Flip the Stack

Most dealerships are built upside down. The DMS is on top and the rider or boater is on the bottom. This is what happens when you turn one right side up, and how I did it inside a company outside our industry.

Every dealership is a stack

Four layers, every time. Software and tools. Processes. People. Customers.

The only question that matters: what is on top? Because whatever sits on top dictates everything underneath it.

The stack most dealerships actually run

Read it top to bottom. The DMS decides the process. The process decides the headcount. And the rider or boater gets whatever survives the tangle.

Flip it.

What the flip buys you

• Ownership replaces the web. Same headcount, but now every rider and boater knows exactly whose hands they are in. No more nine customers bouncing between five inboxes.
• The process straightens itself. When it starts from the customer instead of from the DMS, the windy path becomes a straight line.
• The DMS finally has a job. It stops dictating how you work and starts serving how you work.

Nice picture. Now the honest part: how do you actually build it? I did it inside a company outside our industry. Here is the playbook.

It started with one agent

I stepped into an experiential design and fabrication company in New Orleans, about a hundred people. An industry I had never worked in. Before touching anything, I built an agent to ingest and translate the operation into the way I see a business: the roles, the workflows, the handoffs, where value is created and where it leaks.

• Its charter says who it is, what it does, and where it stops: it drafts, it never sends, posts, or spends, and it never stores a judgment of a named person
• Every meeting, document, and interview goes through it, so the picture keeps sharpening
• That same agent grew into the orchestrator agent for the entire organization

The tool I used to understand the company becomes the tool that helps run it. That is the E, started on day one.

The agent told me what to pull

No composite to compare against. No standard chart of accounts. So I asked the agent which reports to pull from QuickBooks to build a composite from scratch. It produced the report list and the mapping.

• I interviewed the staff with a pocket recorder, with their permission. Transcripts went in verbatim and came out as facts by person and by workflow. Never judgments of people.
• Out came the roles, the processes people actually use, the tools, and the handoffs
• Then a swimlane of one project end to end, redrawn with the team until it matched reality. An opportunity map ranking where the money is. An open-questions log where every unknown stays TBD.

The agent drafts. I decide. In a dealership the same question goes against the DMS: which exports produce which gauge.

Then the customers wrote the plan

With the operation mapped, I used the same ingestion to start understanding customer lifetime value, and then went straight to the source: interviewing customers about their ideal experience.

• What does a great experience look like from your side of the counter?
• Where do we make you repeat yourself, wait, or wonder what is happening?
• Their answers became the spine of the process plan

This is the flip in action. The process gets designed from the customer down, not from the software up.

The software problem

Before I arrived, the company had vibe-coded its own software and hoped to run on it. All front end. Nobody could see how the data layer or the application layer were structured underneath.

The instinct was right. The owner knew the business needed its own software. It just needed a real foundation under the paint.

So we split the tech stack into three chunks

Then the people, starting with nerves

None of this works if the team is scared of it. So we run AI training on a weekly cadence: plain language, real workflows, no hype.

• Open to whoever shows up
• What AI is, how it works, and the vocabulary to talk about it
• Every session ends in an open forum: concerns, opinions, and the questions people have been sitting on

Training earns trust. Nobody builds with a tool they are scared of.

What the training covers

Session 1 is How AI actually works. Five stops.

The public version is at ownex.io/ai-training-session-1. Run it with your team as is.

The questions they actually ask

The open forum is the most important part of the session. Four questions come up every time.

• Can it take my job?
• What can it not do?
• What about the water and the power the data centers use?
• Is our data safe?

Everyone builds their own agent

Training earns trust. Building earns buy-in. Session 2 is Build your agent. Each person builds a personal agent whose charter, skills, knowledge, and memory are organized exactly like the orchestrator's.

• Working memory is a desk. It can only work with what is on the desk, so you learn to run the desk.
• Agents run a loop. The check step is QC. Guardrails are where the loop must stop.
• People build tools that fix their own daily annoyances first. That is where the excitement comes from.

Because every agent shares one structure, nothing anyone builds is a dead end.

Four parts. That is the whole agent.

The Ownex Agent Builder interviews you and writes them: the charter, the memory, a skills roadmap ranked by time saved, and a security review. Free at ownex.io/agentbuilder.

The first two weeks: one sharp skill

Done on build day looks like one agent with one skill. The next two weeks are about making that one skill sharp.

• Fix your own daily annoyance first. Not the company's biggest problem. Yours.
• Done looks like a skill that runs the same way every time

One sharp skill beats ten half-built ones. Sharp is what earns a promotion.

The promotion path

A personal skill that proves out does not stay personal. It gets promoted into the orchestrator so the whole company gets it.

• Incoming. The skill goes into the orchestrator's skills/incoming folder and waits for review.
• Review. It is checked against a written governance file: who reads, who writes, how skills enter.
• Nest. It gets re-parented under the orchestrator and retested. Work travels, personal memory stays.

One shop, not a hundred sheds. One person's shortcut becomes everyone's capability. That is the B feeding the E.

The orchestrator moves to a server

The translator agent from day one, months of context by now, came off my laptop and onto a cloud server everyone can reach.

• A governance file sets the access tiers, who can write, and how skills enter
• A leadership tier is withheld from the general team
• Credentials live in a secret store, never in the files. A janitor routine keeps memory tidy.
• Sub-agents built for specific jobs nest under it and inherit its rules

Day one's translator is now the company's orchestrator. It was never a separate project.

It was never theoretical

Two things already running inside the company.

• One team member built a tool that generates a work-back schedule from the account questionnaire. Built in a day. Then came back with six more ideas.
• A quoting skill trained on the company's own past decks and invoices drafts a first-pass itemized quote, then grades itself against what the job actually invoiced. A person approves every line.

The stack gains an agentic layer

The agentic layer works with the employees on one side and the tech stack on the other. The process splits clean around it, and the customer row grows without the employee row growing. More riders and boaters. Same payroll.

And it compounds

• Every skill an employee builds makes the whole company faster, not just one desk
• The orchestrator keeps learning the business, so the agentic layer gets smarter every week
• Headcount stops being the ceiling on growth. The stack scales, the payroll does not.

That is the whole play. Flip the stack so the rider or boater is on top, then build A through E underneath them.

In a dealership

• DMS exports instead of QuickBooks
• Service department swimlane, write-up to delivery
• RO audit skill
• Weekly scoreboard from DMS exports
• Customer lifecycle cadence

The public example is already live on ownex.io. A labor quoter trained on hundreds of thousands of invoices. It answers labor time questions and keeps learning.

Three free tools. Pick one.

When you build something cool, tell me.