The website project that changed how I think about AI.

About three years ago, I started using AI in my day-to-day work as a Director of Information Security for a large international law firm. That environment has shaped the way I approach almost everything: a report should be clear enough to support a decision, a project should be controlled without becoming bureaucratic, and a process should work on an ordinary day, not only when the right person remembers every step.

AI gave me more range and speed. It did not give me those standards. That distinction became much clearer when I tried to help my wife's dental practice.

The problem was close to home

My wife, Dr. Nana Dickson, owns Allegra Dental Center. She had recently invested in a new website when a dental marketing consultant suggested the next set of improvements: create genuinely useful content around the questions prospective patients were searching for, build lead funnels for people who were not ready to schedule immediately, and connect those inquiries to a CRM-supported follow-up path.

The website itself was only one piece of the problem. The practice needed content, code, analytics, landing pages, and follow-up to work together. Its existing website provider did not have experience with the CRM work the practice wanted.

I offered to see what I could build.

I used Codex to work directly with the website repository, writing and revising content and code, then running the checks around it. Today, my broader operating model also uses ChatGPT Work for research, analysis, and deliverables. OpenAI describes Work and Codex as separate experiences, and I use each for the kind of work it handles best.

What mattered was how much more of the problem I could keep connected without repeatedly translating the practice's goals from one disconnected specialist to another.

What caught my attention

The first website version came together far faster than I expected, and the consultant's reaction got my attention. But speed alone would have been a poor reason to build a company. Fast generic work is still generic work.

The more interesting result was continuity. A conversation about the practice could inform a search topic. That topic could become a useful article, a clearer landing page, a measured inquiry path, and a follow-up workflow. When something changed, I could trace the consequences across the system instead of treating the website, advertising, content, and CRM as separate assignments.

That was the light-bulb moment for me. I began to see a better operating model for managing dental growth, with AI supporting the work instead of becoming the product the practice had to learn.

My first business model had missed the real need

The original version of Smarter Practice focused on the question of how dental offices could use ChatGPT while addressing HIPAA requirements. I spent about a year trying to make that model work. It was not widely adopted.

That experience was useful because it forced me to separate an interesting capability from a problem a practice owner already wants solved. The owners I want to serve care less about acquiring another AI workspace. They want the practice to be easier to find and choose. They want marketing work to connect. They want someone to notice what is not working, explain what matters, and take responsibility for the next step.

Allegra Dental Center gave me a practical place to apply that lesson. I moved AI behind the service, where it could expand what I could examine and execute without giving the practice another product to operate.

Experience changes what the tools produce

It is easy to describe AI as expertise on demand. I think that overstates what is happening.

An AI system can compare more material than I could review manually. It can draft alternatives, work across technical files, repeat checks, and help keep a complicated project coherent. It can also produce something polished, plausible, and wrong. It does not automatically know which business question matters, which claim is defensible, or when a technically valid change makes the customer experience worse.

My professional background contributes the part I do not want to outsource: judgment. Years spent in a high-standard environment taught me what a useful executive report looks like, how accountable projects are run, why evidence and conclusions must stay distinct, and why a workflow needs a clear owner. Those lessons transfer directly into AI-assisted work.

The work is AI-assisted. The judgment and accountability are mine.

A website change can start in ordinary language

A practice owner should not need to name a website component or write a technical specification. The request can arrive as a note or voice recording: “Make it clear that Google Ads management is included, but that ad spend is separate.” Before I use a recording or transcript, it must be appropriate for the work and free of patient information, credentials, and unrelated confidential material.

I give Codex the request together with the website repository and the standards the change must preserve. It can find the affected language, update the underlying content or code, run the relevant checks, and prepare a local candidate. Git shows exactly what changed. I then read the promise as a prospective client and inspect the page on desktop and mobile before deciding whether it is ready for review.

There is still an engineered website underneath this experience. The difference for the dentist is that a useful request can move directly from the practice's language to a visible candidate without passing through a chain of account, writing, design, and development handoffs. This example describes a local Smarter Practice candidate. It does not establish a production client result or guarantee a turnaround time.

Where I stay involved

I want Smarter Practice to use AI and automation extensively for routine content requests, technical changes, research, monitoring, and quality checks. That is how a lean company can bring substantial capability to a practice without asking the practice to coordinate a conventional collection of separate providers.

I also want the time that automation returns to go somewhere valuable: listening to clients, understanding what is changing inside the practice, preparing for strategic conversations, and improving the service itself.

That is what “founder-led, system-supported” means to me. A client should not have to explain the same situation to a salesperson, account manager, writer, developer, advertising specialist, and CRM contractor. I remain responsible for the relationship and the standard of the work. The system helps me coordinate the execution.

I am building this deliberately. Staying close to each client gives me the opportunity to listen, see where the model breaks down, and improve it before considering broader scale. The personal relationship is not something I intend to automate away.

This is an operating direction, not a claim that every capability is unattended or available in every engagement. The active scope, connected systems, approvals, and responsibilities still have to be clear for each client.

A practical test for any AI-assisted provider

A better question than whether a provider uses AI is whether AI sits inside a responsible operating model.

A practice owner should be able to ask who understands the business context, who checks the source material, who decides what should change, and who remains accountable when an automated step produces the wrong answer. If the only explanation is that the technology is powerful, the service model is incomplete.

Worksheet: compare the operating model

Use one copy for each provider you are considering. Record the provider's exact answer, ask for something you can inspect, such as a sample, workflow, report, or test, and note what remains unresolved. Compare the completed sheets before deciding.

Provider:

Business problem I need this provider to own:

1. Who will understand my practice and remain accountable?
Provider's answer:
Evidence or example supplied:
Unresolved concern:

2. What will AI or automation support, and what still requires judgment?
Provider's answer:
Evidence or example supplied:
Unresolved concern:

3. How are facts, claims and recommendations checked?
Provider's answer:
Evidence or example supplied:
Unresolved concern:

4. How do website, search, advertising, content, funnels and follow-up connect?
Provider's answer:
Evidence or example supplied:
Unresolved concern:

5. What happens when evidence is missing or a proposed change is not justified?
Provider's answer:
Evidence or example supplied:
Unresolved concern:

6. What will I review or approve before consequential work goes live?
Provider's answer:
Evidence or example supplied:
Unresolved concern:

Decision: continue / clarify / decline

Reason:

What Smarter Practice is becoming

Smarter Practice has evolved from the ChatGPT workspace idea I first imagined into Managed Dental Growth. The service starts with a premium website and connects it to search visibility, content, Google Ads when elected, lead funnels, CRM-supported follow-up, reviews, measurement, and continuing improvement.

AI makes that breadth practical inside a lean operating model. My job is to make sure the breadth does not become noise: keep the practice's real objective in view, connect the work, hold the standard, explain the limitations, and stay accountable for what happens next.

That is the part of this model I find most compelling. The technology can keep changing. The customer should still know who is responsible.