Feb. 26, 2026

AI Agents are here: ChatGPT, Google Pomelli & OpenClaw

AI Agents are here: ChatGPT, Google Pomelli & OpenClaw
Apple Podcasts podcast player iconSpotify podcast player iconOvercast podcast player iconYouTube podcast player icon
Apple Podcasts podcast player iconSpotify podcast player iconOvercast podcast player iconYouTube podcast player icon

ChatGPT launched ads this week and they're underwhelming. No personalized targeting, no memory integration, just basic in-line search ads that feel like Google in the 90s. Meanwhile, OpenClaw is setting the internet on fire as the first AI agent that actually works—controlling your browser, joining meetings, managing calendars, and doing everything a computer can do autonomously. Created by Austrian developer Peter Steinberger, it's the most famous open source project of all time with 200,000 GitHub stars and sparked a bidding war between Meta and OpenAI. It's getting normies excited about AI in a way nothing else has, with people buying Mac minis and naming them like employees.

This week Sonia sits down with Martin O'Leary (head of marketing at Tap Payments, writer of Uncharted newsletter) to break down ChatGPT's boring ad rollout, Google's Pomelli Photo Shoot AI tool turning product snapshots into studio-quality marketing images, F1's Netflix tie-in with Damson Idris, AMC's AI film controversy, and why Spotify's Bad Bunny Billions Club Live event is a masterclass in using CRM data to reward actual fans instead of chasing new customers.

The big thesis: craft still matters more than AI, but AI is abstracting complexity in ways that create new opportunities for builders. ChatGPT's ads prove OpenAI doesn't understand personalization yet. Google's Pomelli tool will make everyone's product photos look the same unless brands invest in distinctive assets. OpenClaw is bringing the future forward by letting one person build what used to require teams. And Spotify is doing what every brand sitting on CRM data should be doing: rewarding top fans with exclusive experiences instead of gating events for clout.

We're talking about:

  • ChatGPT ads: Why they're underwhelming, lack of memory integration, missing personalization opportunities, and how they compare to Facebook's data trove

  • F1's Netflix crossover: Damson Idris (star of the F1 movie) appearing in the Driver to Survive promo, Apple taking over F1 broadcasting, and Carlos Sainz acknowledging Thea (the girl who created Sparkles the unicorn mascot)

  • Google Pomelli Photo Shoot: AI tool turning basic product photos into studio-quality marketing images, targeting e-commerce sellers, integrating with Google Ads and Merchant Center, and competing with Amazon and Shopify

  • Why AI product photography risks: Reverting to the mean, losing brand differentiation, DoorDash's AI food photos looking too clean, and why craft + prompting beats automation alone

  • Real estate using AI: Justine Moore's observation that realtors are early adopters, creating multiple views of homes (day/night/sunset/party mode), and why it works (showing aspirational living without hiring drone operators)

  • OpenClaw phenomenon: Austrian developer Peter Ticksteinberger's open source AI agent, 200,000 GitHub stars, bidding war between Meta and OpenAI, normies buying Mac minis as "employees," and why it's the first AI tool exciting people outside the tech bubble

  • How OpenClaw works: Persistent memory across conversations, soul.md and identity files for customization, integrations with Slack/WhatsApp/Telegram/Gmail, and why it's a chief of staff + EA + COO combined

  • Designing for agents vs humans: Websites optimized for AI scraping (pricing, specs, structured data) vs brand storytelling, billboards becoming prompts for humans to prompt their agents, and why real-world activations (Super Bowl, Olympics, World Cup) will matter more

  • Spotify's Bad Bunny Billions Club Live: Invite-only Tokyo show for top listeners, leveraging Bad Bunny's 28 tracks in the Billions Club, rewarding actual fans instead of gating for clout, and why more brands should use their CRM data this way

  • CRM data goldmines: Banks, cable companies, phone companies rewarding new customers over loyal ones, why discount codes only support churners or new users, and flipping the model to reward top fans with exclusive experiences

  • AMC AI film controversy: Igor Farov's Thanksgiving short made with Gemini Nano Banana Pro, Screen Vision competition, backlash over AI-generated work in cinemas, and the craft vs livelihood debate (echoes of Snow White in 1937, animation not in Oscars until 2002, Luddites protecting livelihoods not rejecting technology)

  • Sam Altman's bad week: Responding to Anthropic's Super Bowl ad with an essay (if you're explaining you're losing), saying humans need training too (tone-deaf in America), and why OpenAI needs a designated hater

  • The designated hater thesis: Every company needs someone independent and well-paid whose only job is to say "do not release that crap" (McDonald's AI Christmas ad, Coca-Cola AI disaster, Mount Rushmore metaphor)

Plus: Why F1 movie sequels are in production, Martin's theory that OpenClaw agents will game Spotify's top listener algorithm, and the Don Draper "I don't think about you at all" moment OpenAI should've had with Anthropic