The Three Waves of AI: Tools, Teammates, and the Software that Remembers

It's a Tuesday morning.

Your service drive opens at seven.
By seven-fifteen, eleven cars are already in line.

At nine, a customer walks into the showroom.
She's been on your website for three weeks…
looking at the same SUV every Sunday night

Today… she finally came in.

At 9:20, she’s on a test drive.
At ten-thirty there's a deal on the desk.
At eleven the F&I manager is structuring it.

By noon there's an invoice in service for a different customer's brakes,
and a phone call coming in from a third customer asking about a trade.

This is a snapshot of your everyday interactions.

Your store has a hundred of these moments before lunch.
And in every one of them — every call, every walk-up, every deal, every RO – there's information being generated about your customers, your inventory, your people, and your market that, until very recently, the software running your store had no idea what to do with.

That's what we're here to talk about.
Just how big of a problem not knowing what to do with all that information has created.There are two waves of AI hitting our industry right now.

The first one you've already used — most of you have had something write you an email or summarize notes for you this week.

The second one is just arriving, and most haven't seen it yet.

I want to walk you through both — where each came from, where each lands inside your store, and what they're each still leaving on the table.

The first wave was generative AI.
The cleanest way to think about it is this: generative AI is the intern.

The intern writes you a draft.
Answers your question. Summarizes the document. Suggests three options.
The intern is bounded — you hand them a defined task, they hand back a defined output. You still do the work.

You've all used it. ChatGPT. Copilot in Office. Email summaries. Image generation. This is the wave that's already here.

At Reynolds, we bet early that this could do more than just write copy.

In 2024 we launched Spark AI as a unified data layer underneath the Retail Management System, and on top of it we delivered four products.

Conversation AI transcribes every call your store takes and surfaces the hot ones to your sales managers.

Engagement AI handles text follow-up with leads and pings staff when a customer is ready to move.

Prospect AI predicts which of your customers will buy, what they'll buy out of your inventory, and how they'll pay.

Merchandising AI writes the vehicle descriptions from the VIN, cleans the photo backgrounds, and places consistent overlays on every vehicle image across every channel.

These work. Dealers see value the day they turned them on. Our bet was correct — there's enormous productivity locked up in bounded tasks a model can do faster, and at least as well, as a person doing it for the four hundredth time that week.

Efficiency has increased dramatically and these tools have opened the door for the second wave – agentic AI.
And if generative AI is the intern, agentic AI is the employee.

The employee doesn't wait for a bounded task.
The employee reads the lead. Checks inventory. Sends the email. Logs the interaction. Schedules the callback.
The employee reasons across the whole store and decides what to do next.
The work gets done.
The generative steps get automatically sequenced together.

This is an architectural shift, not just a smarter model.

The industry had to stop bolting on and start building a proper home for AI inside of software systems.
We had to design and build Vector databases.
We had to create function calling.
And we needed to leverage the Model Context Protocol and build servers to make data accessible and usable.

Once all that was done, we could build out Rey AI, our agentic AI that doesn’t  just suggest — it acts.

And worth saying out loud: it's a spectrum, not a switch.
Some agents recommend and wait for you to approve.
Some take action with guardrails.
Some run the workflow end to end.

The further you go, the more autonomy — and the higher the stakes of getting it right.

You see this in our work with Rey. Rey reads across the Spark AI data layer, takes real action on your behalf inside the software, and starts to feel less like a tool and more like a teammate.

Rey is part your team.

Picture your Tuesday morning with this layer turned on.

Your service manager doesn't start the day by reading reports —
Rey hands them the three ROs that need attention and tells them why.

Your sales manager doesn't spot-check calls —
Rey already routed the ones that mattered.

Your GM doesn't reconstruct yesterday — Rey reconstructed it overnight.

Your controller doesn't pull deal-level analysis on Friday — Rey ran it Thursday and flagged what's worth a conversation.

Decisions get made earlier in the day, with better information, by fewer people. That's what the agentic wave actually feels like in your store. It's not science fiction.

It's a faster Tuesday.

This is where the industry is heading right now.

It's where most serious R&D dollars are going, ours included.

It's real and it's valuable.

Every vendor worth taking seriously is racing to get good at this, and we at Reynolds are proving we are the best at it.

But there’s still a ceiling here.

The challenge is that even with the best agentic software: it only evaluates when it’s called up.

When your customer comes in for a second visit the agent is looking at the same data they had the last time she was there and figuring out what to do next from scratch.

Agents can’t solve this the way they are built today. They are still taking action in static software.

But every interaction is unique, every customer is unique, and they are constantly evolving.

So the question becomes how can the software adapt to the dealership, to the salesperson, and to the unique interaction with each customer?

This challenge is where I’m spending my time. We're building a proprietary infrastructure, in our own data centers, where the records themselves carry agentic memory.

A platform that senses the business as it runs instead of waiting to be asked.

So: what does software look like when the system itself remembers?

I told you at the start there are two waves of AI hitting our industry. That's true. But there's a third one coming behind them, and it doesn't have an industry-wide name yet.

I call it cognitive software.

The way I think about this is a story every dealer in this room already knows.

Every store has had a thirty-year veteran on the floor.

The one who knows every customer's kids' names.
Who can tell from a service ticket who's about to trade.
Who walks past a deal at the desk and says one sentence that closes it.

And every store in this room has watched that person retire — and watched a new hire open the same customer's file the next morning and start from zero.

That gap — between what the veteran knew and what the file holds — is the gap cognitive software closes.

This is the shift.

Where AI lives in the data, not adjacent to the data.

For this to happen, it requires two large structural changes to how we think about software.

First: the records themselves need to have a memory.

A new salesperson on his first day opens the customer's file and walks onto the floor with every insight about them — because the record has been compounding those details across every interaction your store has had with her for six years.

This doesn’t just make Rey smarter; it makes your salesperson smarter too.

Underneath, this is a different architecture than anything that's come before.

Some people in tech are talking a lot about "institutional memory" in the next generation of AI.

We call it layered intelligence — Rey on the surface, the records carrying their own memory, and the substrate connecting them so they sense and update each other as the business runs.

There's a second shift underneath this one that takes it beyond just memory, and it's where we are working on the hard science.

Every enterprise system you've ever used — every DMS, every CRM, every ERP — has been event-based.

Something happens, a workflow fires, the system goes back to waiting.

Cognitive software is signal-based.

It's continuously sensing the state of every record and understanding where it is on it’s journey.

When a signal shifts —
a customer opens an email,
a vehicle's days-on-lot crosses a threshold,
a deal stalls at the desk — the related records update accordingly. The signal travels, and the system responds as a whole.

The absence of an event is a signal too.

The customer who always opens your emails and didn't open this one.

The vehicle that should have moved by now and hasn't.

The salesperson whose close rate dropped this week and nobody flagged it.

Event-based software can't see any of that, because nothing happened.

Signal-based software treats nothing happened as one of the loudest things the system can hear.

You can think about it as managing the exceptions, but these are exceptions you would have never seen.

Event-based software waits to be asked. Signal-based software notices. That's the difference.

Your software needs to start noticing.

Here’s what I want to leave you with:

First, the foundation stays solid.

The accounting is still the accounting.
The workflows you trust to run your store still run your store.

None of this is interesting if it makes the books wobble or the close take longer.

The cognitive layer makes what works fit better. It does not replace what works.

Second — and this is the imoprtant one to understand clearly — the fit that builds up inside the system from running your business serves your business. The patterns your store creates tune the system to your store. They don't tune it for the dealer down the street. The foundation is shared across every store on the platform. The fit is specific to yours.

Third, the fit deepens over time. This isn't a switch you flip. The platform arrives complete and capable on day one. The fit takes longer. A month in, the system is sharper. Six months in, sharper still. A year in, you'll feel it in places you didn't expect. Eventually, the gap between a cognitive platform that's been running your store and a flat-schema platform that ran the same store for the same period isn't a feature gap. It's a compounding gap. Their software resets every morning. Yours doesn't.

Everyone thinks about agents right now, whether they say it out loud or not, measuring value in headcount you can remove. That's the agentic pitch — count the seats you don't need. Cognitive software answers a different question. The fit that builds up in your system is your people getting better at their jobs. The new hire walking onto the floor with details that used to take twenty years. You don't capture that value once and bank it. You compound it, every day your store runs, because your people are the ones generating the signal the system learns from. Take them out of the loop and the loop stops.

Every interaction, every outcome, even every non-action feeds the cognitive software.  Not just the siloed, digital interactions, but the personal interactions.  Your people will feed the software, and in turn it makes them the best version of themselves. And with that your business becomes the best version of itself.

One more thing worth saying out loud. This isn't unique to our industry. It's how the next decade of serious enterprise software gets built — across every category, in every industry. We're in a particularly good seat to see it here at Reynolds, because the workflows and the data and the trust have been accumulating in this industry for a long time. But make no mistake, this wave is bigger than any we have seen before and it will almost certainly create a massive gap between the leaders and the followers.

So here's where this leaves us.

The first wave gave us features. It was the intern.

Dealerships are using these solutions and they’ve seen success.

They’ve created real, measurable value.

The second wave gave us agents.It is the employee.

It’s happening right now. It’s revolutionizing our industry and…

We are leading the charge with Rey AI

And the third wave – the one I’m most excited for,

Is where everything finally comes together.

It’s the cognitive, signal-based system that actively learns from the dealership it runs inside of.

This is what we’re building. 

Let’s revisit that Tuesday morning five years from now.

The service drive opens at seven.

The line is shorter— not because there are fewer customers, but because the system has been listening to your techs all week and half the vehicles pulling up this morning were flagged for an issue before they got on the schedule.

The parts came in yesterday.
The loaners are already staged.

The advisor's screen already shows what each vehicle needs, what your store's history says about how long the job takes on this model in this season, and which bay it should go to.

The morning rush isn't a rush anymore.

The customer walks in at nine.
The salesperson meets her, takes her out for a drive, sits down at a desk. The screens look the way you'd expect — clean, fast, intelligent and full of helpful agents and insights. But underneath those screens, the system has spent five years tightening around this store, these customers, this market, these people. The salesperson knows things about how to work with this customer that the system surfaced before he asked — things a thirty-year veteran would have known. The desk knows things about how this deal is likely to structure that no generic system could know. The service department knows things about the vehicle that no log file could hold.

Same morning. Same store. Same business you've been running for thirty years. A version of the software that, over time running your operation, fits it the way nothing in enterprise software has ever fit a business before because it KNOWS your business.

Think about the places in your life where you feel known.

Where they don't ask — they just know.

There's a reason you value that.

That's the difference between a transaction and a relationship.

That's what your store can be at scale when the software remembers.

That's what the next era of automotive software looks like.

And its being built by Reynolds, the same people who built the last one.