Building with AI: Beyond the First Draft

THE STORY BEHIND THE FIRST DRAFT

We had to revise the story of Insights before we could tell it.

Our own Insights section is a small example. It began as a handful of short takeaways. Those still matter: a busy person should be able to get one useful idea in a couple of minutes. But they could not carry the whole story of how the work gets done, or why a good website is never simply generated and left alone.

So we reshaped the section: Blog became Insights; dates, images, topics, and filters made it easier to browse; the writing gained room for fuller stories; and we kept the quick reads in the mix. That sequence is the point. Even a site about useful ideas needs a purpose, structure, voice, and details that hold together. The first pass helped us see what was missing.

Who is Oli?

Oli, or Oliver, is my AI collaborator: a customized ChatGPT agent that has worked with me since 2024. I have shaped how he works through instructions, examples, brand standards, and feedback. He can help organize research, explore options, draft, and refine copy, prototype layouts, support implementation, and spot issues during review.

He is not the person responsible for the work. I bring the client context, experience, strategy, and creative direction. I decide what problem we are solving, test whether a recommendation fits the evidence, challenge the parts that feel generic, and approve what goes live. With collaborators and clients, those decisions stay with the people who understand the stakes. Oli helps us move through possibilities faster; he does not own the outcome.

The version history tells a more honest story.

A snapshot taken just before this article update contained 418 saved versions across six live site projects in our portfolio. Pixs.ca accounts for 209. TaraODoherty.com has 109. The CXPR has 52, Science of Next 30, Voxie 10, and Christine Fitness 8.

That number is a record of saved site versions, not a count of prompts, hours, or individual design decisions. The first saved version is included for each project, and a later version can bundle many changes. Still, it makes one thing visible: the work has continued well beyond a first draft.

Counts reflect each project’s latest published Site version immediately before this article update. Tara was at version 109; this update is the next Tara release. A version is a checkpoint, not a one-to-one measure of revisions. Separate beta and private preview projects are excluded.

What changed across those versions?

There is no single “make it better” button. Iteration means looking at a specific problem, deciding what to adjust, then checking the result in context. Across our work, the changes have included:

• Positioning and audience: clarifying who a site serves, what it offers, and what a visitor should understand first.
• Structure and journeys: changing page order, navigation, and calls to action so people can find what they came for. On this portfolio, we brought the breadth of client work forward before case studies.
• Copy and proof: tightening language, checking claims against real experience, and making the next step clear.
• Design and usability: refining layouts, image choices, contrast, responsive behavior, and small interaction details for different screens and readers.
• Search and launch readiness: improving page titles, descriptions, crawlable links, structured data, sitemap coverage, and the details that help both people and search engines understand a page.
• Quality checks: revisiting pages after changes, catching inconsistencies and checking that a polish pass did not break something already working.

Our client work spans very different needs: SickKids Foundation and other nonprofit journeys, telecom, and digital services, financial services, healthcare, and diagnostics, and consumer products. A donation journey, a wealth experience, and a diagnostic workflow should not be pushed through the same template. Context changes the right answer. The work behind SickKids, AT&T, Alcon and Insurance Institute of Canada starts by understanding that difference.

Why the follow-through matters

A polished first pass can still be wrong for the audience, too vague about the offer, difficult to use on a phone, or disconnected from what the organization can actually deliver. If nobody reviews it, those gaps do not disappear. They become the experience visitors have.

Iteration is how intent meets reality. A team can bring in customer evidence, notice where people hesitate, see what a stakeholder needs to explain, or catch how a page behaves at a smaller screen size. Then someone has to weigh the trade-offs. That is where experienced judgment earns its place: it helps teams spend their effort on changes that solve the real problem instead of endlessly polishing whatever AI produced first.

A practical way to use AI

A useful starting point is one page, one audience, and one decision the page should support. AI can help explore options; the chosen direction still needs to fit the task, brand, and facts. Check the page on a phone, verify its links and claims, and see whether a real person can use it.

Then keep a record of what changed and why. The next revision should have a better starting point. If you need a partner to help make those calls, see how I work or explore what’s possible together.

Prefer a short read? Give AI a clear job. Keep people in charge.