Last week, I shared 50 things I've been doing with the AI agent Instinct. It convinced quite a few of you to request Instinct invites, and Instinct dutifully replied to all of you who requested one with a personalized invite. You people love trying out AI.

What isn’t Instinct doing for me yet? That exercise is often at least as instructive. If I came up with 50 ways I wasn't using it, I'd feel the need to say why, though, and that would be a slog of a column to read. Is it because of privacy concerns? Did I try, but it failed? What's stopping me, and what might stop you?

Here's a shortlist. I'm sharing scenarios where I’d consider enlisting an agentic AI app like Instinct, as opposed to countless hypothetical examples.

Vibe Coding

I enlisted Instinct to maintain and optimize a bunch of websites I built using Lovable. Then, I wanted to focus more on one of them, and I realized how much I missed using Lovable directly as a design and development partner. There's a conversation flow unique to certain apps. It's like the difference of going from working with a generalist to a specialist. This is also why I often prefer not to use MCPs and like to work in certain software directly. User experience still matters, at least sometimes.

Presentation Design

Instinct is very good for knowledge management, including a meta version of knowledge management where it keeps track of its own work. I used it to create a shell of a presentation for me based on everything I've written about vibe coding and agents. The information ls coherent, accurate, and well-organized. Visually, it's boring. I tried suggesting frameworks to use for visuals. No dice. Claude Design is better for presentations, and I'd imagine Google Slides will become a default AI presentation builder for many. Instinct fell short here, and this is also an area where it’s not worth going back and forth with an agent. It’s best to see results directly.

Capturing My Voice

I trained Instinct to sound more like me for email drafts, and it's good enough now. It remarked, as if surprised, that I don't curse in email like I do via text. Instinct is actually good enough to send a transactional email on my behalf. If an AI Insiders speaker is looking for a link to the event, for instance, I'm fine with Instinct responding (it does so only with my permission). But if a speaker asks for more nuance, like understanding AIMG's audience better, I'll step in. The more important a message is, the less AI is involved.

Client Work

This involves permissions. With all the experimentation I do with AI, if a client doesn't allow it, Instinct won't touch it. I use very different AI stacks for client work.

Wish Me Happy Birthday

It's my birthday month. I asked Instinct, "What are you getting me for my birthday?" It said, "A clean [Google] Search Console and Lovable credit overruns. Wrapped." Wow, it really gets me.

But then it said, "Real answer: When is it? I'll pick a $30-50 physical thing and send you one approval tap with the exact total."

This baffled me. It has done genealogy research for me. It has access to my email and calendar. I'm already getting some birthday month promotions from certain businesses (thanks for the early greetings, Charles Schwab).

Given how smart AI can seem, it can also be very, very dumb. It is still, and will always be, artificial intelligence. Don't assume agents' algorithms have any special superpowers, and you can probably find some smart ways to put them to work.

— David Berkowitz, Chief Community Officer, Marketecture Media

1

BaD Mktg's Good Math: Two People, $2 Million

Who: Agency Owners, Independent Agency Executives, Creative Directors, Freelancers

What: BaD Mktg, a two-person Toronto agency run by holdco veterans Ian Buck and Yotam Dor, has generated more than $2 million in revenue over three years by running its back office and concepting on LLMs, Digiday reports. Their second proprietary tool, Goodie, automates priorities, payments, and vendor management with human financial checks, and the pair say they operate like a shop with a dozen staffers, cutting concepting time from six weeks to six days or less. Their twist on AI creative: They now pitch clients full AI-generated spots instead of storyboards, while keeping final production in human hands.

Why it matters: The two-person, seven-figure agency is no longer a thought experiment. The telling detail: They sit on finished work so clients don't see how fast it actually was, because the billing model hasn't caught up with the production model. Want to meet them virtually? Join our upcoming AI Insiders event.

(Digiday)

2

SheaMoisture Treats Every Second as a Checkout

Who: Brand Marketers, Creative Directors, CPG Marketers, Influencer/Social Leads

What: Unilever-owned SheaMoisture is rebuilding content around the idea that any moment can be a purchase moment, creative director ShaNiece Pyles told The Drum: testing CTA placement in the first three to five seconds, loosening influencer briefs, and casting everyday consumers over reach. On AI, Pyles draws a clear line: It speeds post-production and explores early ideas, but it does not make creative decisions, because "if the casting and the lighting and the hair are wrong, AI is gonna put out whatever it puts out." Her bigger worry is talent: If automation strips out the learning-by-watching, juniors "won't even know how to prompt the robots."

Why it matters: This is a brand creative chief articulating the human-in-the-loop model in specifics, not slogans, plus the mentorship-erosion problem nobody's AI deck mentions.

(The Drum)

3

An Agency Buys CTV From Inside a Chatbot

Who: Agency Media Buyers, CTV and Video Teams, Ad Tech Executives, Performance Marketers

What: Independent agency InterMedia is running agentic CTV buying in production, wiring Vibe.co's Claude integration for reporting and actions and Olyzon for cross-channel decisioning. Staff query campaign data in Claude's dashboard and order changes conversationally; Vibe says more than 1,000 brands have adopted its MCP integration and MCP-enabled buying is now a large share of platform spend. The agency's reframe of expertise: "Instead of being an expert in how to navigate a platform and push the buttons, you're an expert in being judicious about what changes to make."

Why it matters: Agentic buying left the demo stage in CTV before most of the industry noticed. The job description shift in that quote, from button-pushing to judgment, is the one every media team is about to have.

(AdExchanger)

The room every marketer wants to be in is returning to New York City.

On March 9-10, 2027, Marketecture Live will be back at The Glasshouse NYC for a two-day event that will convene leading executives across brands, agencies, publishers, and adtech.

Wondering what the hype is all about? Marketecture Live takes a different approach: expertise first, no panels, no filler, no fluff. Across more than 40 sessions and breakouts, every conversation is designed to be sharp, relevant, and worth your time.

The event has consistently sold out, so don’t wait to lock in your ticket at the best price you’ll find.

SECURE YOUR SEAT