Welcome to another edition of the Car Dealership Guy Podcast Recap—a rundown of key lessons from top operators, founders, and execs shaping the future of auto retail.
Today’s guest is AJ McGowan, VP of Research and Development at Reynolds & Reynolds.
He breaks down why fragmented software platforms are quietly limiting what AI can see and do, uses a service drive scenario to show how a fully connected system could surface a high-value trade opportunity in real time, and explains why Reynolds is investing in its own hardware infrastructure to make that scale.
There are three distinct phases of AI in automotive, and most dealers are still in phase one.
Most of the industry's conversation about AI conflates three fundamentally different things, which makes it hard to know what to actually invest in or expect.
"When you think about what that phase really looked like, that was, you know, generative AI, right? It was, 'How do I summarize this email?' Or, 'How do I come up with a a great recipe?' Or, you know, any of these other things where it was pretty bounded task that you would just ask for."
The shift from generative to agentic is the shift from intern to employee, and also happens to be the shift from agentic to cognitive, which is the shift from employee to veteran, where the AI has accumulated enough institutional memory to make decisions with full context rather than just responding to requests.
Agentic AI is like having an employee, and less so a tool you prompt.
The distinction between asking AI to do something and having AI that proactively does things matters practically for how dealers should think about what they're buying.
"What we're fundamentally talking about is true companion software, right? I like to say sometimes generative is like an intern, and agentic is like having an employee, where it's a real companion that we are teaching how to use the other software that we use, right?"
Ray (Reynolds' agentic platform in development) is being built on the idea that an agent can goal-seek across the DMS, CRM, AutoVision, and other data sources to answer a dealer's business question and eventually take action on it.
Reynolds is investing in its own hardware for a reason.
Building data centers to run AI inference in-house is an unusual move for a software company, and the reasoning directly affects what dealers can expect from the platform.
"We need to control costs… so that we can continue to provide value to all of our customers at a great rate. So, that's that's kind of number one. Same thing around quality, right? We don't want to be beholden on somebody else that might go down."
The other reason is privacy, given that the most interesting AI use cases involve data a dealer would never want to hand to a third party, and owning the infrastructure is what makes keeping it in-house possible.
Cognitive software is what happens when AI is so embedded that removing it breaks the product
The jump from agentic to cognitive is a structural change in how the software itself is built.
"One kind of fun definition that I've heard about cognitive software is… if you remove the AI, then the software doesn't work anymore. That's the jump from agentic to cognitive, where it truly becomes not just a system that's built for agents to be able to manipulate it easily, but a system where AI is deeply embedded inside of every transaction, every interaction, and the software itself in a lot of ways melts with the AI to become truly a learning system."
He added that generative is dropping a slice of lemon in a drink, agentic is squeezing it as hard as you can, and cognitive is making lemonade: something net new.
Presented by:
1. Matador AI - Most dealerships are losing leads in the follow-up. Matador AI fixes that. It's not generic automation. It's dealership-trained AI, built on millions of real conversations and optimized around what converts. In a thin-margin business, every conversation matters. Book your demo today and get your first month free @ here.
2. Zurich - Zurich Advisor IQ is Zurich’s AI-driven training and coaching platform built to help F&I teams perform more consistently and sell more effectively — using real transaction data, not theory. By analyzing actual F&I transactions, Zurich Advisor IQ helps identify behaviors and trends influencing results, delivers actionable insights and roleplay scenarios, and gives dealership leaders visibility into performance across managers, stores and rooftops. Connect with your Zurich representative to request a demo and see how Zurich Advisor IQ can help turn F&I insight into stronger dealership performance. Discover more @ here.
3. Reynolds and Reynolds - Turn cars faster and increase profit with AutoVision, an end-to-end inventory management suite that optimizes every step of the used vehicle lifecycle. From acquisition to sale, AutoVision gives you a clear way to manage inventory. Visit AutoVision.com for more information.
Dealer workflows haven't changed meaningfully yet, but the tools underneath them really have.
McGowan says the discrete tools feeding into that workflow are producing outcomes that simply weren't possible before.
"For most dealers, that insight into what their BDC is doing is net new, and has huge changes to their behavior. Or, if you look at AutoVision and… being able to recommend on cars, we can actually run that both at appraisal time so it helps in their workflow when they're inspecting cars, but it also can run every single night and look at all of their inventory and say, ‘Above or below a threshold, here's where you're at relative to where Ray says that you should be.’"
The next phase is wiring those discrete tools together so a dealer can set a monthly goal and have Ray figure out how to pursue it across the entire system.
The service drive is the richest untapped AI opportunity in the dealership.
A customer pulling in for an oil change with two payments left on their lease, in a car that retails well and moves fast in the current market, is a solid acquisition opportunity, but only if every relevant data point is visible at once.
"The net out of, which would be, 'Hey, this is a car we probably want to try to acquire. Let's send her a text and say we may be able to give her a great offer and have her walk onto the sales floor to go talk to somebody.' That type of insight, while it seems, you know, straightforward, actually requires a lot of different pieces of data that come from different parts of the system."
AI can't act on what it can't hear, and that’s why fragmented data across disconnected systems is the single biggest barrier to getting this kind of insight in real time.
Consolidated platforms will have a compounding advantage over point solutions as AI matures.
The case for having everything under one roof has historically been about training simplicity and fewer logins. In an agentic world, the argument becomes structural.
"When you think about an agenic future, having all of your tools under one roof is going to become much more important than it was in the past."
With that, the opportunity cost of fragmentation will keep rising, because every new AI capability will require access to data from every relevant system, and every gap between those systems will be a ceiling on what the AI can actually do.
Until singularity, if a person can't do it, neither can the AI
A principle from a machine learning engineer McGowan worked with 15 years ago has held up through every iteration of the technology since, and it's a useful frame for calibrating expectations about what AI can and can't be trusted with.
"If a person can't do it, the AI can't do it. Like if you could not teach a person to do this, then the AI is not going to be able to do it."
In other words, wherever a skilled person reliably produces a good outcome, AI can eventually match it, and wherever people consistently fail, the AI will too.
The dealer's edge in a cognitive software world is still the people
Every wave of AI (generative, agentic, cognitive) ultimately hits a ceiling defined by context and data quality. The thing that keeps the loop improving on the other side of that ceiling is the humans feeding the system.
"The dealership is really about the people that work there, and in a cognitive software world, those people continue to feed the software in this virtuous loop that gets you better and better and better."
This actually means the people side of the business, invested in properly, compounds alongside the technology rather than competing with it.
AutoVision started as a data science problem, which is why it ultimately became something bigger.
Before there was an inventory management platform, there was a single question: can we get materially better at valuing cars?
"The thing that really distinguished us in the marketplace and the problem that we set out to solve was how do we get better at valuing cars? That was the fundamental before frankly we'd even decided we wanted to build an inventory management platform."
Avery (which learned to use AutoVision's tools the way a trained user would) came directly from that foundation, and he says Ray is the next expression of the same idea.
