloud Code in December, a subtle acknowledgement that their own in-house tools were falling behind.

Sources told The Verge that Anthropics tools were extremely popular, maybe a little too popular. Now they noted that alongside promotion of in-house tools, there's also a financial factor. The licenses will be terminated at the end of June just in time for the beginning of Microsoft's new financial year, with management reportedly seeing Claude Code as an easy place to cut some costs.

Look, I don't think this is that insane, especially given that Microsoft has a competing product that they really need to be up to snuff. I think you could argue this both ways, use one of the best tools at the moment to actually help make yours better, or take away that tool to create more internal incentive to improve what you've got. But in either case, it certainly feels like another part of the competitive strategies of these companies firming up.

Finally today, one of the big sub-themes running through the AI industry ever since the announcement of Claude's mythos has been security issues with everyone trying to figure out just how real the cybersecurity implications are. Reportedly, security researchers have now used Claude Mythos to find a new way to exploit Apple's operating system. The researchers claim that Mythos was able to link together a pair of bugs to execute an attack that granted access to kernel memory on Mac OS.

Mythos was used to both discover the vulnerability and carry out the attack, which caught a lot of people's attention as macOS is generally considered one of the more security hardened systems available. Researchers seemed blown away by the capabilities, writing. Mythos preview is powerful. Once it has learned how to attack a class of problems, it generalizes to nearly any problem in that class. Mythos discovered the bugs quickly because they belong to known bug classes.

This is the latest in a series of reports that suggest that Mythos was not just marketing hype. Last week Mozilla announced that Mythos had helped them find and patch 423 bugs in the past month, which was more than they had found in the previous 15 months combined, and Anthropic has also released an updated checkpoint for Mythos, which massively boosts its cybersecurity capabilities.

The UK AI Security Institute tested the new update and found that it could complete their automated cyberattack benchmark in six out of ten attempts. The previous benchmark run had two successes out of ten, with GPT-5.5 having one successful run. TLDR, it seems that indeed there are some pretty big capability jumps on the horizon. For now that is gonna do it for the headlines. Next up, the main episode.

And the good news, these behaviors are teachable at scale. If you're trying to move from AI access to real capability, KPMG's research on sophisticated AI collaboration is worth your time. Learn more at kpmg.com slash US slash sophisticated. That's kpmg.com slash US slash sophisticated. Today's episode is brought to you by Granola. Granola is the AI notepad for people in back-to-back meetings.

You've probably heard people raving about granola. It's just one of those products that people love to talk about. I myself have been using granola for well over a year now, and honestly, it's one of the tools that changed the way I work. Granola takes meeting notes for you without any intrusive bots joining your call.

During or after the call, you can chat with your notes, ask Granola to pull out action items, help you negotiate, write a follow-up email, or even coach you using recipes which are pre-made prompts. Once you try it on our first meeting, it's hard to go without. Head to granola.ai slash AI daily and use code AIDALY. New users get 100% off for the first three months. Again, that's granola.ai slash AI daily.

One thing I keep seeing in enterprise AI: companies hedging across every cloud, every model, every framework, or paying a GSI for a pilot that never ends. The team's actually shipping, they've picked a lane and they move fast. That's one of the reasons I like today's sponsor Robots and Pencils.

They've gone all in on AWS. They're an advanced tier and AWS pattern partner, and they ship production AI coworkers in 45 days. That's led to them doing some of the more interesting work I've seen on AI coworkers. And by that I'm not talking about chatbots. I'm talking about actual agentic systems that sit inside a business architecture and do real work.

That kind of focus matters if you're an enterprise leader trying to get something real into production, or an AWS rep trying to move a customer from interested to deployed. Request an AI briefing at robotsandpencils.com. One conversation with robots and pencils, and you'll know. So, coding agents are basically solved at this point. They're incredible at writing code. But here's the thing nobody talks about.

Coding is maybe a quarter of an engineer's actual day. The rest is stand-ups, stakeholder updates, meeting prep, chasing context across six different tools. And it's not just engineers. Sales spends more time assembling proposals than selling. Finance is manually chasing subscription requests. Marketing finds out what shipped two weeks after it merged.

Zencoder just launched Zenflow Work. It takes their orchestration engine, the same one already powering coding agents, and connects it to your daily tools. Jira, Gmail, Google Docs, Linear, Calendar, Notion. It runs goal-driven workflows that actually finish. Your stand-up brief is written before you sit down. Review cycle coming up, it pulls six months of tickets and writes the prep talk.

Now you might be thinking, didn't OpenClaw try to do this? It did, but it has come with a whole host of security and functional issues, which can take a huge amount of time to resolve. Zencoder took a different approach. SOC2 Type 2 certified, curated integrations, tighter security perimeter. Enterprise grade from day one, model agnostic and works from Slack or Telegram. Try it at zenflow.free.

We're actually going to start with some news from OpenAI. It's no secret that at this point, Codec and coding adjacent knowledge work use cases are where OpenAI's big focus is. They've gone from just a couple hundred thousand to more than four million codex users per week. And those codex users, as we've been discussing for the past couple of weeks, kind of represent something categorically different because they are building with agents. They don't represent just individual seats.

They represent a huge amount of token spend, either on their own or as part of their organizations, meaning that that 4 million are going to be punching way above their weight class, even relative to the 800 or 900 other million consumers who are using ChatGPT on a regular basis. Alongside a lot of anthropics communications troubles.

OpenAI has really taken advantage of the To put Codex for the first time in a really leading position relative to the other harnesses out there, specifically Claude Code, which you see not only in the conversation among developers and AI enthusiasts on Twitter, but also in more direct signals, like the fact that the slash goal primitive

came to Codex before it came to Claude Code. Now real ones also know that it was in Hermes first, but the point is that Codex is clearly where a ton of OpenAI's emphasis is and is making big gains in the industry. With that in mind, earlier this week, Codex team member Thiba from OpenAI said that the company was beginning to think about having a stable release cadence with a larger release each week on Thursday. We're going to announce things whenever they come up.

But the flip side, of course, is that it means that we get predictably something new each week. He did confirm later on that this would be their approach going forward, and this was the first Thursday where that came to bear. What was announced was codex in the ChatGPT mobile app.

Now, there's been a lot of memeing recently around people carrying half-open laptops everywhere they go. Basically, when you're running Codex or Claude code locally, the computer has to stay on for you to continue working. So you've got all these people who are shoving their thumbs in there as they go about other parts of their business.

Now Enthropic had released some features trying to make it easier to interface between Cloud Code locally in your mobile app, specifically in the remote control feature, but Codecs in the ChatGPT mobile app takes it a step farther.

This is not just remote control, but a full-fledged experience where you can genuinely work completely, including initiating new work, reviewing outputs, steering the execution, approving next steps, all from within this app. Now the discussions around this fell into one and three categories. The first was OMG, thank you. I've wanted this.

The second was, boy, that feels a little bit buggy. Are you sure you want to push things out every week? But the third and more significant is the one that recognizes that this is not just a feature update, but a continuation of the change in modality of how we do work.

Zord calls this a much bigger shift than people realize, writing, this is the beginning of AI agents becoming persistent operators, not just chat interfaces. You start tasks on your phone, Codex keeps executing on your Mac mini, laptop, or dev box. And you step in only when approvals or direction are needed. That changes the entire workflow dynamic for builders, researchers, and developers. We're moving from AI Helps Me Code to AI works alongside me continuously.

So much of what we have been talking about for the last couple of weeks, but really all year, is exactly this shift. from doing and producing things to managing fleets of synthetic intelligences that do those things for you. And it makes sense that now from an interface perspective, if you are indeed managing fleets of digital intelligences, That the labs that are creating those intelligences are trying to free you up from being chained to your traditional laptop type of environment.

OpenAI's Aiden McLaughlin writes, Codecs running while you cook with your partner, codecs running while you push your kid on the swing, codecs running while you call your mom. Codec running on the thing you dreamed about for years but never had time for, codecs doing this while you hang with your loved ones. Meanwhile, Nick Baumann from OpenAI gave a specific example of how this shifts had changed the way that he worked. Nick wrote

My laptop has become a satellite device since I started using Codex for my phone, and my Mac Mini has become the home. It's clunky, but the end state feels more like how we're going to be working in the near future.

I'm currently running the Codex app on two devices, my MacBook and my Mac Mini. My laptop isn't reliably connected to Wi-Fi enough, so I keep a Mac Mini on my desk that is always connected. When I kick off new threads from my phone, I start them on the Mac Mini. When I'm working from my desk, I run them there too. The cool part is that I've added my MacBook and Mac Mini as connected devices to each other. That means I can start or resume threads from either device.

So if I'm in a meeting but want to continue a thread on my laptop that was started on my Mac Mini, I can do that. What this means? I have an always-on codex that is accessible from my phone with its own dev environment. All threads are always accessible from any of the three devices. I can run heartbeat threads that stay on 24-7.

It's a little makeshift today, but the shape of it feels very real to me. Codex is no longer tied to whichever computer happens to be open in front of me. It starts to feel like something I can stay connected to across whatever device I'm using. Early open claw hackers are going to see a lot of the motivation and goals they had for setting up those types of environments on display in how Nick is now using just the straight up normal offering from OpenAI.

Lapo Cherisi expands out the implications even farther. OpenAI, he writes, is launching Codex Mobile, so you can monitor and approve your AI coding agent from your phone. This is the real tell. When AI agents work unsupervised, the bottleneck isn't generation speed, it's human review cycle.

The mobile interface isn't a convenience feature. It's an admission that we're entering a world where your job is triage, not execution. And the SLA is now how fast can you approve the next step while you're in a meeting? B2B will follow the same path.

AI SDRs, AI content engines, AI campaign builders, all optimized for async human checkpoints, not replacing the human entirely. The UX question isn't can the AI do it anymore? It's how do we design approval flows that don't become the new bottleneck? Most Martec vendors haven't even started asking that question yet.

Now, here's where it starts to get really interesting, especially in light of the Google conversation. On the one hand, this feels a little bit like ChatGPT becoming the super app that OpenAI has been talking about for some time. Adam GPT from OpenAI writes

Codex in ChatGPT definitely feels super app-ish to me, to be clear directionally towards a super app and not that it is the super app. But there is another possible route, which is not that Codex fits inside ChatGPT, but that Codex kind of becomes ChatGPT. And what you start to see is this weird tension. It feels much more like it did in 2025 and 2024, that work AI and non-work AI are diverging.

There was an essay last year that became fairly influential, especially when people were in the scaling bubble kind of debate moment, called AI as Normal Technology. And what it argued was not that AI wasn't significant, but that while people were acting like its disruption was going to be faster, broader, and more extreme than previous technologies,

The essay's authors, Arvind Narayanyan and Syesh Kapoor, actually said no, AI was going to be pretty normal in terms of its pattern of diffusion, the inertia with which it hit human systems. And this year, I think it wouldn't be unreasonable to argue. that as a consumer technology, AI is, while extremely impressive and certainly faster growing than things we've seen.

Still ultimately normal. In fact, one of the reasons for AI pushback is regular consumers who aren't using this for work having AI features shoved down their throats in places that they don't really want them. Meanwhile, over on the work side of the house, We simply cannot get enough. We cannot get updates to the models and harnesses fast enough. We cannot get unlocked access to more tokens fast enough. I think when it comes to work in other words.

AI is an extremely abnormal technology, which is not to say that it won't deal with normal problems like institutional inertia. There's a reason the big labs are spending so much money building out forward-deployed engineering organizations. But the disruption to the way that we work.

truly doesn't look like anything that came before it. If I am correct, In this assertion that I keep making, that big chunks of knowledge work are moving from doing the thing to managing AI agents that do the thing for us, that is a category shift in how we work and what we do, not just a change in how we accomplish the same old goals. Which puts, I think, companies trying to build for both consumers and the work user in kind of a tricky situation. Over the last six months.

OpenAI made a very clear decision, best expressed in their shutdown of Sora, that although they certainly weren't going to abandon their hundreds of millions of consumer users, the big game for them and where they needed to place all of their emphasis was on that work user.

Anthropic, meanwhile, never really had the benefit of going after both and had always been on the work user train. Microsoft, by default, given where it sits in the ecosystem, was always going to be work-related AI, and on the other side of the ledger, both Apple and Meta were obviously always going to be on the consumer side.

Because it's not exactly clear how much the normalness of AI as a consumer technology has to do with AI itself or just the woeful underperformance of those two companies in finding a way for it to be useful for consumers. All of this leaves Google. Google is the one other company besides OpenAI that for some time

has pursued the consumer AI and the work AI uses in equal measure. Now Google is even more voracious than that, working on new categories of models and world models, deep multimodal and video type of things, and more. In fact, one of the challenges for people using Google's tools is product sprawl, which brings us to next week's I.O. What are Google planning to announce? And what is it going to say about how they're choosing to handle the emerging difference between consumer and work AI?

First glances suggest that unlike OpenAI, they are not going to pick a lane but continue to pursue both. One of the reports is for a new, always on, personalish AI agent called Spark. This screenshot of a welcome screen for Gemini Spark has been flying around, describing itself by saying, Let Gemini do more as your everyday AI agent, ready 24-7 to help with your inbox, online tasks, and more.

The big promise is similar to what we've talked about in the context of Google and Apple before, that Google already has a whole bunch of contextual knowledge about you and can theoretically be using that context to build better experiences for you. And indeed, on that same launch screen it says, the more you use Gemini Spark, the better it understands you and what you want to accomplish. To work on your tasks, it uses your info from sources like connected apps,

skills, chats, tasks, websites you're logged into, personal intelligence location, and more. Now it doesn't say that Gemini Spark is strictly personal, but to me it feels like the language leans in that direction. In another part of that same screenshot it says,

While it is designed to ask for your permission before taking sensitive actions, it may do things like share your info or make purchases without asking. And at least at first glance, people seem at least a little bit enthusiastic. Andrew Curran writes,

I got a new permissions request pop up a few days ago, so they're getting everything ready ahead of the launch. Spark is a great name and Gemini will be very good at this. All its best qualities are made for this. Google I.O. is in five days, the agent wars are about to begin. Brasser X rights

If this Gemini Spark leak is real, it actually looks promising. A twenty four seven agent that learns from you, runs across apps, chats, tasks, and logged in websites is exactly where consumer AI has to go. Google I.O. could be spicy. Jan Kronberg writes. The winning assistant won't be the smartest empty chatbot, but the one with the deepest context about your actual life. Google has had that data for 20 years, and Spark is finally the product built on top of it.

However, I can't help but feel a little bit like Peter Gostev from Arena AI, when he tweets a quote from the posting, The more you use Gemini Spark, the better it understands you and what you want to accomplish, and adds, It's funny. I feel like I've seen that line from Google for about eight years with product name changed once in a while. Hope it will actually work this time.

Now, I don't know about eight years, but it certainly feels like this has been the promise of personal agents since we started talking about them. The question is, what the personal use cases for agents actually want to be. To the extent that we're talking about work agents, Jan Kronberg's assessment that the winner won't be the smartest empty chatbot, but the one with the deepest context about your actual life.

I don't know if I agree with There's a reason that context bleed can be one of the biggest challenges with agents. Because I use AI and agents to talk about so many different ideas, many of which I get to at least second base before abandoning to move on to something else. When Claude or Codex have all of the context about me, I spend a lot of time telling them to just ignore or remove some past things from their memory because they're no longer relevant.

I can't even tell you how confused it gets about my entrepreneurial or builder plans because it's always comparing them to whatever I was talking about a few weeks or a few months ago. Now certainly I think Work AI wants better tools to curate context. and make it available easily, but it's not just gonna be this big old pot of everything about you.

On some intuitive level though, it makes more sense that this would matter for consumer agent type use cases than People probably aren't going to want to take a bunch of time to be perfectly curatorial in what an agent has access to.

What I'm not sure yet is what the uses for those agents are actually going to be. I find myself feeling like a curmudgeonly old man when I look at people's attempt at shopping agents or travel booking agents and just get really skeptical that that's something that people are going to want agents for.

But I certainly don't have enough confidence in that to argue that people shouldn't try. And when you have all of that data and all of that context about people and billions of users for whom your tools are already integrated deeply into their personal lives.

it makes sense that that would be something that Google would be out on the front trying to figure out. On the other side though, it also does seem like Google is going to try to jump back more aggressively into work AI as well.

Disappointingly to some, early reports are that while we will get new Gemini models, they will not be state-of-the-art. Anything that we got, it sounds like would be closer to somewhere between 4-7 and 5-5, certainly as opposed to anything like Mythos, and even that might be pushing it.

Where the work AI side of the conversation gets interesting, however, is in the context of this broader conversation we're having about the end of AI's experimentation era, the end of the subsidy era, and the beginning of the era of actual factual trade-offs. AI entrepreneur Bindu Reddy writes. Gemini 3.2 Flash. Rumors are that benchmarks show it's hitting 92% of GPT-55's performance on coding and reasoning tasks while being 15 to 20x cheaper on inference costs.

The latency improvements are insane, sub 200 milliseconds for most queries. Google's distillation and sparsity techniques are paying off massively. They've essentially compressed a frontier model into a flash variant without the usual quality cliff. This strikes me as a path where Google could easily become incredibly relevant for all these coding and coding adjacent work-related use cases.

There are companies all around America right now trying to decide if they can get over their concerns to start running a Chinese open source model. If Google can swoop in with a 20x cheaper inference. that's at Opus 4.5 or 4.6 type of levels, a lot of those companies will breathe a cool sweep of relief and redesign their systems to use that model for a lot of their big workloads.

Even if Google got there accidentally, if I were them, I would be thinking about how to triple down on this as a real significant opportunity. But even if that's the case, there is still a big question around the harness. This week Ethan Moloch posted, really curious when Gemini is going to join the co-work and codex race to build a local app that isn't just for developers. Antigravity hasn't posted updates to X in a month, and remains very software focused.

Meanwhile, we see accelerated updates and releases from OpenAI and Anthropics. Hader writes, OpenAI is putting its energy into codec. XAI just launched Grok Build for coding. Anthropic already has Claude Code, but Google still looks a bit unclear on its main agentic coding platform, Gemini CLI, AI Studio, Jules. Really hoping the upcoming Google IO brings some clarity.

That I think would be another huge win if there were clarity and more than just clarity, consolidation around what the core agentic harness was going to be for the Google ecosystem.

So, if at the end of next week after I.O., we're sitting there having announced Spark for consumers, an Opus four five or four six class model for fifteen to twenty times less money, and a clear consolidation on which agentic harness that you're supposed to be working through, and hopefully some updates to that harness, how would I think people will have received that? I wouldn't be surprised if the market doesn't really know what to make of it.

Unfortunately for them, I think most Wall Street investors are not listening to this show, staying close to conversations like harness engineering and the subsidy era ending. And I think there might be a divergence between what they think is good, i.e., you gotta think a lot of them are expecting to hear that Google has a state-of-the-art model, at least at the level of GPT-5.5 or Opus 4.7. But I think when it comes to actual builders, there is a clear lane for Google to get back in their game.

Now, of course, it's also possible that Google just decides that they don't care that much about that type of usage, but I'm skeptical that that's how it will go. In any case, it is going to be an interesting week. Give some rest this weekend and get geared up. For now, that's gonna do it for today's AI Daily Brief. Appreciate you listening or watching as always. Until next time, peace.