From Code to Commerce: The AI Agent Takeover

Alex: Hey everyone, welcome back to Daily AI Digest! It's November 6th, 2026, and we've got a jam-packed show for you today.

Jordan: We really do. We're talking Cisco betting the farm on AI rewriting its entire codebase, Perplexity jumping into the agent wars, and Delaware potentially letting AI run actual businesses with zero humans in charge.

Alex: Plus a skeptical take on AI research hype and a browser automation tool that's apparently just... faster. Lots to get into.

Jordan: But before we dive in, did you see that BBC Verify piece asking whether data centers swayed the US midterms?

Alex: I did! Apparently nothing unites people faster than a giant humming building draining the local power grid.

Jordan: Turns out AI can write your code, but it still can't win you an election.

Alex: Yet. Give it a few more news cycles. Okay, speaking of AI writing code—let's get into our first story, because this one is wild.

Jordan: This is from Hacker News, originally reported on, and it's about Cisco. They've apparently handed a 27-year-old engineer what's being called 'unlimited' AI tokens and a 24-month mandate to rewrite 98% of the company's legacy codebase.

Alex: Wait, 98%? That's basically the whole company's software. And one engineer is in charge of this?

Jordan: One engineer, championed personally by Cisco's president, which tells you this isn't some skunkworks side project. This has executive air cover and, apparently, an open spending tap for AI compute.

Alex: Okay, 'unlimited tokens' is the phrase that's breaking my brain a little. What does that even mean practically? Like, there's no budget ceiling at all?

Jordan: Functionally, it sounds like it. Which is a fascinating risk bet, because normally in enterprise IT you're fighting tooth and nail for budget approval. Here they're saying cost is not the constraint—speed and coverage are.

Alex: But legacy code at a company like Cisco isn't just some old app nobody uses. We're talking networking infrastructure, routers, switches—stuff that if it breaks, breaks the internet a little bit.

Jordan: Exactly, and that's the part that should make everyone's palms a little sweaty. Rewriting 98% of a codebase isn't like refactoring a function here and there—it's an enormous surface area for subtle regressions. Things that pass every unit test but behave differently under some edge case nobody thought to check.

Alex: So how do you even validate something at that scale? You can't have a human review every line if AI is generating that much code that fast.

Jordan: Right, which means they almost certainly have to lean on AI for the validation too—automated test generation, behavioral diffing against the old system, shadow deployments. It becomes AI checking AI's homework, with humans spot-checking the spot-checkers.

Alex: That feels like it could go really well or really, catastrophically not well.

Jordan: That's exactly why this is such a big deal to watch. This isn't a startup experimenting—this is a Fortune 500 company treating AI coding tools as the primary engineering driver, not just an assistant that autocompletes your functions.

Alex: It's kind of a bellwether, right? If this works, every enterprise CTO is going to point at Cisco in their next budget meeting.

Jordan: And if it doesn't, it becomes the cautionary tale in every AI risk management deck for the next decade. Either way, 24 months from now, this is a story people will still be talking about.

Alex: No pressure for that 27-year-old engineer, by the way.

Jordan: None at all. Just the entire company's technical future resting on their shoulders. Speaking of AI taking on bigger and bigger roles—let's talk agents, because Perplexity just threw their hat in the ring.

Alex: This is The Verge story, right? Perplexity's new thing?

Jordan: Yep, The Verge reported that Perplexity has launched something called 'Teammates'—always-on AI agents that can take on multiple roles within a workspace. They're directly going after OpenAI and Anthropic's agent products.

Alex: Teammates. Okay, that's a very deliberate branding choice. They want you to think of this thing as a coworker, not a tool.

Jordan: Exactly, and that's the whole shift happening right now. We've moved from single-purpose chatbots—ask a question, get an answer, done—to persistent, role-based agents that have memory, context, and can run workflows on their own over time.

Alex: So like, instead of me asking a chatbot to summarize a document every time, I'd have a 'teammate' that's always kind of there, doing ongoing work?

Jordan: Right, and these integrate with something called Perplexity Computer, which lets them deploy automated workflows pulling from your existing projects, skills, and memory. So it's less 'answer my question' and more 'go handle this recurring piece of my job.'

Alex: Okay but how is this actually different from what OpenAI and Anthropic are already doing? Because it feels like every major AI company announced some version of 'agents for work' this year.

Jordan: That's the real question, and honestly the honest answer is: differentiation is thin right now. Perplexity's angle has always been search and retrieval—they're very good at grounding answers in real-time information. So their pitch is probably going to be agents that are more reliably accurate and sourced, versus just fluent.

Alex: So less 'creative agent that writes your emails' and more 'research agent that actually gets its facts straight.'

Jordan: That's my read on it. But we're clearly in an arms race phase. Every few weeks there's a new agent product launch, and most of them sound remarkably similar on paper—memory, roles, workflows, integration with your existing tools.

Alex: It's like the streaming wars but for AI agents. Everybody's got a slightly different flavor of the same basic thing.

Jordan: Pretty much, and the real test is going to be which ones actually get adopted inside companies and stick around, versus which ones are more flashy announcement than daily-use tool.

Alex: Well, speaking of agents getting more autonomy—this next story takes that idea to a genuinely wild extreme.

Jordan: Oh, this one's a doozy. Also from Hacker News—Delaware is considering legislation that would let AI agents legally run entire businesses without any human oversight at the top.

Alex: Wait, hold on. Run the business. Like, be the CEO? No human has to be in charge at all?

Jordan: That's the proposal, yeah. Delaware's a big deal here because it's where a massive chunk of US corporations are legally incorporated, so whatever Delaware does on corporate law tends to ripple out nationally, sometimes globally.

Alex: Okay, my brain immediately goes to: who's liable when the AI makes a terrible decision? Like, who gets sued?

Jordan: That is THE question, and honestly nobody has a clean answer yet. Right now corporate liability frameworks are built around humans—officers, directors, people with a name and a conscience who can be held accountable. An AI agent doesn't have that.

Alex: So is it the company that built the AI? The person who deployed it? The shareholders? That seems like an impossible legal tangle.

Jordan: It really is, and that's exactly why this is such fascinating and uncharted territory. It's not just about whether AI agents are capable enough to run a business—a lot of them probably are, for simpler operations—it's about whether our legal system has any framework at all for treating a non-human as a decision-maker with consequences.

Alex: This feels like one of those things that sounds like sci-fi until suddenly it's just... the news.

Jordan: Right, and if Delaware actually passes something like this, it could set precedent that other states, and honestly other countries, look to when figuring out their own approach to AI agent personhood in commerce.

Alex: AI agent personhood. That phrase alone is going to keep law professors employed for the next twenty years.

Jordan: Oh, easily. We're watching the tools-to-legal-actors transition happen in real time, which honestly connects right back to our first two stories—agents writing code, agents doing office work, and now potentially agents legally running the whole show.

Alex: It's the theme of the day, basically. Okay, but let's pump the brakes a little, because our next story is kind of a reality check on all this AI momentum.

Jordan: Yeah, this is a good palate cleanser. Also from Hacker News, there's a widely discussed LessWrong post—67 points, 50 comments, which for that community is a lot of engagement—arguing against the idea that AI is driving genuinely ingenious R&D breakthroughs.

Alex: So this is basically someone saying, hold on, is AI actually doing groundbreaking science, or are we just telling ourselves that story?

Jordan: Pretty much. It's pushing back on this narrative you hear a lot—that AI is accelerating scientific discovery, finding new materials, solving open problems. The author's basically saying: show me the actual ingenuity, because a lot of what's being celebrated is more incremental or just AI being a very fast, very good search-and-synthesis tool.

Alex: Which, to be fair, is still useful! Being a fast research assistant isn't nothing.

Jordan: Totally, and I think that's the nuance here. The pushback isn't 'AI is useless for research,' it's 'let's not conflate useful tooling with genuine creative breakthroughs.' Those are different claims, and the hype cycle tends to blur them together.

Alex: And the fact that it got 50 comments on LessWrong tells you it touched a nerve.

Jordan: Definitely. That community is pretty deep in AI research circles, so when a post like this generates that much discussion, it usually means people have been quietly thinking the same thing and this gave them permission to say it out loud.

Alex: It's a good reminder, especially after talking about Cisco rewriting its entire codebase and Delaware letting AI run companies—like, let's not get too far ahead of what these systems can actually do.

Jordan: Right, grounding expectations matters. The capability is real, but so is the hype, and separating those two is genuinely hard work right now.

Alex: Okay, well, on the more concrete, down-to-earth end of the spectrum—let's talk about our last story, which is a lot more technical and, frankly, kind of fun.

Jordan: Yes, this is a Show HN post, also Hacker News. It's a tool called Mantis, and the headline claim is that AI agents can browse the web 36% faster with it than with Playwright MCP, which is one of the more popular existing solutions.

Alex: Okay, for people who don't live in developer land, what is Playwright MCP, quickly?

Jordan: Playwright is a browser automation framework, and MCP stands for Model Context Protocol—it's basically a standard way for AI agents to connect to tools and data sources. So Playwright MCP lets an AI agent control a web browser—click buttons, fill forms, read pages—as part of doing some task.

Alex: And Mantis is saying it does that same job, but over a third faster?

Jordan: That's the claim, and speed here actually matters a lot more than it might sound. If you've got an agent doing research, or shopping, or data entry across dozens of web pages, that 36% compounds fast. Slow, unreliable browsing is one of the biggest bottlenecks in agentic workflows right now.

Alex: Because the web wasn't built for robots to click around on it, it was built for humans with eyeballs and patience.

Jordan: Exactly, pages are messy, layouts shift, there's ads and popups and cookie banners—agents have to parse all of that just to do what a human does without thinking. So any tool that makes that faster and more reliable is a genuinely big deal for anyone building real-world agents.

Alex: It's kind of the unglamorous plumbing side of the agent boom, isn't it? Everyone talks about the flashy 'AI runs your business' stuff, but somebody's got to make the thing actually click the right button.

Jordan: That's such a good way to put it. And honestly, we're seeing a growing ecosystem of these MCP alternatives and complements—tools built specifically for agent tooling rather than retrofitted from human-browsing frameworks. That infrastructure layer is going to quietly determine how good all these flashier agent products actually are.

Alex: So Mantis is kind of a small story today, but it's the sort of thing that props up everything else we talked about—Teammates, Delaware's AI-run businesses, all of it.

Jordan: Right, it's all connected. The agent wars, the autonomous business question, even Cisco's code rewrite—underneath all of it is this unglamorous question of whether the tooling is actually solid enough to trust with real stakes.

Alex: Which honestly feels like the theme that ties our whole episode together today.

Jordan: It really does—from code, to agents at work, to agents literally running companies, the question keeps coming back to: how much can we actually hand off, and are we ready for what happens when we do?

Alex: A question I don't think anyone has a clean answer to yet, but it's going to keep coming up all year.

Jordan: Oh, for sure. We'll be right here tracking it as it unfolds, story by story.

Alex: That's all for today's Daily AI Digest. Thanks so much for listening, everyone.

Jordan: We'll be back tomorrow with more. Until then, take care, and maybe double-check what your AI teammate is up to.

Alex: See you next time!