What a CPBO Learned Building a Personal Fleet of AI Agents

Michael Affronti is the Chief Product and Business Officer at DriveCentric, the number one CRM and engagement platform for automotive dealerships, and one of the most hands-on product leaders building and using AI agents today. He spent three and a half months earlier this year back in the code, learning how agents actually work from the bottom up, memory, state management, harnesses, before returning to the executive seat. In this episode of Product Talk, Sonjoy Ganguly sits down with Michael to unpack how he runs a personal fleet of AI agents (named Jarvis at home, Q at work), how his AI Labs team ships two production agents a month, and what product leadership actually looks like when the team is a mix of people and agents.

Join us for new conversations with leading product executives every week. Roll through the highlights of this week’s podcast below, then head on over to our Events page to see which product leaders will be joining us next week.

Show Notes:

1. Michael’s career has been defined by a consistent pattern: finding new markets where technology hasn’t yet reshaped consumer or business behavior, and then building inside them. From Microsoft Outlook to Salesforce Payments to Bumble’s trust and safety operation to DriveCentric’s automotive CRM, each move was deliberate exposure to a new industry and a new set of problems.
2. At Bumble, the most surprising and instructive part of Michael’s role turned out to be trust and safety, not the consumer AI work he expected to lead. Spending a year thinking about how to protect members in a two-sided marketplace at scale gave him a form of systems thinking about safety and risk that he now applies directly to how he builds and deploys AI agents at DriveCentric.
3. The three and a half months Michael took off earlier this year to watch YouTube videos about AI and get back into coding is, in his view, one of the most valuable professional investments he has made. He could not have built or managed agents well at the executive level without first understanding from the bottom up how harnesses work, how memory and state management function, and where agents fail.
4. Michael runs two personal agent fleets. At home, Jarvis lives on a Mac Mini and manages over a dozen sub-agents handling meal planning, Instacart shopping, travel logistics, packing lists, and weather prep. At work, Q serves as his agentic chief of staff, running competitive intelligence, morning briefs, customer meeting analysis, conference talk prep, and deck creation. Every time he encounters a repetitive loop in his life, he builds another agent.
5. Building agents in his personal life, where they have access to deeply private information, forced Michael to do the serious work on security architecture and memory management that most enterprise builders skip. That experience now directly informs how DriveCentric builds and pitches the agents it sells to automotive dealerships.
6. The most important thing AI does for product managers, in Michael’s view, is shorten the distance between a problem and a decision. Instead of spending days explaining an idea across meetings, he now drops a prototype and a voice note into Slack and gets feedback the same day. That compression of the feedback loop is what makes everything else possible.
7. At DriveCentric, the team is now building a prototype a day in certain areas and sharing them directly with connected customers over Slack to get real feedback before anything goes to development. The speed from customer conversation to product change has collapsed to the same day in some cases, which Michael describes as profoundly exciting but also an open question about whether it scales across the full company.
8. DriveCentric ran what Michael calls the SDLC next-gen project: a month of internal research followed by intensive analysis, mapping every step of how an idea moves from anyone’s brain through to a shipped product. The team documented the entire pipeline, then systematically identified where time was being spent on low-judgment, non-essential work that could be automated or cut.
9. One of the key outputs of the SDLC project was counterintuitive: product briefs need to get better and more thorough, not shorter. They need to answer what is being built, why, who it is for, and what success looks like clearly enough that anyone who reads it months later, including engineers and marketing, can orient themselves without asking. The rest of the process can be compressed and automated; the front-end thinking cannot.
10. AI is shifting product management from artifact production to judgment, craft, customer empathy, and strategic thinking. The question is no longer whether you can write a 60-page requirements document, it is whether you are building the right thing, getting the right customer feedback, and deploying your product people into conversations rather than documentation cycles.
11. Engineers at DriveCentric are also getting time back, and Michael is deliberately redirecting that time toward customer conversations and upstream brainstorming sessions. Engineers now join customer meetings and take features across the line as product-engineer hybrids. The blurring of the line between product and engineering is not something to manage around, it is something to accelerate.
12. DriveCentric created AI Labs, a small internal team of single-person full-stack product engineers who own everything from the brief through the customer conversation through the code through the launch. Each person operates like a one-person company. The team is currently shipping two production agents a month, with many more in pilot, and the decision-making speed is unlike anything Michael has experienced before.
13. AI Labs works because the path to decisioning is fast. A team member comes out of a customer meeting, makes edits to a live AI product, sends it back to the customer that day, and gets feedback that day. Whether that model can scale across a larger organization is an open and active question, but within its current scope it is producing real results.
14. DriveCentric has built AI adoption into its OKR structure at the product organization level. Each quarter, specific agent-building projects are assigned to product people as deliverables, and completed agents go into a shared harness available across the company or department. Competitive intelligence agents, customer meeting scrapers, bug surfacing tools, all built by people on the team and deployed to everyone.
15. One of Michael’s clearest observations from advising other companies is that being too restrictive with AI access is the biggest mistake software companies are making right now. There is no golden playbook. The only way to find what works is to experiment, and excessive caution about security policy and access is the main thing preventing teams from learning fast enough.
16. The one-on-one has changed. Michael still holds them, but far less frequently, and almost all project management has been removed from them. The freed-up time goes into conversations about the person’s development and career. The status updates can be automated or pulled from AI tools. The human conversation should be about the human.
17. Michael’s thesis on the future of product leadership is that the best leaders a few years from now will manage a mixture of people and long-running horizon agents that do work in parallel with the human team. That transition is already beginning in his organization, and he expects it to become the norm rather than the exception.
18. What will not change as AI takes over more execution is the value of judgment at the strategy level. Portfolio management, what you are building, how initiatives move from incubation through execution through launch through scale, what the frameworks are for making those calls, is the skill Michael believes product leaders need to be actively doubling down on right now, precisely because everything around it is about to change.
19. Michael’s background in his family’s Cadillac dealerships in Queens is part of why DriveCentric appealed to him. Automotive is an industry built on personal, relationship-based selling, and his thesis is that AI should amplify that relationship rather than fundamentally disturb it. Technology as an amplifier of human connection, not a replacement for it, is the guiding principle behind how DriveCentric thinks about its product.
20. The most important question a product person can answer still has nothing to do with AI: what are we building, why are we building it, who is it for, and what does success look like? No amount of automation changes the quality of thinking that has to go into that document. The rest of the pipeline is compressible. That is not.