Personal AI and Always-On Agents at Builders Club

October 16, 2026

The day after Techtoberfest ended at the Tannery, I went to a smaller one at Builders Club on King Street in Kitchener. Personal AI & Always-On Agents was run by Waterloo Agentic, the local chapter of Canada Agentic, and it was one of the most useful evenings I've spent at a tech event this year.

25people on the third floor
12questions sent in before the night
2fireside chats
$11a ticket

Melissa Cutler hosted it, in a dirndl. She heard Communitech was running Techtoberfest, loved the name, and made her own version for the day after.

The format is what made the night. About 12 people sent questions when they signed up, and Melissa built the evening around them. She read each question, asked the person to explain what they were really after, and passed the mic to whoever in the room had an answer. Everyone got to talk, and nobody had to sit through a pitch.

5:30 pm
Arrive and connect, with pretzels from Beertown and beer from Block 3 Brewing
6:40 pm
Fireside one: trust and oversight, and finishing builds
7:20 pm
Break
7:50 pm
Fireside two: business and everyday workflows
8:30 pm
Takeaways and round tables

Trusting an agent

The first question was mine. I asked which tasks we should and should not give to always-on agents. I write a lot of code with Claude Code, and it makes things up.

Steve Hanov has worked at Corel, Microsoft and BlackBerry, and is building an AI assistant for real estate agents with his co-founder Noah Deskin. His answer was that you can't fully trust the models because of how they are trained: people pick the answer they agree with, so the models that survive are the agreeable ones.

“
You are smarter than the smartest model.

He no longer reads every line the AI writes. He has it write ten plan files in a row, reads the plans, corrects them, and lets it work while he's away.

Always-on agents

Colin Knapp keeps each always-on agent to one job. He said an agent's heartbeat is a cron job: it runs on a schedule. For anything that really matters, he asks the model one clean question at a time and takes a one-word answer, so ordinary code decides what happens next.

Business and everyday work

The second half was where the room really opened up. Bilal Shaukat runs a door-to-door fundraising agency that has been around since 1999, and had been using AI agents for two weeks. His old boss spent two hours a day cutting turf maps out of Google Maps screenshots. Bilal had AI read census data and charities' financial statements to find the best postal codes, then count roofs to cut the turfs. It was the best example of the night.

120-150doors in each turf
30 minfor a month of maps
2 hrsa day, the old way

Melissa shared her own system too. She has recorded her calls for two and a half years, and that became the context for a ChatGPT project she uses to "duplicate myself". One of its commands is "what should be remembered?", which saves her corrections and her best examples to Google Drive.

Vera Cui, who leads Google Developer Group Markham, drove an hour and a half to invite everyone to GDG Markham's first hackathon on October 24 in Thornhill.

Takeaways

Steve said this feels like the early web, and that everyone in the room is early. Arun Kumar talked about building a "minimum learnable experience" instead of a minimum viable product: the smallest thing you can put in someone's hands to learn what you need.

I went because I didn't know much about always-on agents. I came home with a plan: an agent that is triggered by something, does some analysis and emails me about it. I'll be trying that for weeks. I also saw systems like the ones I build on my own used for sales and marketing, which I never would have thought of.

Thank you to Melissa and Waterloo Agentic for a great night, to Joe Jatou for keeping the mic moving as MC, to Builders Club for the space, and to Beertown and Block 3 Brewing for the pretzels and beer.