The final day of MAICON 2026 opened with the most disturbing session I attended all week. It closed with the one that left me most hopeful. I have spent the day since trying to hold both of them in my head at once, because I think both were right.
Karen Hao made me look at who pays for all of this
Karen Hao, the investigative journalist who wrote Empire of AI, opened the morning. Her session was extremely disturbing and scary.
Her argument is that the largest AI labs operate like historical empires. They take data nobody agreed to give. They depend on cheap labor, including workers in Kenya paid very little to label graphic violence for OpenAI, at a serious psychological cost. They keep research locked inside a few companies. The data centers behind their models pull water and power from the communities around them. She contrasted the giant general-purpose models she calls rockets with smaller, purpose-built systems she compares to bicycles, like AlphaFold.
Then she was asked whether she uses these tools herself. She doesn't. She abstains from generative AI built by the major labs, partly on ethical grounds and partly because the company running a closed platform can review anything you type into it. For a reporter protecting sources, that alone rules them out.
I use Claude every day. I run an AI consultancy and teach law firm marketing teams how to put these tools to work. So this was a hard hour to sit through, which is exactly what it should have been. A handful of companies now have an enormous influence on where our society goes. I use their products all day and rarely weigh that.
It has me seriously considering open-weight models, which anyone can download and run on their own systems, for more of our work. Hao described them as a transitional step: comparable performance, without every prompt flowing back into a company's next model. I haven't made that switch yet, though I'm thinking hard about it. For law firms, where client confidentiality is the business, I suspect a lot of marketing leaders should be thinking about it too.
Mike Kaput and Mitch Joel both start with a person
Mike Kaput, chief content officer at SmarterX, gave the most informative session of the day for anyone with content marketing in their job description. He walked through the five-stage playbook his team now runs: original inputs, anchor assets, a content portfolio, living knowledge and an agentic engine.
What I love is where it starts. Stage one is the material AI can't produce on its own: expert interviews, customer conversations, first-party survey data. Mike has rebuilt his process around AI, but the value still comes from one person sitting down with another and finding out what they really think. AI scales it from there. When SmarterX ran its State of AI for Business survey, drafting the report went from four weeks to about one.
The later stages are just as useful. Once an anchor asset exists, his team mines it for external content and for internal material like sales decks and customer success briefs. Then everything goes into a library that people and AI can both search. They pull from a 246-episode podcast archive to build new courses and presentations. I would add one more use. Once all of it sits in one place, you can analyze it to see what your audience keeps asking about, which points to what you should offer next. For a law firm, that turns every client alert, webinar and attorney interview into raw material for training and pitches, instead of a page nobody revisits.
Mitch Joel reached the same place from the other direction. He described two "cities." In the intelligence city, AI personalizes what customers see and makes internal teams more capable. In the intimacy city, brands use AI to build close, personal relationships with each customer at scale. His warning was against "automating the life out of marketing." AI should remove friction. People still have to create the meaning. He pointed to Cisco rolling out agents to 90,000 employees, with a person setting the objective, supervising each step and staying accountable for the result.
Both want personalization at scale with a person still involved. Mike starts his process with a real conversation. Mitch ends his with one.
Chris Penn and Dan Slagen turned everyone into a builder
Chris Penn of Trust Insights ran the most practical session of the conference for me, a hands-on workshop. AI is bad at math because it predicts words, so he had us direct it to write and run Python code, which gives the same result every time. We took Google Search Console data on AI Overview citations and built an analytics dashboard from it, borrowing methods from economics and even agronomy along the way. One more prompt turned the whole analysis into a reusable skill, so next month's report starts from a data drop instead of a blank page. I enjoyed building it and plan to turn it into a skill of our own.
For me, the best session of the day was Dan Slagen's. He is chief marketing and AI transformation officer at Zapier. I had heard much of his story when Mike Kaput interviewed him on The Artificial Intelligence Show. It resonated even more in the room.
Early in 2026 he declared what he calls a red alert and ran a five-to-six-week buildathon. Every marketer spent three weeks learning a developer tool like Cursor, Claude Code or Codex, then built a working project. The team turned in more than 60. He reviewed every one, wrote handwritten feedback on each, and paired the beginners with the most advanced builders so everyone could see that the difference between them came down to hours spent with the tools. A designer built the event platform Zapier used to run its own conference. One marketer now manages more than 150 creators as a team of one.
Underneath it sits what he calls a marketing OS, a shared brain that pulls in market activity, company context and live data, then briefs the team each morning on what happened yesterday and what needs attention today. It can only read work done in the open, so Zapier grades its executives on how much of their Slack activity is public.
I was most curious about what comes after the buildathon. His second red alert is the 2027 org chart. For planning, he has thrown out the traditional split of demand gen, brand and product marketing, and he expects something flatter. He wants curious, flexible generalists. He is also working out how to pay marketers for the pipeline they generate and reward the people who build systems instead of asking for more headcount.
Law firm marketing departments will face the same questions, probably later and with more caution. At Zapier, one marketer and a system now run paid social. When a law firm's proposal work goes the same way, where does that person sit, and what happens to the rest of the team? I don't have a clean answer yet, but a law firm marketing team could run its own buildathon next quarter. It could also start putting more of its work in shared spaces now, with the general counsel helping decide what belongs there.
Paul Roetzer and Mike Kaput brought it back to earth
The conference closed with Paul Roetzer and Mike Kaput sitting down to give their view of where things are going. I really appreciated it. After a day that opened with Hao's warning and peaked with Zapier tearing up its org chart, it gave the whole conference a needed balance.
Paul expects enterprise adoption to be gradual. In his view, the near-term value of agents is in deterministic, rules-based work like recurring reports. Risk-averse organizations won't turn swarms of autonomous agents loose anytime soon. That is the principle behind Chris Penn's workshop, applied to the whole company. Mike described marketers becoming builders and systems managers, which is Zapier's story told across the industry.
Paul also said leaders have an ethical responsibility to reskill the people they already have and give them new roles rather than replace them. He named long-tenured staff, who know how the organization really runs, as strong candidates for AI operations jobs. On Day 2 he asked who learns the work when AI does it. On Day 3, part of his answer was to start with the people already in the building.
Their advice for all the uncertainty was plain: do the next thing.
I came away from Cleveland inspired and invigorated. I also left a little more careful about which models I use than when I arrived. I'm already looking forward to MAICON next year.