|
Almost Timely News: 🗞️ The Road to MAICON 2026 (2026-10-11)
How I’m using my AI agents to get ready for MAICON 2026
|
| |||||||||||||||
Almost Timely News: 🗞️ The Road to MAICON 2026 (2026-10-11) :: View in Browser
 |
The Big Plug
✍️ Enroll in my new course, AI for Writers and learn how to make AI write better.
My Unofficial LinkedIn Algorithm Guide is updated for October 2026!
👉 Get the paid version here, USD 20, with one-pagers and AI Skills.
Content Authenticity Statement
100% of this week’s newsletter was made by me, the human. Learn why this kind of disclosure is a good idea and might be required for anyone doing business in any capacity with the EU in the near future.
Watch This Newsletter On YouTube 📺
Click here for the video 📺 version of this newsletter on YouTube »
Click here for an MP3 audio 🎧 only version »
What’s On My Mind: The Road to MAICON 2026
This week, I’m packing up to head to Cleveland for MAICON, the Marketing AI Conference, now in year 7. I have a special fondness for this event; long before generative AI, Paul Roetzer and team had built this community of AI-forward marketers and technologists to talk about the future of marketing with AI. I have the privilege of being the very first MAICON keynote, back in 2019, and that talk is still available though very, very outdated now.
It’s funny because looking back, all the classical AI things I talked about - forecasting, predictive analytics, natural language processing - are hot once again because generative AI gives non-experts access to these methods. You can talk to Claude Code or ChatGPT Codex or what have you and have it create any of those methodologies as software that you can run without having to pay anything for it.
But today, let’s talk about the road to MAICON and what I’m doing with my various tools to get ready for the event.
Part 1: Before I Leave
MAICON is also unusual for me because I like to drive there. it’s about a 600 mile / 1000 km drive, a nice straight shot on I-90 from Boston to Cleveland with plenty of stops and scenery along the way. One year, I had a friend from the UK visiting and we made a stop at Niagara Falls. Without traffic or stops, it’s about 9.5 hours of driving, easily doable within a day. Most years, it’s about 11 hours with stops for things like meals, tourist attractions, etc.
This year, I’ve built a few pieces of support for me along the way so I can think and work a bit as I drive. The first is what I call slopcasts. These are AI-generated podcasts, read aloud by a TTS (text to speech) system based on academic papers I’ve been meaning to read but haven’t gotten around to. Google’s Gemini Notebooks (formerly NotebookLM) do this very capably, but with a fair number of restrictions and a methodology that isn’t terribly flexible. You also can’t do any customization of the voices or of their artificially created speech disfluencies, which can be tough to listen to after a while.
For myself, I’ve built a skill for my Hermes Agent that assembles these for me using lots of deterministic tools. I’ll tell my agent what papers I want it to retrieve, like the gold medal papers from an academic conference, then ingest those papers, draft a speaking script from the paper that summarizes the key points in my preferred dialect, and feeds the speaking script to my TTS, Breeze2-TTS, a local TTS that runs on my Mac and generates good quality spoken word audio for free. Once that’s done, it puts them in a place any audio player can reach, like a Dropbox folder, and then I can listen to the papers summarized my way.
What makes this work for me is the integration with my agent. I build the skill in a frontier model like Claude Code, then load the skill onto the agent so it doesn’t have to try to build it itself. Agents like Hermes, when plugged into free, local models like Qwen3.6 35B are capable of doing, but not always great at thinking or coding. If I build the skills first in the best models, then the local models only have to execute the task, not build it themselves.
And because it’s under my control, I can tune the speaking script exactly the way I want without having to fight the built-in system prompts like in Gemini Notebook.
The second setup within my agent is a connection to the MAICON conference app. The SmarterX team very sensibly created a conference app this year that’s a web app, which means any agent that can use a browser can access the app and interact on our behalf. My agent has instructions to check the app once an hour and notify me if there’s anything worth me responding to, and I message it right from Discord to do that.
As I’m out and about for the conference, I’ll be able to have the agent keep looking on my behalf, so I don’t have to constantly be on my phone. Discord can message me when there’s something I need to pay attention to.
My third and final skill I built just for the road to MAICON is a heavily revised version of Andrej Karpathy’s LLM-wiki skill. LLM-wiki, as he designed it, is a simple markdown-based wiki that an agent uses as a journal of sorts. That’s a fine start, but I want it to do much more. I’ve reconfigured the stock LLM-wiki skill built into Hermes Agent into a true multimedia journaling app.
From Discord or the communications app of my choice, I can load video, audio, or images to the Hermes channel. Behind the scenes, it fires some Python scripts and local AI to convert and transcribe video and audio, storing that in the wiki. For images, it uses Qwen3.6’s vision capabilities to digest photos and images. Then inside the wiki itself, I’ve installed the SilverBullet server along with QMD (query markdown) that provides lexical and semantic search.
Other things I’ve bought or prepped for this trip include a portable, battery-powered espresso maker (no more conference/hotel coffee!), plenty of power banks and power strips, AC adapters for the car to power a heating pad for my back (yay being in your 50s), and my usual bag of AV tricks. On site, Katie Robbert and I will be doing a live version of the Trust Insights podcast, In-Ear Insights, so I’ve got my own camera and mics for that as backups for the main show ones.
I’m also filling my external hard drive with piles of local models. I’ve promised a few friends like Andy Crestodina, Becca Bultsma, Chad Treadway, and Angie Bailey that we’ll take some time to do an informal mini-workshop on setting up an agent like Hermes Agent that can have the same functionality as an agent like Meta Muse, but much safer and more private. Some of the models I’m packing include Qwen3 Embedding, EmbeddingGemma2, Qwen3.6 35B-A3B, Gemma 4 26B-A4B, Gemma 4 E4B, and others - the handful of models I know work well with agents and can run on robust hardware.