AI Agents and the Company OS at Builders Club

October 22, 2026

Thursday was a long day. I went straight from the Accelerator Centre to Builders Club, where I first interviewed Steve Hanov and Noah Deskin in a side room, and then slipped into the main room for the demos. I missed a couple of them while I was out in the hallway talking to people, so this covers the ones I saw.

The demos I did see came from people running real parts of their work with agents, and a lot of them don't write software for a living. They were also honest about where their setups break.

7thmeetup in the series
5demos
10 minfor each demo

How the night runs

AI Agents × Company OS is a monthly meetup that Artem Zhutov and Babajide Okusanya run on the third floor of Builders Club. Each presenter gets ten minutes to show agents doing real work on their own screen, followed by a couple of minutes of questions, and pitches aren't allowed.
Artem opened with a slide he told everyone to screenshot. It was a prompt from Claire Vo that has your AI go through your email to find every service you have signed up for, look through your connected apps for old keys and integrations you no longer use, check your code for vulnerable dependencies, and put everything in a spreadsheet with what is at risk and what to do about it.

Artem also showed his own setup. He has a separate agent for each part of his life, from video editing and content to health, goals and immigration, and one "chief" agent that hands out the work. His point was that his attention has become the limit, because the agents produce more than he can read. He is building an inbox where their work lands for him to review, the way a team reviews code before it ships.

A new hire that onboards itself

Bailey Chittle showed Quartzite, an open source knowledge base he built with Claude Code, where a person's notes and an agent's documents live together and link to each other. His talk itself was a Quartzite document.
Bailey then spoke to Claude out loud, in front of the room, and told it to onboard itself onto the company's documents. It read a Go codebase it had never seen and explained it back to him. Then it made a short video of the knowledge base as a night sky, with each document as a star and the links between documents drawing the constellations. As an encore, Bailey asked it for a goodbye animation for the audience, and it made one and thanked us for being a great room.

Bailey's four steps for trying this yourself are to put everything you know into one folder, open Claude Code and ask it to read everything and tell you what it is, give it a concrete job, and then review what it makes together with it. He said a folder of plain files works better than a big retrieval system, because the agent is capable of searching the files on its own.

One dashboard for every client's agents

Syed Ahmed deploys agents for companies, and his clients range from a $20 million business to a $7 billion real estate firm. Syed said nobody wants to open pull requests and nobody wants to review them, so his consultants change a client's agents (their memory, rules, prompts and skills) by uploading files to Payload, an open source CMS.
For his own work, Syed runs T3 Code across three machines, including an old Fedora computer that only his agents use, and he closes finished tasks from a sidebar until it is empty at the end of the day. When someone asked how he checks his agents' work, he said that anything with a number the agent can check itself, and that he checks a new feature or a change to the interface himself.

Seeing the path

Ali Shaverdi, a founder who describes himself as non-technical, is building Vizi. It draws the steps an agent is going to take as a map, the same way most people would rather look at Google Maps than listen to the directions. You leave comments on the map instead of writing another prompt, and Vizi shows which model will run each step and what it should cost.
Ali's live demo broke because a Google server in Toronto went down five minutes before he went on. He promised the room he wasn't pitching ("Jesse hates pitches"), walked us through what he could, and asked for beta users. Every live demo night has one of these moments, and he handled it better than I would have.

The agents ran the company

Bahubali Shete runs his company, TapInPass, with no employees, and his slides are online. One slide covered that same Thursday, a day he spent in meetings while the agents ran the business.

17support and sales threads answered
4sales meetings prepped and recapped
7kits shipped, tracking sent
150+customer meetings set up since May

The agents also tracked four supplier orders across Amazon, Alibaba and JLCPCB, and found two defects that customers reported, with fixes under way. Each task left a record he could check.

Go-to-market from a non-developer

Vivek Sai is building Fractional Manager. He has 18 years of experience in go-to-market and growth work, and he doesn't write code. He showed the scripts he added after his Claude sessions kept colliding with each other: one keeps parallel sessions in separate copies of the code, one stops to run tests before changes are merged, and one makes Claude ask what a page is for before it builds it, the way a product manager would. Without that last one, he said, if you ask Claude for a pricing page it gets very excited and builds whatever it imagines.
His go-to-market engine works down from a strategy to a 90, 30 and 14 day content calendar, with an agent for each kind of content and a board like Trello, called Huddle, where he and the agents meet. It finds the right people to contact on LinkedIn, but Vivek writes those messages himself, because he finds the automated ones annoying too. He wrote up how to build it step by step and shares his Claude Code setup on GitHub.

What the hosts are building

Babajide Okusanya, who co-hosts the series, is an AI consultant and runs Disposable by Default, which sets up two-week AI experiments for engineering, product, sales, marketing and operations teams. Artem runs a Claude Code and Obsidian lab, a program for setting up a team of agents on top of your own notes.

What I took away

Two of the five demos came from people who don't write code, and both of them had real systems running. That was the part that stayed with me. Bailey's advice about putting everything in one folder of plain files is the one I'm going to try first.

A couple of people asked me afterward for a plain-language explanation of what everyone means by an agent, a skill and a knowledge graph, so I'm planning to write one for the newsletter.

Thanks to Artem, Babajide and the presenters, and to Builders Club for the room and the pizza.