What If You Could Just Talk To Your Website? | Practical AI Ep 43

Practical AI: Episode 43

What If You Could Just Talk To Your Website?

Watch on YouTube

Published: May 29, 2026

TL;DR

  • A non-developer edited a live website by talking to it. Olga connected Claude to her real business site (running on PageMotor), and just by describing what she wanted, it built a new on-brand page and added a navigation menu linking to it — no code, no page builder.
  • The week’s theme: AI stopped being about the best model and became about who can use it. The “land grab” moved to the operational layer of every business.
  • Anthropic raised $50 billion — the second-largest private round in history, behind only OpenAI’s $122B in March — the same week it joined a Vatican AI encyclical and opened a Seoul office.
  • An AI found 10,000+ security flaws in 30 days, including one hiding in OpenBSD for 27 years.
  • $52.3 billion went into AI this week, and $50 billion of it went to one company. Strip Anthropic out and all of AI raised $2.3B.

Table of Contents

About This Show

Practical AI is a weekly live show (Fridays 11am CT) hosted by Olga Pechnenko and Chris Pearson that cuts through AI hype to deliver news, trends, and hands-on tips for builders and founders. Unlike technical AI podcasts, Practical AI focuses on business applications and ROI — what actually works, what’s hype, and what you can implement Monday morning.

What You’ll Gain

  • See what “talking to your website” actually looks like. A non-developer edited a live business site — a new page and a new navigation menu — just by describing it to an AI. Watch the exact workflow and why it matters for anyone who owns a site.
  • Understand the real shift this week. AI is moving from “who has the best model” to “who can operate it,” and the businesses using it now are pulling ahead. Learn where that leaves you.
  • Learn why one company raised more than the rest of AI combined. Anthropic’s $50B round, what it signals about consolidation, and why the other 111 funded companies split just 7% of the money.
  • Discover the security wake-up call every business owner needs. An AI found 10,000 flaws in a month, including a 27-year-old one. The question is no longer “is my software secure” but “when does an AI find the hole.”
  • Gain a clear read on where your customers actually are. A platform-by-platform breakdown of who’s on Facebook, Instagram, LinkedIn, and X — and why most people are marketing to the wrong room.

Biggest Takeaway to Implement: Stop waiting on a developer or a page builder. Connect an AI to the tools you already run, describe the outcome you want in plain language, and let it do the work. The ceiling on what one person can build just moved — the people using this now are the ones who win.

PageMotor and Practical AI Updates

Free, informative, and FUN!

Frequently Asked Questions

Can you really edit a website just by talking to it?

Yes. In this episode Olga connected Claude to her live site (built on PageMotor) and, by typing plain-language requests, had it build a new page and add a navigation menu — matching her existing design, with no code and no page builder. Read more below.

How much did Anthropic raise, and is it the biggest round ever?

Anthropic raised $50 billion (some outlets report a $65B total at a ~$965B valuation). It’s the second-largest private round in history — OpenAI’s $122 billion in March is still the biggest. Read more below.

What is Claude Mythos and what did it find?

Mythos is Anthropic’s unreleased security model. Across 50 companies in 30 days it flagged more than 10,000 high or critical software flaws, including one hidden in OpenBSD for 27 years. Over 90% were validated as real. Read more below.

What are Claude Code’s dynamic workflows / “ultracode”?

A new Anthropic feature where Claude spins up a fleet of AI agents (about 16 at a time, up to ~1,000 across a job) that build and check each other’s work, doing in a night what used to take a team months. Read more below.

Why is everyone talking about Taiwan?

Nvidia announced ~$150 billion a year in Taiwan and AMD added more than $10 billion. Almost all of the world’s advanced AI chips are made on one island — a huge concentration of risk. Read more below.

Is Facebook really charging now?

Yes. Meta rolled out paid tiers — $3.99/mo for Instagram and Facebook, $2.99 for WhatsApp. The compute cost of AI is starting to crack the free-internet model. Read more below.

Practical AI: What If You Could Just Talk To Your Website?

Key Definitions

What is an AI-native CMS?

A content management system built so an AI agent can read and edit the site directly through an API — not just a human clicking around an admin panel. PageMotor is the example in this episode: Olga points Claude at her live site and it makes real changes, because the site is structured for an agent to operate.

What is “the operator era”?

The shift, named on this episode, from a world where value comes from building software to one where it comes from operating AI well. As AI does the building, the advantage moves to whoever can direct it, knows their own pain, and owns their audience.

What is “agent debt”?

Like technical debt, but for AI agents. When you hack together agent workflows fast and never clean them up, system prompts conflict, memory gets polluted, and tools overlap — until months later the agent behaves strangely and nobody knows why.

What is MCP (and why “the new SEO”)?

The Model Context Protocol is the open standard that lets AI agents connect to tools and data. The idea raised this week: if agents can’t find and use your business through MCP, you become invisible to them — making MCP the new version of being findable.

Quotable Moments

We’re moving away from who’s got the best model. We’re moving to the operational and practical application of it. And whoever is using this information right now is winning big time.

— Olga, on the week’s theme

I just talked to my site and it does things for me. I didn’t have to touch anything. I didn’t have to do anything.

— Olga, on the live demo

This idea that some knight in shining armor is just going to show up and fix all your problems is ridiculous. Or is it? The knight just walked in the door — now you complain to the AI, and it actually fixes the problem.

— Olga and Chris, on the operator era

Within five years, PageMotor will be the winner in software that writes its own software. It’ll beat Cognition even though they just raised a billion dollars.

— Olga’s bold prediction (Chris: “we’re writing it down”)

01:30 Mythos: 10,000 flaws in 30 days

The Mythos Numbers

Across 50 companies in 30 days, Anthropic’s unreleased Mythos model flagged more than 10,000 high or critical flaws across 1,000+ open-source projects — including a flaw hiding in OpenBSD for 27 years. Over 90% were validated as real; Cloudflare alone found 2,000 bugs in its own systems.

Two weeks ago the show covered Mythos finding that 27-year-old bug. This week the full scorecard landed, and it’s staggering. The reason a machine catches what thousands of brilliant engineers missed: humans read code sequentially and can’t hold an entire sprawling codebase in working memory, while the AI ingests the whole thing and maps every interaction at once. The unit of work went from a person reading one file for an hour to a machine analyzing millions of interactions in minutes.

The strategic point cuts both ways. Attackers now have the same near-free capability — a scan costs a few hundred dollars. But defenders can finally scan and patch their own code overnight. The question is no longer “is my software secure,” it’s “when does an AI find the hole that’s been hiding for a generation.”

Pro tip: Olga predicted Anthropic would open Mythos to the public by the end of May. It didn’t — still gated to the invited group over cybersecurity concerns. She owned the miss on air, which is exactly the kind of honesty that builds trust.

05:30 Claude’s agents that check each other’s work

Anthropic shipped dynamic workflows in Claude Code, switched on by a mode called “ultracode.” You ask for a workflow and Claude becomes an orchestrator, spinning up a fleet of agents — about 16 at a time, up to roughly 1,000 across a job. They aren’t just splitting a to-do list: half build a solution while the other half try to tear it apart, arguing and iterating until the result holds up, with no human mediating.

The 3am Engineering Team

The workflow can take a 3-month corporate migration and turn it into an overnight job. The cost is “token burn” — hundreds of agents arguing generate enormous compute. But if you can afford it, you have an entire engineering team and a security team at your fingertips.

Paired with Mythos, the takeaway is the week in one line: in a single week, one company shipped an AI that does the work of a security team and an AI that does the work of an engineering team. The ceiling on what one person can build just skyrocketed.

08:20 Anthropic’s $50B week

The Raise In Context

Anthropic raised $50 billion (some reports cite a $65B total at a ~$965B valuation) — the second-largest private round ever. Only OpenAI’s $122 billion in March is bigger.

The money is only part of the story. The same week, Anthropic opened a Seoul office (South Koreans use Claude at about 3.5x the population-adjusted rate), and its researcher Chris Olah took part in the rollout of Pope Leo’s 42,000-word AI encyclical on protecting the human person. It also quietly tightened its own safety policy around chemical and biological risks. One company is now a financial story, a geopolitical story, and a moral story at once.

Key Takeaway: When a single company can replace a security firm and an engineering team in the same week — and raise more than the rest of the industry combined — it stops being a software vendor and starts acting like a foundational pillar of the economy.

13:20 Why almost every AI chip comes from one island

You can have $50 billion and brilliant algorithms, but you can’t conjure the hardware. Nvidia’s Jensen Huang announced ~$150 billion a year across Taiwan’s ecosystem and broke ground on a new campus, calling the island “the epicenter of the AI revolution.” AMD followed with more than $10 billion of its own.

The Concentration Risk

Nearly all of the world’s most advanced AI chips are manufactured on one island. Every autonomous agent in every other story this week runs on that silicon. The entire industry is tethering its future to a single geographic point — because demand for compute is effectively infinite.

16:00 The AI that checks its own math

More compute doesn’t help if the AI just hallucinates faster — the only way to use it is if the machine can verify its own reasoning. That happened this week. A day after OpenAI claimed a math breakthrough that still needed human experts to confirm, Google DeepMind’s AlphaProof Nexus solved nine open Erdős problems — two unsolved for 56 years — and verified its own proofs automatically.

The mechanism matters: DeepMind paired a creative language model (the student tossing out ideas) with a formal logic solver (the ruthless professor that rejects any flawed step), looping until the proof is perfect. Once a machine can formally verify its own logic, it graduates from text predictor to reasoning engine. Notably, DeepMind’s Demis Hassabis tempered the hype — “still not AGI,” he said, calling it the foothills of the singularity, roughly four years out.

18:00 AI is inventing proteins to fight disease

The same autonomous reasoning hit the highest end of human health. The Chan-Zuckerberg Biohub launched an AI “world model” of protein biology, mapping nearly 7 billion proteins. A world model isn’t a database — it learned the underlying grammar of how proteins fold, the way ChatGPT learned English, so it can invent proteins that never existed in nature. Researchers have already used it to design novel binders targeting cancer and immune diseases, compressing years of drug discovery into hours.

20:00 AI moves into your everyday tools

The shift isn’t only in labs. OpenAI shipped a ChatGPT add-in that builds and edits PowerPoint slides natively, and Google wired Gemini directly into Adobe, Canva, and CapCut. The reasoning engine is becoming an invisible layer inside the software you already use — no separate browser window required.

Pro tip: Having the best AI wired into your tools does nothing if you don’t open it. It’s uncomfortable at first. Go play with the new AI features in the apps you already pay for — that’s the cheapest edge available right now.

22:00 Facebook is charging now (and made your community an asset)

Meta started charging for its apps — $3.99/mo for Instagram and Facebook, $2.99 for WhatsApp. The cause is undeniable: the compute cost of AI is forcing the end of the free, ad-supported internet. We’re moving from an internet of human content paid for by ads to an internet of machine reasoning paid for by subscriptions.

At the same time, Meta quietly launched Forum, a Reddit-style app that pulls in your Facebook Groups and adds an AI “Ask” tab. It looks like a clone, but it’s a data play: Meta owns the biggest pile of real human conversation on earth and is using it to ground its own AI — it even says so on screen. The deeper read for builders is the platform-audience reveal: X and LinkedIn are the smallest rooms by active users; Facebook and Instagram are far bigger. Most people teaching and selling AI are shouting in the wrong room.

Key Takeaway: Your community is a real asset, but a community that lives on a platform is rented land — you can be blocked or buried overnight. Build where your people are, then own the connection (your email list) so the relationship survives the platform.

45:00 The demo: editing a live website by talking to it

What Actually Happened

Olga connected Claude to her live business site (running on PageMotor) through its API. By typing plain-language requests, she had it build a new on-brand newsletter page AND add a navigation menu linking to it — matching her homepage design, with a GEO summary written in automatically. Zero code. Zero page builder. The page part took about four minutes.

This is the episode’s title moment and the clearest proof of the week’s theme. Olga didn’t open an admin panel or fight a page builder — she worked in her Claude window, said “find my newsletter and put it on my site,” and it built the page. Then she noticed nothing in the navigation pointed to it, so she asked it to add an “Insights” item to the main menu — and it edited the live site’s structure to do it. The point isn’t that she built something. It’s that a non-developer ran and edited a live website by talking to it. If she can, anyone can.

The website of the future is two websites: one for humans, and one for the AI agents that now read and operate it.

67:00 The operator era

The deep dive built on Greg Isenberg’s dispatch from a week in San Francisco, where every founder and investor was circling the same idea: the build era is ending, and the operator era is starting. Billionaires are buying SaaS companies and rebuilding them agent-first. The hottest role is the “forward-deployed engineer” who sits between the agent and the customer making sure things actually work. And new hazards are emerging, like “agent debt” — the AI version of technical debt.

Pro tip: Model loyalty is fading. Olga uses Claude, Codex, and GPT for different jobs. The skill worth building isn’t picking one model — it’s knowing which tool to bring to which task, and keeping your work portable.

86:00 Funding: $52B into AI, $50B to one company

The Week’s Funding (Crunchbase)

$52.3 billion went into AI this week — 93% of all venture dollars. $50 billion of it was Anthropic. Strip Anthropic out and 111 other AI companies split just $2.3 billion — about 7% of the total. This wasn’t a funding boom; it was one historic round.

Below the giant, the money went to the picks and shovels: Modal Labs ($355M, compute on demand), Moffett AI (specialized chips), and OpenRouter ($113M, a single doorway to any model). Olga closed with a bold call: within five years, PageMotor will be the real winner in software-that-writes-its-own-software, beating Cognition despite its fresh $1 billion round. Chris’s response: “We’re writing it down.”

Keep Learning