May Cognition eat Google’s catered lunch
And: Waymo is soaring while everyone else is still trying to find the accelerator.
Welcome to Cautious Optimism, a newsletter on tech, business, and power.
Friday. Intel’s earnings were a bit of a horror show. Revenue came in ahead of expectations ($12.86 billion versus $11.92 billion), but the company’s third-quarter profit estimates were light. The company is curtailing fabs in Germany and Poland, and merging some of its assembly efforts in Vietnam and Malaysia as part of an effort to “grow [its] capacity based solely on the volume commitments and deploy CapEx lockstep with the tangible milestones and not before.”
But most worrisome was admission by the company that it’s as behind in AI as we thought. Recently installed CEO Lip Bu Tan said that his company must “move up the abstraction stack into system and software,” areas “where Intel has traditionally been weak or entirely absent.” How to do that? “First understanding emerging and real AI workloads,” then by working “backwards to design software, systems, and silicon to enable best outcome for those particular workloads.”
Admirable and probably the right move. But we might have expected Intel to be a bit further down that particular road. — Alex
📈 Trending Up: Combatting corruption in Ukraine … rents, when rates rise … fintech on the S&P500 … open-source AI … no shit … tariff impacts on automotive profits … India-China relations? … shelling between Cambodia, Thailand …
📉 Trending Down: Combatting corruption in Texas … food security in Gaza … comprehensible finance products … air safety in Russia … day 1,247 … Vegas … domestic monetary policy … Chinese kindergarten employment …
Two cheers for founders, and welcoming AI to the culture wars
The demise of the OpenAI-Windsurf deal allowed to Google to strip-mine the successful startup of its founders and many of its core engineers. The transaction left behind the husk of a company with capital and ARR, but a void where its ability to innovate had previously been found.
As Google competes with Windsurf — it has a host of tools on offer to help you write code — perhaps the search giant felt that it was being given the chance to eat twice while paying once. Worse, Windsurf staff that were offered a spot at Google as part of the deal didn’t get that square an offer. Here’s Prem Qu Nair, who writes that despite being the second employee at Windsurf, and being offered a place at Deepmind:
I was given an offer that would explode same day. I had to forfeit all of my vested shares earned over my 3.5+ years at Windsurf. I was ultimately given a payout of only 1% of what my shares would have been worth at the time of the deal.
In going to Cognition, I’ve chosen a different direction.
A 24 hour countdown to explode your vested shares for a pittance is barely an offer. It’s a threat: Take this, or watch as the asset you helped build dies after we yoinked its brain.
GPT-5 is pipped for an August release, The Verge reports.
Thus, two cheers for the founders at Cognition who have scooped up what was left of Windsurf and tucked it into their own operations. Cognition, the startup behind the Devin coding agent, had just the sort of talent that Windsurf had just lost, and the combined entity would have two startups’ worth of ARR to raise against. And raise it is: Forbes reports that the newly-expanded startup is working on a round that could value it at $10 billion, up from $4 billion this March. Good. May Cognition eat Google’s catered lunch.
Wait, what about woke AI? Back in mid-July, the WSJ reported that the Trump administration was working on an executive order targeting AI models that it considered ‘woke.’ Given that defining the term is akin to trying to sheetrock with overcooked linguini, the story was worrisome.
Well, we got our hands on the EO — here — and it’s interesting. To wit, the document admits that while “the Federal Government should be hesitant to regulate the functionality of AI models in the private marketplace,” it intends to enforce notable restrictions on AI technology that it itself purchases:
One of the most pervasive and destructive of these ideologies is so-called “diversity, equity, and inclusion” (DEI). In the AI context, DEI includes the suppression or distortion of factual information about race or sex; manipulation of racial or sexual representation in model outputs; incorporation of concepts like critical race theory, transgenderism, unconscious bias, intersectionality, and systemic racism; and discrimination on the basis of race or sex. DEI displaces the commitment to truth in favor of preferred outcomes and, as recent history illustrates, poses an existential threat to reliable AI.
I don’t want to sound alarmist, but it sounds that if an AI model understands that trans people are, in fact, people, it will be labeled ‘woke’ and thus find itself precluded from government use. Or if it discusses systemic racism? I don’t know how else to read those words.
The good news is that most AI will remain outside of government censorship. The bad news it that AI companies that want to sell to the government will have to prepare a second, more censored family of AI models to do so. Yuck.
If it drives like a duck, turns like a duck, honks like a duck…
Here are three headlines for you:
Waymo hits 100 million driverless miles as robotaxi rollout accelerates [July 15, 2025]
Uber will let women drivers and riders request to avoid being paired with men starting next month [July 23, 2025]
Lyft Now Lets You Favorite and Block Drivers. Here's How It Works [July 24, 2025]
My contention here is that self-driving cars are popular because they are technological curiosities, because they are safer drivers, and because a lot of people don’t want to deal with other people. If you are a man, ask the women in your life about the worst Uber ride that they have taken. They will have stories. Despite Uber and its rivals building stronger driver checks over time, you can’t screen for creepy in clean criminal records.
So, self-driving cars have a huge wedge into the traditional Uber market: Removal of the driver, and thus the chance of someone learning where you live that you might not want to.
Uber is building out its own self-driving robotaxi fleet, as we’ve reported. But it feels a bit late to save all the company’s future market share. Waymo is soaring, and everyone else is still trying to find the accelerator.
Everyone wants to build you stuff
While we scratch our heads at the Windsurf saga, the business of getting AI to build stuff for customers is heating up. The category — often referred to by the catch-all moniker of vibe coding — has niches. There are products designed to help users code, and there are services that are designed to do the coding work for the user.
Both models are doing well:
Cursor reached $500 million ARR earlier this year, underscoring how much the market will pay for coding tools that, in theory, allow for greater productivity from expensive humans.
Lovable reached $100 million ARR just the other day, detailing how much the market will pay for tools that can turn an idea into a product, with all the messy code writing abstracted to the backend.
Cognition, Figma’s new Make product, TabNine, GitHub Copilot, Claude Code, Mistral’s code-writing AI models, and the coming Grok 4 model built for coding are just some of the other players in both camps.
And the retinue is only growing:
Amazon recently announced Kiro, an “AI IDE that helps you deliver from concept to production through a simplified developer experience for working with AI agents.”
Microsoft recently introduced GitHub Spark, a service that it claims will help users “transform [their] ideas into full-stack intelligent apps and publish with a single click.”
Sound familiar? The good news for us regulars is that the number of well-heeled companies working to build tools to help us build tools is expanding so quickly that we should see healthy feature and price competition. For startups seeing giants crowd into a space that was previously largely the domain of upstarts is worrying; but there’s so much spend up for grabs I suspect that we’ll wind up with a fleet of winners instead of, say, a new Google search-style incumbent.