Category: GenAI

  • Meta’s Advantage+ is Quietly Hijacking Brand Control

    Lately I’ve been feeling that Meta is getting a little too comfortable making decisions for our brands.

    My own experience with Advantage+ has been nothing but frustration and I’ve been hearing more and more complaints from both agency and in-house folks about Meta’s Advantage+ features being enabled by default across custom audiences, ad set targeting, and even worse… creative. On the surface, it sounds helpful… just let the algorithm optimize everything, right? Right?!

    In the end, brands are losing control of their own creative. We’re talking about Meta automatically adding music, backgrounds, and text overlays to carefully crafted brand content. No consideration for brand guidelines, visual identity, or the brand world you’ve spent years building.

    Imagine spending months developing a campaign that perfectly captures your brand world, only to have an algorithm slap a random background and some generic music on it because it thinks it’ll perform better.

    And if you’re in a restricted industry like alcohol, pharma, or financial services? This becomes a compliance nightmare. Alcohol brands have incredibly strict advertising guidelines where every word, image, and implication gets scrutinized. Now we’re supposed to trust an algorithm to automatically modify that content without violating advertising standards or legal requirements? Mind you, this is the same algorithm that pushed gore, violence, and death to people’s feeds in an “oopsie” a few months ago. Hm, no thanks!

    What’s even more frustrating is the operational burden this creates. I’m hearing from agency buyers that they’re having to increase billable hours just to spend extra time making sure these features are turned off. Meta uses some pretty sneaky UI tricks to keep them enabled by default like buried toggles and confusing opt-out flows. There is no singular button to disable it in Ads Manager nor disable it from your ad account or business manager.

    So now agencies are billing clients more time to fight against features that were supposed to make things easier. That’s backwards.

    I get that automation can drive performance. But there’s a line between optimization and completely surrendering brand control to an algorithm that doesn’t understand your brand story, doesn’t know your visual guidelines, and definitely doesn’t care about the consistency you’ve worked so hard to build.

    When did we decide that slightly better CTRs were worth giving up our brand identity? How did we end up in a place where media buyers are “approving” creative edits?

    This is a fundamental role reversal where traditionally the creative process flows from brand → creative team → media team for placement. Now it’s becoming brand → creative team → algorithm → media team scrambling to notice what got changed.

    Somewhere deep in the cubicles of Meta’s offices is a product manager who knows better but has to implement Gen AI features because of a mandate that conflicts with their customers needs.

  • Delegate & Done: How AI Changed the Way I Shop Online

    We make an average of 35,000 decisions a day. By the time you’ve figured out dinner, scanned your inbox, and doom-scrolled through 40 tabs for the “perfect” gift… your brain is begging for mercy.

    That’s why when I needed to find a birthday whiskey gift for my Scotch-loving uncle, I didn’t add anything to cart, I delegated it to AI.

    From Tabs to Tasks: The Agent That Shopped For Me

    I used ChatGPT’s Agent platform (an actual agent, not just the chatbot) to research whiskey options within a $40–$60 budget.

    Here’s what I gave it:

    • ✅ My birthdate (for site age-verification modals)
    • ✅ My uncle’s shipping address in New Jersey
    • ✅ A simple prompt: “He likes Scotch but is open to anything good.”

    And just like that, 20 minutes later I got a tidy Excel sheet dropped into my inbox.

    Here’s a peek at what it delivered:

    Bottle Price Notes
    Johnnie Walker Black $39 A smoky classic. Reliable. Won’t raise eyebrows.
    Talisker Storm $55 Bold maritime peat. Feels like a wave punched you lovingly.
    I.W. Harper Cabernet $60 Bourbon in wine casks. Smooth, slightly fancy.
    (+ 3 more thoughtful picks)

    I didn’t lift a finger. No comparison charts, no Reddit rabbit holes. Just: Delegate → Done.

    Why This Matters (and Why It’s More Than Just Whiskey)

    We spend 10+ hours a month comparison shopping. That’s over 120 hours a year of toggling tabs, reading Amazon reviews from “Karen in Omaha,” and trying to make one smart purchase decision.

    It’s exhausting. The psychologist Barry Schwartz coined this effect “the paradox of choice.” The more options we have, the worse we feel—even after we buy something.

    AI agents break that loop by:

    • 🔄 Automating research without overwhelming you with info
    • 🧠 Reducing decision fatigue and freeing up mental RAM
    • 📈 Giving just enough context to make you feel confident, not confused

    Today it’s whiskey. But tomorrow?
    Try:

    • ✈️ Vacation planning (Flight + hotel + itinerary in one go)
    • 💻 Laptop buying (Specs, reviews, comparisons—done)
    • 🏥 Healthcare plans (Yeah… I’m scared too, but AI might help)

    What I Actually Prompted (In Case You Want to Try This)

    Here’s the exact breakdown I gave the agent:

    Task: Find a whiskey gift for my uncle
    Budget: $40–$60
    Preferences: He likes Scotch but is open to other whiskey styles
    Constraints: Must be shippable to New Jersey, consider local laws
    Extras: Include direct purchase links and tasting notes

    The platform handled everything from price availability, shipping legality, and even a little flourish in the tasting notes that made it feel curated.

    I used OpenAI’s GPT-4o-based assistant and connected it with a research workflow. You could replicate something similar using:

    Traditional Product Research? Useful But Flawed

    Look, I’m not anti-research. Sometimes I want to obsessively compare five espresso machines I’ll never buy.

    But most of the time?
    We’re just trying to get to “good enough” without wasting a Saturday.

    Traditional comparison shopping:

    • ❌ Eats up time
    • ❌ Leads to second-guessing
    • ❌ Often ends with “screw it, I’ll just get the same one as last year”

    AI shopping flips that script.

    The Catch: Let’s Talk Limitations

    🧩 Biases in Results
    AI agents are only as good as their data sources. If your agent pulls only from sponsored listings, you’ll get sponsored outcomes.

    🔒 Privacy Concerns
    Yes, I gave an AI my zip code and a preference for Scotch. No, I didn’t feed it my SSN. But AI-assisted commerce will require stronger consumer protections over time.

    ⚠️ Over-Reliance
    The danger is blindly trusting output without verifying. Delegation shouldn’t mean abdication.

    What’s Next? From Curated Shopping to AI-Led Consumption

    E-commerce is moving fast:

    • Amazon is experimenting with AI-generated buying guides
    • Shopify merchants are building concierge bots
    • Entire startups are focused on AI-led travel planning and itinerary design

    This shift from “what do I buy?” to “what can you just figure out for me?” is a rethink of digital delegation.

    Final Thoughts (and an Invitation)

    AI agents aren’t here to replace your judgment. But they can replace your Saturday of endless Googling.

    Try it. Start with something low stakes like a gift. Then work up to higher-order stuff. You might be surprised by how much time and cognitive space you get back.
    Have you used an AI agent to handle a shopping task?
    Drop a comment below or contact me here. I’d love to collect more stories for a follow-up.

  • How I Turned an Idea into a Fully Functional WordPress Plugin (With a Little Help from Gemini) 🚀

    How I Turned an Idea into a Fully Functional WordPress Plugin (With a Little Help from Gemini) 🚀

    Ever wanted your StoryGraph reading lists to live right on your blog? I did and teaming up with Google’s Gemini as my AI vibe code partner turned it into a crash course in agile development and creative problem-solving.

    Lately, I’ve been on a mission to level up my Gen AI skills and not just in theory, but by actually building things. I’m not a developer by trade or even training, but I’ve been experimenting with what’s possible using tools like Claude, ChatGPT, and Gemini.

    Last night a project sort of took on a life of its own. I wanted to create a WordPress plugin that pulls in my StoryGraph reading lists and displays them on my blog. It started as a “let’s see if this is even possible” experiment and quickly evolved into a messy, fun, but very educational journey through AI-assisted coding and constant pivoting.

    During this project there were three big pivots:

    1️⃣ Scraping Strategy → 403s Everywhere

    Since The StoryGraph lacks an official API, the first idea was to have the plugin scrape my StoryGraph reading lists which was a solution that Gemini shared early on. Technically, it could’ve worked… but we immediately ran into 403 errors. StoryGraph’s anti-bot protection (shoutout to Cloudflare 👋) was not having it. And while the data was public, scraping without permission lives in a legal area that’s not something I wanted to mess with. It was time to pivot.

    2️⃣ RSS Feeds → Behind a Paywall → Deprecated Functionality

    Next up: RSS. We reworked the plugin to pull from StoryGraph’s XML feeds… only to hit another wall. Turns out they’re locked behind a “Plus” subscription. I ponied up and then found out the feeds were deprecated. Classic AI wild goose chase. Those “Check these statements. AI can be wrong.” warnings should indeed be heeded.

    3️⃣ Data Export → Bingo

    Finally, after digging through account settings, I spotted StoryGraph’s Export Data feature. One quick download later and boom: a clean, reliable source of truth. No scraping, no chasing ghosts. Just good old-fashioned data export.

    🔧 The Plugin Takes Shape

    The next pieces to build were the actual plug-in functionality which was much faster than figuring out how to retrieve the data.

    • Admin Upload: Built a simple upload screen in the WordPress admin where I can drop in the data file instead of relying on live fetches.
    • Smart Parsing: Gemini helped map out fields like Title, Author, and Read Status. We ironed out some bugs around file types and capitalization quirks.
    • Cover Art Magic: For a more visual display, we hooked into the Open Library API to grab book covers using ISBN numbers.

    📈 The Result

    What began as a plain text list can now be a polished, dynamic, and visual display of my reading life right on my WordPress site. This project wasn’t just about writing code. It was about thinking creatively, adapting on the fly, and using AI as a true vibe-coding partner to get to a smarter solution.

    🔗 You can check it out it here: jamesk.xyz/books