He Built His Own Jarvis on a $50 Raspberry Pi

Bryan Dennstedt is a Partner and Practice Area Leader at TechCXO, one of the largest fractional executive firms in North America: 120+ C-suite partners, $56M in annual fees, more than 5,000 clients. He works as a Fractional CTO, CIO, and increasingly as a Fractional Chief AI Officer (CAIO) for mid-market companies moving from scattered AI experimentation to something more deliberate.

Before TechCXO: Solution Architect at IBM, Software Developer at Xerox, founding CTO of MDLIVE, a telehealth platform later acquired by Cigna's Evernorth. He also hosts AI with Bry, a weekly podcast and newsletter aimed at executives who want to understand AI without the hype. Find him on LinkedIn.

Jarvis on a Raspberry Pi

About 25 minutes into the episode, Bryan mentioned almost as an aside that he'd built himself a Raspberry Pi assistant. He calls it Atlas.

It listens to his meetings. When a call ends, action items surface on a small touchscreen on his desk. He can tap to delegate them, push them to a follow-up queue, or fire them to his assistant. Before the next meeting, Atlas has already pulled the notes from the last one.

Every client, every conversation, goes into a knowledge base he can query whenever he needs to. "What did I say about this last week?" He asks. It answers. The system connects to his Telegram and WhatsApp and runs as a web app, with the database in the cloud and the Pi as a local interface and voice layer.

Getting there: he handed Claude the SSH key to the Pi and told it to figure it out. Claude worked through the Bluetooth drivers, configured the speaker, selected the voice, built the UI. Bryan nudged it along. There were months of failure before any of it worked, of course. But the end result is incredible.

How AI is Changing Work

Bryan runs 3 to 5 fractional engagements at a time and has seen companies at every stage of AI adoption. The pattern he sees is: the ones winning built the plumbing. The ones struggling gave everyone a login.

Replacing the SDLC, not just accelerating it: Bryan has built Software Factories at four or five companies — end-to-end agentic development pipelines where AI handles requirements intake, UX review, code generation, QA, and code review, with humans overseeing phase gates (rather than writing every line). The developers who go through it start skeptical, but the moment it clicks, he says, is when an agent creates and closes a Jira ticket on its own. After that, they start thinking differently about what they can ship.

"I have to tell the CEO I'm going to double their team's capacity without spending much more money. Secretly I'm thinking 10x, but I can't say that because they won't believe me."

Shadow AI is the new BYOD: 90%+ of employees are already using consumer AI tools. About 40% of companies have an official policy. Bryan's parallel is the BYOD crisis from the early smartphone era: blanket bans didn't work then and won't work now. What does work is giving people a governed environment they actually want to use, one that connects to the tools and context of their real work.

Jevon's Paradox and the job displacement debate: Bryan isn't worried about AI displacement in the way the headlines suggest. When radiology automation dropped the cost of a scan from $500 to $50, the number of radiologists went up. More capacity generates more demand. He thinks we're at the start of something similar: an explosion of ideas from founders and operators who were previously priced out of building.

"The financial barrier of building an idea is gone. We're going to see so many new companies, new careers, new possibilities."

Why giving AI to your CTO was probably a mistake: Most companies handed the AI problem to their CTO because they needed someone technical to own it. Bryan's observation is that building products and getting an entire organization to change how it works are different jobs, and a lot of technical leaders weren't hired for the second one. The companies getting traction are treating AI adoption as an organizational design problem instead of just a tooling problem.

What running 40 fractional engagements teaches you: Working across multiple companies simultaneously gives Bryan pattern-matching that in-house leaders rarely develop. He sees what's working, what's failing, and what the common mistakes look like across industries, sizes, and maturity levels. His 5 Stages of AI Maturity framework (Uncertain, Scramblers, Strategists, Advanced Implementers, Advisors) is the diagnostic he uses with every new client to figure out where to begin.

Feeling Six Months Ahead

What he's showing is impressive: a touchscreen on a desk with action items from a call that just ended; a nonprofit dashboard built in 30 minutes while the board sat waiting; a new website, 10x better than the old one, generated in 27 minutes while the CEO watched. But he put in the hours of failure first, figuring out how it all works.

The leverage and the compounding are real, and it starts with building the infrastructure (the context, the connections, the process) so the AI can do the work before you ask.