The Search Wars Begin

Alex Shannon: I keep going back and forth on this — I think I actually land on the side that this is a good thing.

Sam Hinton: Really? Because I read the same story and I came out the other end deeply uncomfortable.

Alex Shannon: But think about it — Google’s finally being honest about what Search has become. They’re not pretending it’s just ten blue links anymore.

Sam Hinton: Alex, thirty percent of people immediately went looking for an alternative. That’s not users embracing the future, that’s users running for the exits.

Alex Shannon: Or maybe it’s just the vocal minority? I mean, how many people actually know what DuckDuckGo is?

Sam Hinton: Well, apparently a lot more than Google thought.

Alex Shannon: You’re listening to Build By AI, the daily show tracking how artificial intelligence is reshaping our world. I’m Alex Shannon.

Sam Hinton: And I’m Sam Hinton. Today we’re diving into Google’s big bet gone wrong, a massive security hole threatening millions of AI agents, and why the Vatican is suddenly very concerned about robot soldiers.

Alex Shannon: Plus OpenRouter just became the latest AI unicorn, proving that maybe the multi-model future is already here.

Sam Hinton: It’s Tuesday, May 27th, 2026. Let’s get into it.

DuckDuckGo installs are up 30% as users reject being ‘force-fed’ Google’s AI Search

Alex Shannon: Alright, so let’s start with what we were just arguing about. Google made a massive announcement at I/O 2026 — they’re completely overhauling Search. We’re talking about replacing those traditional blue links with AI agents. This is Google basically saying ‘we’re going full AI, whether you like it or not.’

Sam Hinton: And users said ‘actually, we don’t like it.’ The backlash was immediate and measurable. DuckDuckGo app installs spiked thirty percent. Thirty percent, Alex! That’s not a blip, that’s a migration.

Alex Shannon: OK, but let’s be real here. How long have we been complaining that Google Search results are getting worse? All the SEO spam, the content farms, the ads disguised as results. Maybe AI agents cutting through all that noise is actually what we need.

Sam Hinton: But that’s exactly the problem — they’re not giving us a choice. It’s like if Netflix suddenly said ‘we’re only showing you what our algorithm thinks you should watch, no more browsing.’ Sometimes I want to see the ten blue links. Sometimes I want to make my own judgment about what source to trust.

Alex Shannon: That’s a fair point. But isn’t this just the natural evolution? I mean, when was the last time you went to the second page of Google results anyway?

Sam Hinton: Dude, there’s a difference between not going to page two and having an AI agent decide what information I should see. What happens when the agent gets it wrong? What happens when it has bias baked in? At least with blue links, I could see multiple perspectives.

Alex Shannon: Right, so the question becomes — is Google moving too fast here? Should they have kept both options available?

Sam Hinton: Exactly! And look, I’m not anti-AI. But this feels like Google saying ‘trust us, we know better than you do what you’re looking for.’ The thirty percent jump in DuckDuckGo usage tells me people aren’t ready for that level of mediation.

Alex Shannon: But here’s what I keep thinking about — maybe this is actually good for information quality? If the AI agent is pulling from multiple sources and synthesizing them, it could potentially give us better, more comprehensive answers than we’d get from clicking through a bunch of individual links.

Sam Hinton: That’s the optimistic case, and maybe you’re right. But what if the AI agent misinterprets something? What if it prioritizes certain types of sources over others? What if Google’s business incentives start influencing what the AI considers ‘good’ information?

Alex Shannon: Those are valid concerns. I guess my question is whether those risks are greater than the current problems with search. Because let’s be honest, most people don’t fact-check or cross-reference sources anyway. They just click the first result that looks credible.

Sam Hinton: True, but at least they had the option to. That’s what bugs me about this whole thing. It’s not that AI agents are inherently bad — it’s that Google made a unilateral decision to change how billions of people access information, without really asking if that’s what we wanted.

Alex Shannon: The interesting thing is this could actually be a huge opportunity for competitors. DuckDuckGo has been the privacy-focused alternative for years, but now they might become the ‘give me actual search results’ alternative.

Sam Hinton: Yeah, and think about what this means for businesses too. If Google’s AI agents are deciding which information to surface, that’s a massive shift in how companies think about SEO, about reaching customers. The whole digital marketing industry might need to completely reinvent itself.

Alex Shannon: That’s huge. Instead of optimizing for keywords and ranking algorithms, you’re optimizing for what an AI agent thinks is worth sharing. That’s a completely different game.

Sam Hinton: And we don’t even know the rules of that game yet. At least with traditional SEO, it was somewhat transparent. This feels like a black box where Google controls both the input and the output.

Alex Shannon: You know what’s really interesting though? This might force websites to actually create better, more authoritative content. If AI agents are looking for comprehensive, well-sourced information rather than just keyword-stuffed articles, that could improve content quality across the web.

Sam Hinton: Or it could kill smaller publishers who don’t have the resources to optimize for AI agents. If Google’s AI preferentially surfaces content from major media outlets or well-funded sites, we could see even more consolidation in online publishing.

Alex Shannon: That’s a really good point. There’s this whole ecosystem of independent creators and niche experts who’ve been able to reach audiences through search. If AI agents don’t surface that content, those voices could get drowned out.

Sam Hinton: Exactly. And here’s another thought — what happens to citations and traffic? If people are getting their answers directly from the AI agent, they might never click through to the original sources. That could be devastating for content creators who rely on search traffic.

Alex Shannon: Oh man, I hadn’t thought about that. If AI agents are essentially summarizing and synthesizing content without driving traffic back to the original creators, that’s basically taking their work product without giving them the economic benefit.

Sam Hinton: Right! It’s like having the world’s most sophisticated content aggregator that never actually sends readers to the original articles. Publishers could see their traffic crater even if their content is being used to generate AI responses.

Alex Shannon: This is making me rethink my position. Maybe the issue isn’t whether AI agents give better answers — maybe it’s about the broader economic and information ecosystem that search supports.

Sam Hinton: And that’s probably why we’re seeing such strong pushback. People might not be able to articulate all these concerns, but they can sense that something fundamental is changing about how information flows on the internet.

Alex Shannon: Alright, so here’s what I’m watching — does that thirty percent spike in DuckDuckGo usage hold, or do people drift back to Google out of habit? And does Google respond by maybe offering both options, or do they double down?

Sam Hinton: My prediction? Google doubles down because they have to. They’re competing with ChatGPT and Claude for being the go-to AI assistant. But I think they’re about to learn that search behavior is stickier and more personal than they realized.

Alex Shannon: And if DuckDuckGo can capitalize on this moment — improve their results, maybe add some AI features that enhance rather than replace traditional search — they could actually grab meaningful market share for the first time in years.

Sam Hinton: The irony is that Google’s AI-first approach might actually validate the market for AI-assisted search, just not necessarily Google’s version of it. Other companies could build AI tools that help users navigate traditional search results rather than replacing them entirely.

OpenRouter more than doubles valuation to $1.3B in a year

Alex Shannon: Let’s shift gears to some good news in the AI space. Early reports suggest that OpenRouter just closed a massive Series B — we’re talking $113 million led by CapitalG, Google’s growth fund. If confirmed, this more than doubles their valuation to $1.3 billion in just one year.

Sam Hinton: OK, that’s impressive, but what really caught my attention was the usage numbers. Five times growth over six months. That’s not just hype money, that’s real adoption. And I think it tells us something important about where this industry is heading.

Alex Shannon: Remind me what OpenRouter does exactly, because I think a lot of people might not be familiar with them.

Sam Hinton: So OpenRouter is basically the Switzerland of AI models. Instead of being locked into just GPT or just Claude, developers can access multiple AI models through one API. Think of it like Stripe, but for AI instead of payments. You integrate once, and then you can route requests to whatever model works best for that specific task.

Alex Shannon: And that five-times usage growth suggests that developers are really embracing this multi-model approach. They’re not just picking one AI and sticking with it.

Sam Hinton: Exactly! And this makes total sense when you think about it. Different models are good at different things. Maybe you use Claude for writing, GPT for coding, and some specialized model for data analysis. OpenRouter lets you do all of that without building separate integrations for each one.

Alex Shannon: But here’s what I find interesting — CapitalG leading this round. That’s Google’s investment arm betting big on a company that basically commoditizes AI models, including Google’s own Gemini. That seems… counterintuitive?

Sam Hinton: Actually, I think it’s brilliant strategy. Google realizes they’re not going to win by having the single best model for everything. But if OpenRouter becomes the standard way developers access AI, and Gemini is one of the options, Google still wins. Plus, they get data about how their models perform compared to competitors.

Alex Shannon: That’s actually really smart. Instead of fighting to be the only AI model people use, they’re betting on being part of the ecosystem that emerges when people use multiple models.

Sam Hinton: Right, and think about what this means for the average developer or business. Instead of vendor lock-in, you get vendor flexibility. If OpenAI has an outage, you can route to Anthropic. If one model gets too expensive, you can switch to another. That’s powerful.

Alex Shannon: It also puts pressure on the model companies to keep innovating and keep prices competitive, because switching costs are now much lower.

Sam Hinton: Absolutely. And I think this funding round validates something we’ve been talking about — the future isn’t going to be dominated by one AI model. It’s going to be about orchestrating multiple AI models intelligently. OpenRouter is building the infrastructure for that future.

Alex Shannon: But let me push back on this for a second. Doesn’t this create its own problems? If you’re routing between different models for different tasks, how do you maintain consistency? How do you debug issues when you don’t know which model generated which response?

Sam Hinton: Those are valid concerns, and I think that’s actually where OpenRouter’s value proposition gets interesting. They’re not just providing access to different models — they’re presumably building tools for model selection, performance monitoring, maybe even automatic fallbacks when one model is having issues.

Alex Shannon: So they become the orchestration layer. That actually makes the $1.3 billion valuation make more sense. They’re not just an API gateway, they’re potentially the mission-critical infrastructure that sits between applications and AI models.

Sam Hinton: Exactly. And think about the timing here. A year ago, most developers were just trying to figure out how to integrate any AI model into their applications. Now they’re sophisticated enough to want choice, performance optimization, cost management — all the enterprise concerns.

Alex Shannon: The timing feels right too. We’re past the early hype phase where everyone was just trying to integrate any AI. Now developers are getting sophisticated about choosing the right tool for the right job.

Sam Hinton: And the five-times usage growth tells me that this isn’t just a few big enterprise customers. This is probably thousands of developers and companies realizing that multi-model is the way to go. That’s the kind of organic growth that makes investors excited.

Alex Shannon: You know what else this suggests? The AI model market is maturing faster than I expected. We’re already at the point where standardization and interoperability matter more than just raw capability.

Sam Hinton: Yeah, and if OpenRouter can execute on this vision, they’re positioning themselves to be absolutely essential infrastructure. Like, imagine trying to build modern software without AWS or without Stripe. OpenRouter wants to be that indispensable for AI.

Alex Shannon: The question is whether the big model companies try to compete with this directly. Could OpenAI or Anthropic build their own multi-model routing? Or do they recognize that OpenRouter actually helps them by making it easier for developers to adopt AI in general?

Sam Hinton: I think the big players probably see OpenRouter as net positive. It lowers the barrier to AI adoption, which grows the overall market. And competition between models through a platform like OpenRouter probably drives more innovation than if everyone was siloed.

Alex Shannon: Plus, if Google is leading this round through CapitalG, that suggests they see this as strategic, not competitive. They want to make sure Gemini is well-positioned in whatever multi-model future emerges.

Sam Hinton: Keep an eye on this because if they’re right about the multi-model future, this could be one of those companies that becomes invisible but essential — the plumbing that powers everything else.

Millions of AI agents imperiled by critical vulnerability in open source package

Alex Shannon: Alright, now let’s talk about something that should make every developer break out in a cold sweat. Early reports suggest there’s a critical vulnerability in something called Starlette — it’s an open source framework that gets 325 million downloads per week. If confirmed, this could affect millions of AI agents and tools worldwide.

Sam Hinton: Oh no. Oh no no no. Three hundred and twenty-five million downloads per week, Alex. This is like finding out there’s a critical flaw in the concrete that half the buildings in the world are built on.

Alex Shannon: Walk me through what Starlette actually is, because I think a lot of people have never heard of it but apparently it’s everywhere.

Sam Hinton: So Starlette is basically the foundation that a lot of web applications are built on, especially AI applications. Think of it like the foundation of a house — you don’t see it, but everything else depends on it. And because it’s open source, thousands of other projects use it as a dependency.

Alex Shannon: And according to the reports, this vulnerability allows hackers to breach servers and steal sensitive data and credentials. When you’re talking about AI agents that might have access to company data, personal information, API keys…

Sam Hinton: Right, this is potentially catastrophic. AI agents often have elevated privileges because they need to access multiple systems to do their job. So if you compromise an AI agent, you’re not just getting one piece of data — you might be getting the keys to the kingdom.

Alex Shannon: And this highlights something we’ve been warning about for a while — the security implications of AI aren’t just about the models themselves. It’s about all the infrastructure and dependencies that power these systems.

Sam Hinton: Exactly. Everyone’s so focused on prompt injection and model safety, but here we have good old-fashioned software vulnerabilities that could affect millions of AI deployments. It’s like worrying about whether your car’s AI is too smart while ignoring that the brakes might fail.

Alex Shannon: What really worries me is the scale. Because this is open source and so widely used, patching this isn’t just a matter of updating one application. It’s potentially thousands of companies that need to identify if they’re using Starlette and then update their entire stack.

Sam Hinton: And some of them might not even know they’re using it! This is the classic dependency hell problem. Your AI application uses framework A, which depends on library B, which depends on Starlette. Three layers deep, and suddenly you’re vulnerable.

Alex Shannon: Think about all the AI startups that have moved fast and built their entire platforms on top of frameworks like this. They might have no idea what’s in their dependency tree, and now they’re potentially sitting on a massive security hole.

Sam Hinton: And here’s what’s really scary — if this affects millions of AI agents like the reports suggest, we could be looking at a coordinated attack surface. Hackers don’t have to find individual vulnerabilities in each AI system. They just have to exploit this one flaw across thousands of targets.

Alex Shannon: That’s terrifying. Especially because AI agents are often deployed in production environments where they have access to databases, APIs, internal systems. A mass exploitation could be devastating.

Sam Hinton: And think about the timing. We’re in this phase where everyone is racing to deploy AI agents, often without fully understanding the security implications. A vulnerability like this could set back AI adoption by years if it leads to major breaches.

Alex Shannon: This feels like a wake-up call for the industry. As AI becomes more critical infrastructure, we need to take supply chain security a lot more seriously.

Sam Hinton: Yeah, and it’s not just about patching vulnerabilities. It’s about having visibility into your entire dependency tree, having rollback plans, having monitoring in place to detect if something goes wrong. The basics that we should have been doing all along.

Alex Shannon: But let’s be honest — how many companies actually have that visibility? Especially the smaller ones that are building on top of frameworks and assuming they’re secure.

Sam Hinton: Probably very few. And this is where the move-fast-and-break-things mentality of a lot of AI companies could really come back to bite them. Breaking things is fine when it’s a feature, but not when it’s your entire security model.

Alex Shannon: What’s the responsible thing to do here if you’re running AI systems? Like, if you’re listening to this and you have AI agents in production, what should you be doing right now?

Sam Hinton: First, audit your dependencies. Figure out what open source packages you’re using, especially web frameworks. Second, have an incident response plan ready. And third, consider whether your AI agents really need all the permissions they currently have.

Alex Shannon: That last point is huge. The principle of least privilege applies to AI agents just like any other system. If your AI agent only needs to read certain data, don’t give it write access to everything.

Sam Hinton: If you’re running AI agents or tools in production, this is probably a good time to audit what open source dependencies you’re using and make sure you have a process for security updates.

Alex Shannon: And honestly, this might be the push that finally gets companies to invest properly in DevSecOps for AI. Because the alternative is waking up one morning to find out that your AI agents have been compromised for months.

Sam Hinton: The silver lining here is that this is a solvable problem. It’s not some fundamental flaw in AI technology — it’s a traditional software vulnerability that can be patched. But it requires the industry to take security seriously and not just treat it as an afterthought.

AI Agents Plunged the Tech World Into Chaos. Here’s Exactly How That Happened

Alex Shannon: Alright, so there’s this fascinating deep-dive from Wired about how something called Claude Code and OpenClaw apparently triggered what they’re calling ‘computing’s biggest transformation’ and caused chaos in the tech world. The details are still coming out, but this sounds like a major milestone in AI’s impact on the industry.

Sam Hinton: Wait, Claude Code and OpenClaw? I’ve heard whispers about these but I thought they were still experimental. If Wired is saying they caused chaos, that suggests they actually got deployed at scale.

Alex Shannon: That’s what caught my attention too. We’ve been talking about AI agents theoretically for so long, but this sounds like the moment they actually started reshaping how software gets built and deployed.

Sam Hinton: You know what this reminds me of? The early days of cloud computing when companies started realizing they could spin up servers instantly instead of waiting weeks for hardware. Except this time it’s about spinning up entire development workflows.

Alex Shannon: But ‘chaos’ is an interesting word choice. That suggests this transformation wasn’t smooth. What kind of chaos are we talking about here?

Sam Hinton: Well, think about it — if AI agents can suddenly write, test, and deploy code autonomously, that’s going to disrupt a lot of existing processes. Maybe we’re talking about version control systems getting overwhelmed, or code review processes breaking down, or just humans not being able to keep up with the pace of change.

Alex Shannon: Or maybe it’s more fundamental than that. If AI agents are making architectural decisions and writing core infrastructure code, how do you maintain oversight? How do you ensure quality? How do you even understand what your own systems are doing?

Sam Hinton: That’s terrifying and exciting at the same time. On one hand, the productivity gains could be massive. On the other hand, you could wake up with a codebase that no human fully understands.

Alex Shannon: And think about the security implications. If AI agents are writing code that other AI agents then deploy and maintain, we’re creating these complex systems where the humans are increasingly removed from the loop.

Sam Hinton: Right, and that ties back to the Starlette vulnerability we just talked about. Imagine trying to patch a critical security flaw when half your codebase was written by AI agents using patterns that might not be immediately obvious to human developers.

Alex Shannon: That’s a really good point. If Claude Code and OpenClaw are generating massive amounts of code automatically, how do you audit it all? How do you know what coding patterns they’re using, what dependencies they’re pulling in?

Sam Hinton: And here’s another thought — what happens to the human developers in this scenario? Are they becoming more like architects who design systems that AI agents then implement? Or are they getting squeezed out entirely from certain types of work?

Alex Shannon: I’m really curious to see the full Wired article because the phrase ‘computing’s biggest transformation’ is pretty bold. Bigger than the shift to cloud? Bigger than mobile? That’s saying something.

Sam Hinton: It might be, though. Those were platform shifts, but this could be a fundamental change in how software gets created. Instead of humans writing code, we might be moving to humans describing what they want and AI agents figuring out how to build it.

Alex Shannon: If that’s true, then every software company needs to start thinking about how their development processes, their hiring, their entire business model might need to change.

Sam Hinton: Yeah, and the companies that figure this out first are going to have a massive advantage. But the ones that get it wrong might find themselves with systems they can’t maintain or understand. That’s probably where the chaos comes from.

Alex Shannon: I wonder if this is connected to that OpenRouter funding we talked about earlier. If AI agents are becoming the primary way software gets written, then the infrastructure for managing and orchestrating different AI models becomes even more critical.

Sam Hinton: That’s a really interesting connection. Maybe Claude Code and OpenClaw are examples of specialized AI agents that need to work together with other models. You might have one agent for writing code, another for testing, another for deployment — all orchestrated through something like OpenRouter.

Alex Shannon: And if that’s the case, then the chaos might not just be about AI agents writing code — it might be about entire software development pipelines becoming autonomous. Imagine code getting written, tested, deployed, and monitored without human intervention.

Sam Hinton: That’s both incredibly exciting and terrifying. The speed of development could be unprecedented, but the potential for things to go wrong at scale is enormous. No wonder Wired called it chaos.

RAPID FIRE

Alex Shannon: Alright, let’s rapid fire through some other stories that caught our attention today. First up — early reports suggest Stord just raised $250 million to deploy something called ‘Physical AI’ across their fulfillment network.

Sam Hinton: Physical AI in logistics makes total sense. We’re probably talking about AI-powered robots, autonomous sorting systems, predictive inventory management. Amazon’s been doing this for years, but now it’s becoming accessible to smaller players.

Alex Shannon: Two hundred and fifty million dollars is a massive bet though. That suggests they’re not just adding some AI features — they’re fundamentally rebuilding their entire fulfillment infrastructure around AI.

Sam Hinton: And the timing makes sense. With labor costs rising and consumer expectations for fast delivery getting higher, physical AI might be the only way to stay competitive. That’s a quarter billion bet that the future of warehouses is fully automated.

Alex Shannon: I’m curious about the ‘Physical AI’ branding though. That feels like they’re positioning this as something different from traditional warehouse automation.

Sam Hinton: Yeah, probably AI that can adapt and learn in real-time, rather than just following pre-programmed routines. Think robots that can handle unexpected package shapes, or systems that optimize routing as demand patterns change.

Alex Shannon: And speaking of regulation, Colorado apparently passed a new AI Act targeting automated decision-making for what they call ‘consequential decisions.’

Sam Hinton: Colorado’s becoming the testing ground for AI regulation. This is probably about things like AI making hiring decisions, loan approvals, criminal justice recommendations. The fact that states are moving faster than the federal government tells you how urgent this is becoming.

Alex Shannon: The focus on ‘consequential decisions’ is interesting because it acknowledges that not all AI use cases are equal. Using AI to recommend movies is different from using it to decide who gets a job or a loan.

Sam Hinton: Exactly. And Colorado has been progressive on tech policy, so this could be a model for other states. We might see a patchwork of state regulations before the federal government figures out what to do.

Alex Shannon: Which creates compliance nightmares for companies operating across multiple states. But probably necessary given the pace of AI deployment.

Sam Hinton: Better to have imperfect state regulations than no regulations at all while AI systems make decisions that affect people’s lives.

Alex Shannon: Now here’s an interesting one — the Vatican is urging strict regulation of artificial intelligence in warfare, according to Human Rights Watch.

Sam Hinton: When the Vatican weighs in on tech policy, you know we’ve reached a moral inflection point. They’re probably thinking about autonomous weapons, AI-powered surveillance, the ethics of machines making life-and-death decisions. This isn’t just about technology anymore — it’s about the soul of humanity.

Alex Shannon: And the fact that Human Rights Watch is amplifying this message suggests there’s growing international concern about military AI. We could be heading toward some kind of international treaty or agreement.

Sam Hinton: The Vatican has moral authority that transcends national boundaries. If they’re calling for strict regulation, that could influence countries that might otherwise be hesitant to limit military AI development.

Alex Shannon: It also raises questions about where the line is between legitimate defense applications and problematic autonomous weapons. That’s a conversation we’re going to need to have as these technologies advance.

Sam Hinton: And it’s not just about weapons. Military AI includes surveillance systems, interrogation tools, border security. The Vatican is probably concerned about the broader militarization of AI technology.

Alex Shannon: And finally, Google introduced something called Gemini Omni, though details are still pretty sparse from their blog announcement.

Sam Hinton: Gemini Omni sounds like Google’s answer to GPT-4o — probably multimodal, real-time, designed to handle text, voice, images, maybe video all in one seamless experience. Given the Search backlash we talked about, this might be Google’s way of showing they’re still innovating even when users are pushing back.

Alex Shannon: The timing is interesting though. Announcing a new AI model right after getting backlash for forcing AI into Search feels like Google saying ‘we’re doubling down, not backing down.’

Sam Hinton: Or maybe it’s their way of giving users more AI options rather than just the one-size-fits-all approach they took with Search. Gemini Omni could be positioned as a separate product rather than a replacement for existing services.

Alex Shannon: We’ll have to see what the actual capabilities are, but the name suggests they’re going for comprehensive, all-in-one AI interaction.

Sam Hinton: Which brings us full circle to the multi-model conversation we had about OpenRouter. Maybe the future isn’t one omni-capable AI, but multiple specialized AIs working together.

BIGGER PICTURE

Alex Shannon: If you zoom out and look at everything we covered today, there’s this interesting tension emerging between AI progress and user acceptance.

Sam Hinton: Yeah, it’s like we’re in this phase where the technology is advancing faster than people’s comfort level with it. Google pushes AI Search, users revolt. AI agents write code autonomously, chaos ensues. The capability is there, but the social systems to handle it aren’t.

Alex Shannon: And at the same time, you have companies like OpenRouter getting billion-dollar valuations by making AI more accessible, while security vulnerabilities are affecting millions of AI deployments. It’s progress and peril happening simultaneously.

Sam Hinton: What strikes me is that we’re moving from the experimental phase to the infrastructure phase. These aren’t just cool demos anymore — they’re becoming the foundation that other systems depend on. And that brings both massive opportunities and massive risks.

Alex Shannon: The OpenRouter story really illustrates this. A year ago, most developers were just trying to figure out how to use any AI model. Now they want sophisticated orchestration, cost optimization, performance monitoring — all the enterprise-grade features.

Sam Hinton: And the Starlette vulnerability shows the flip side of that maturation. When millions of AI systems depend on the same open source foundations, a single flaw can have massive ripple effects. We’re building critical infrastructure on top of infrastructure we don’t fully control.

Alex Shannon: Then you have the regulatory response, which feels like it’s trying to catch up. Colorado passing an AI Act, the Vatican weighing in on military AI — institutions are recognizing that this technology has moved beyond the experimental phase and needs governance.

Sam Hinton: But the pace of change is so fast that regulation almost feels reactive rather than proactive. By the time we figure out how to govern AI Search or autonomous coding agents, the technology has already moved on to the next breakthrough.

Alex Shannon: The question I keep coming back to is whether we’re building the right guardrails as we go, or if we’re moving so fast that we’re going to have to retrofit safety and security and user choice after the fact.

Sam Hinton: I think 2026 might be remembered as the year AI went from ‘interesting experiment’ to ‘critical infrastructure.’ And with that transition comes all the messy human questions about control, transparency, and who gets to decide how these systems work.

Alex Shannon: And there’s this interesting divergence happening. You have companies like Google making top-down decisions about how people should interact with AI, while companies like OpenRouter are building more flexible, user-controlled infrastructure. Two very different philosophies about who should be in control.

Sam Hinton: That’s a really good observation. Google’s approach with Search is ‘trust us, we’ll give you the best experience.’ OpenRouter’s approach is ‘here are the tools, you decide how to use them.’ Those represent fundamentally different views of user agency in the AI era.

Alex Shannon: And the user revolt against Google’s AI Search suggests that people aren’t ready to give up that agency, even if the AI provides objectively better results. There’s something about having choice and control that matters beyond just efficiency.

Sam Hinton: Which ties into the broader regulatory questions. If AI systems are making decisions that affect people’s lives — whether it’s what information they see, whether they get hired, whether they get a loan — then questions of transparency and user control become matters of democratic participation.

Alex Shannon: That’s probably why institutions like the Vatican are getting involved. This isn’t just about technology efficiency — it’s about preserving human dignity and agency in an increasingly automated world.

Sam Hinton: And the security vulnerabilities add another layer. If we’re building critical systems on AI infrastructure, but that infrastructure has fundamental security flaws, we’re creating systemic risks that could affect entire sectors of the economy.

Alex Shannon: Keep an eye on user pushback as a leading indicator. When people start actively rejecting AI features they used to be excited about, that’s a signal that the industry needs to slow down and listen.

Sam Hinton: And watch for the companies that figure out how to give users more control rather than less. The future probably belongs to whoever can harness AI’s power while preserving human agency and choice.

Alex Shannon: The Claude Code and OpenClaw story we talked about feels like a preview of what’s coming. AI agents that can autonomously develop and deploy software could create incredible productivity gains, but also unprecedented risks if we don’t get the governance right.

Sam Hinton: And all of this is happening while the technology itself continues to advance at an exponential pace. The AI models being deployed today will seem primitive compared to what’s coming next year. We’re trying to govern a moving target.

OUTRO

Alex Shannon: That’s a wrap on today’s Build By AI. Thanks for diving deep into the messy, complicated reality of AI with us.

Sam Hinton: If you’re getting value from these conversations, hit that subscribe button and tell a friend. We’re back tomorrow with more stories from the front lines of artificial intelligence.

Alex Shannon: I’m Alex Shannon.

Sam Hinton: And I’m Sam Hinton. See you tomorrow.