I think the most important thing expert, AI-leading marketers should be doing right now is shifting the narrative: I think (my opinion, per usual I could be wrong so let's chat in the comments!) we need to stop telling the world that "anyone can build an AI agent" and start emphasizing everything required to build a truly successful, secure, and performant AI system for PPC.

Where's this coming from? I see brand operators saying, "Why pay an agency? I'll just use Claude," with zero idea of the extra risk and energy they are taking on. I also see non-PPC experts beginning to frame themselves as offering competitive services to PPC experts "because Claude is the media buyer now". We aren't properly communicating the real complexity of a production-grade PPC agent experience. Both in the complex, individual real-time skills to do PPC well, and the technical side to building a proper agent that will actually last.  This means leaning out of the "anyone can do this!" and leaning into the messy reality behind the scenes in AI: complex data organization, security, edge-case management, and why relying on Claude's base model (or a quick pre-built connector) isn't enough.

IMO, non-PPC and non-AI experts need a healthy dose of doubt about their ability to maintain these systems and confident sounding Claude isn't properly framing this... until it gets to a moment when pushed when it finally admits (AFTER doing the thing) "you're absolytely right, I was wrong about that and I apologize. I don't actually have an answer to that". The hidden cost of learning, building, and maintaining a live agent is real, and for most brands, it is far safer (and cheaper) to buy a tested, ready-made system or software instead.

Look at the work folks like Alfred Simon, Mike Rhodes 💡, and Frederick Vallaeys are putting into their administrative, deeply-tested systems. These three aren't just building agents, so it's why I point to them: they're building agentic infrastructures.

That kind of infrastructure should be the foundation for whatever the rest of us normies are building.I truly believe that in one year, FEWER people will be even attempting to build agents from scratch. More people will go back to paying for established software or proven frameworks because the first wave of costly mistakes, broken agents, and security slip-ups will have rolled in. (Think about it, a software has an entire team of developers testing and implementing fixes before they push it out, and then they charge pennies on the dollar of the time it takes for you with ChatGPT or Claude to figure out what's going on).

Definitely some sort of combination (MCP!!) is the future, individualistic management and advice really is fantastic... but it's got to be built on a structure. This mental shift feels like an important maturity milestone for AI in business. We're moving from the "look what I built" phase to the "is this actually stable and safe" phase.