AI Policy

The short version

I use AI heavily to build. Your material, meaning files, documents, and correspondence, goes through AI providers so I can analyze and synthesize it. That is not a corner being cut. It is the reason you get in weeks what used to take months.

If you want something kept out of that, tell me which parts and I will work them by hand. It costs more and takes longer, and I will quote it honestly before I start.

AI in the build is not AI in your product

This is the distinction that matters most, and almost every concern people raise comes from the two being blurred together.

• In the build. Reading your existing material, research, planning, drafting, and writing code. This is where AI genuinely changes what is possible, and it is always happening.
• In your product. A feature that calls an AI model while your customers use it. Sometimes that is the right answer. Often it is not. It happens only when the job calls for it, and it is always spelled out in your scope of work before it is built.

A recent engagement is a fair illustration of the normal ratio. Roughly 1,500 of the client's past documents were analyzed to understand how their business actually writes. The system delivered to them makes two AI calls, each writing a single paragraph. Heavy AI in the build. Almost none in the product.

What happens to material you provide

Material you send me may be transmitted to third-party AI providers through their APIs so it can be processed during the engagement. Anthropic is the primary provider. There are two paths, and they are governed differently:

• Development work runs on an AI subscription with the model training setting turned off. That excludes the work from being used to train future models, and places the account on the shorter of the provider's two data retention tiers.
• Anything your delivered product calls in production runs on commercial API access, which under those providers' commercial terms is not used to train their models by default.

Your material is never sold, never published, and never used for any purpose other than performing your engagement.

Two honest limits. Providers reserve the right to review conversations their systems flag for safety, regardless of the training setting. And I can tell you which provider controls are enabled, but I cannot make guarantees on a provider's behalf beyond the controls they offer and the terms they publish.

Working copies of your material also live on my own machines for the length of the engagement, which is true of any consultant who has ever been sent a file. That copy is the one I actually control, and it is the one I will make commitments about.

What never goes to an AI provider

• Credentials, API keys, access tokens, and passwords. These operate your systems and are never transmitted.
• Payment details and financial account numbers.
• Government identifiers.

On a typical engagement, the personal information that does get processed is ordinary business contact data: names, addresses, email addresses, phone numbers. That is real personal data and it is handled accordingly. It is not the same category as the list above, and the two are not treated the same way.

Automated processing

Some of that processing is automatic. I run internal systems that keep my own project and client records current on a schedule, rather than only when I am sitting at the keyboard. They run on the same providers and the same settings described above, on material you have already sent me.

Keeping something out

You can designate specific material as excluded from AI processing. Identify it in writing before or at the time you provide it, and it will be handled manually.

Where exclusions materially change the work, I will quote the additional time and cost before proceeding. That is usually a significant difference. Reading fifteen hundred documents by hand is a genuinely different job from reading them in an afternoon.

There is also a middle option. Some work can run on a model hosted locally on hardware I control, where nothing is transmitted to a third party at all. It is slower and less capable than a frontier model, and far faster than doing it by hand. If your material is sensitive enough to warrant it, ask and we will scope it.

What you own when it is done

Worth understanding, because it is the part most people building with AI will not raise with you.

Under current United States law, material generated purely by AI cannot be copyrighted, because copyright requires human authorship. Prompting alone does not create authorship, however detailed the prompt. Human contribution does: substantial editing, deliberate compositional decisions, and combining generated output with original human work.

In practice this rarely affects a working system, where the protection that matters is your trademark, your contract, and your access to the code. It matters a great deal for brand assets like a logo or a wordmark, where being able to stop someone else from using it is the entire point of owning it.

AI material you bring me

Bring it. If you generated a layout, drafted copy, or built a rough version of what you want before hiring me, that is useful and I would rather see it than not.

It gets treated as a brief, not as a deliverable. It tells me what you are drawn to and what you are trying to say, which is often the hardest thing to get out of a conversation. What ships is built properly, to the standard the rest of the work is held to.

Two practical notes. Where AI-generated material is going to carry your brand long term, see the ownership section above, because it applies to what you made too. And if you want an existing AI-generated site or system taken over and finished properly, that is its own scope, quoted on its own, because repairing generated work is often more effort than building it once.

Changes to this policy

This page will change as the tools and the law change, and both are moving quickly. The date at the top reflects the current version. Where a change is significant, clients with active engagements will be told directly rather than left to notice.

Questions about any of this are welcome, including hard ones. If something here does not work for your situation, say so early. It is a much better conversation before a project starts than in the middle of one.