How we use different AI models to help us build projects.

We work with five AI models, and each helps with a different part of our work: building, reviewing plans, checking code, project discovery and artwork. Here is what each one does, why we chose it and how its work is checked.

This is how we use different AI models to help us build projects, as of October 2026. We work with five of them at Whitelam Media, and each one helps with a different part of our work. We're proud of our herd, so we made a short film about it. In the film, each model is one of our AI friends, an animal matched to the task it's naturally good at.

0:49

Claude Opus 5.5 builds our websites and runs our growth agents.

In the film it's the beaver, nature's engineer. Claude Opus 5.5, from Anthropic, writes most of the code on the websites we build. It also runs the scheduled agents that grow our own site: a weekly search review, a content writer, an outreach drafter and a quarterly strategist.

Why this model. Building a website is long work. The model has to keep the code, the plan and last week's report in view, and keep going for a long stretch without losing the thread.

How it's checked. It never reviews its own work. Every code change runs through automated checks and a review by a model from another company. Anything that matters, such as deleting a page or sending an email, waits for our approval.

Claude Fable 5.1 reviews our plans before we build.

In the film it's the elephant, the one that sees the big picture. Before we build anything that handles payments, sign-in or client data, Claude Fable 5.1, from Anthropic, reads the plan against the real code and looks for what would fail or be hard to undo. It reviews new page designs the same way.

Why a separate reviewer. A plan is cheapest to fix before any code exists, and a reviewer that didn't write it sees what the writer took for granted.

How it's checked. We go through every finding and accept, change or reject it before the build starts.

GPT-6.1 Sol checks every change to our code.

In the film it's the owl, up all night checking every line. GPT-6.1 Sol, from OpenAI, reviews every proposed change to our code against our written standard. Every night it also sweeps whatever changed since the night before.

Why this model. Claude writes most of our code, and a reviewer from the same family tends to share its blind spots. A reviewer from another company brings independence.

How it's checked. A reviewer can be wrong too. Each finding is checked against our written rules, then fixed or answered with evidence. Our guide to who checks the code when AI writes it has every step.

Claude Haiku helps us discover new projects.

In the film it's the fox, the explorer of the herd. Claude Haiku, from Anthropic, helps with project discovery. Every night it looks for requests for proposals (RFPs) we could help with, and every week it identifies businesses that might need help with a new website. It also powers our Free Search Check, which shows whether AI names your business when buyers ask.

Why this model. Discovery runs many web searches. Haiku is quick and inexpensive, and a bigger model would cost more without finding more.

How it's checked. Everything it finds goes on a list we read, and no one is contacted without our approval.

GPT Image helps us with artwork and images.

In the film it's the peacock, the artist of the herd. GPT Image, from OpenAI, helps us with artwork and images. We've used it, for example, to sharpen a firm's own project photos, turning small, soft images into clean, larger ones.

Why this model. It understands what's in an image, so its work looks natural, without the smeared look of older enlarging tools.

How we use it. We run it by hand. When it works on a firm's own photos, it's told to keep everything in the picture and add or remove nothing.

Which model does what, and how it's checked.

Different animals for different tasks.

Our herd works well because each animal does what it's good at. We choose the model for each task and write the rules it works to. When a task needs a different model, we change it: in October 2026 we moved all our code reviews to GPT-6.1 Sol.

Nothing ships without a check, and every change to our code is reviewed by a model that didn't write it. For client content, AI tools help with first drafts, and we edit and check every piece before you approve it.

The film ends with our AI friends, and with a promise: more animals are coming to help us build more great things.

The rules every model follows.

Every model gets things wrong sometimes, so each one's work is reviewed, and a few rules hold for all of them:

• Messages. No message to a client or prospect goes out without our approval.
• Live data. A change to a live database needs our go-ahead.
• Prices and deletions. A new price, a deleted page or a redirect waits for our approval.
• Volume. AI makes it easy to turn out pages in bulk, which Google's spam policies call scaled content abuse. We publish at a steady pace, and every page carries something only we have.

On our own site, the growth agents publish small changes themselves once the checks pass, such as a new search title. We read the weekly report, and every change can be undone.

Questions about how we use AI.

Do you write with AI? Yes, for first drafts. We edit every line and check every claim, and you approve each piece before it goes live.

Do Anthropic or OpenAI endorse Whitelam Media? No. We pay to use their models like any other customer.

Is our website tied to these models? No. The code is ordinary Next.js in a repository you own, so any developer can work on it.

See how we build on Next.js and Payload, or book a call and ask which model does what on your project.