Designing this site with Claude instead of Codex Mixed
Ran out of Codex credits mid-project and switched. First pass of this page came from a single brief. Judging on how far the second and third rounds get.
Building an agency with AI, in public
I run a marketing agency. I'm rebuilding how it works with AI, and showing the work.
Red Hills Lab serves gyms, markets, talk series and other businesses with a physical space. Every project here is a real brief, what AI did, what it didn't, and the tools I used. Take the playbook, or hire me to run it for you.
$ show projects where ai was used to…
Red Hills Lab website Live Rebuilt the agency site on Webflow with AI doing the first pass on structure, copy and CMS content. Human judgement on positioning and final copy. Brief · Outcome · Stack · Limits ↓
Problem. The old site described services generically. It didn't say who the agency is for (businesses with a physical space and an experiential offer) or show the work.
Goal. A site that qualifies leads before the first call, built in days rather than a two-month agency project, and cheap enough to iterate every month.
- Client
- Red Hills Lab (own agency)
- Timeline
- Placeholder: 2 weeks, evenings
- Role
- Strategy, prompts, final edit, Webflow build
- Sitemap, page structure and first-draft copy generated from a one-page positioning brief, then edited by hand.
- Case-study pages set up as a CMS collection so new work is a form fill, not a rebuild.
- Placeholder metric: enquiries mention a specific service ~2× more often than before, because the site names them.
What I'd tell a client: AI got the site to 70% in an afternoon. The last 30% (voice, what to leave out, what the offer actually is) was still the job.
Positioning brief→Claude: sitemap + copy→Codex: Webflow custom code→Webflow CMS
The one technical piece worth showing: the prompt that produced usable copy asked for constraints, not adjectives.
# positioning brief → page copy Audience: owners of gyms, markets, event series (UK) Tone: plain, confident, no agency jargon Rule: every service line names a real deliverable and a timeframe Output: H1, 2-line intro, 6 service blocks (title + 25 words each)
Generated copy defaulted to hype ("transform", "elevate"). Needed a banned-words list.
Webflow layout still had to be built by hand. AI was useful for custom code embeds and CMS field planning, not for the visual design.
Positioning decisions can't be outsourced. The model will confidently write a site for the wrong audience if you let it.
LX3 Shopify store In progress Building a Shopify store for LX3. AI is drafting product descriptions from a spec sheet, generating the theme customisations, and structuring collections. Brief · Outcome · Stack · Limits ↓
Problem. Placeholder: a product range with a spreadsheet of specs and no web copy, images or store structure.
Goal. Launch-ready store: collections, product pages with consistent copy, and a theme that matches the brand without a custom build budget.
- Client
- LX3
- Status
- Build phase, launch date TBC
- Role
- Store setup, brand, copy system, theme edits
Being written as it happens. So far:
- Product copy generated in batch from the spec sheet (one prompt, one CSV in, one CSV out, imported via Shopify's bulk importer).
- Theme section edits written in Liquid by AI, checked in a dev theme before pushing.
Spec sheet (CSV)→Claude: copy per row→Shopify bulk import→Liquid edits via CLI
{% comment %} placeholder: badge if stock is low {% endcomment %}
{% if product.selected_or_first_available_variant.inventory_quantity < 5 %}
<span class="badge">Only a few left</span>
{% endif %}
Batch copy is consistent but samey. Every 10th product needs a human rewrite so the range doesn't read as templated.
AI-written Liquid works, but it doesn't know your theme's existing sections. Always test in a duplicate theme first.
Agency operations, rebuilt around AI In progress Reworking how Red Hills Lab runs day to day: briefs, proposals, content calendars and client reporting, with AI handling the first draft of each. Brief · Outcome · Stack · Limits ↓
Problem. A mid-tier retainer model only works if delivery time per client keeps dropping. Most of the hours went to first drafts and admin, not thinking.
Goal. Cut time-to-first-draft on the four recurring documents (brief, proposal, monthly content plan, client report) by more than half, without lowering quality.
- Placeholder: proposal first drafts from a 10-minute call transcript, edited in 20 minutes instead of written in 3 hours.
- Monthly content calendars generated from the client's brand doc and last month's performance notes.
- Placeholder: reporting narrative written from the analytics export, charts still done by hand.
This is the project that turned into the consulting offer. If it works for a marketing agency, it works for most service businesses.
Call transcript→Proposal draft→Human edit→Tasks in ClickUp
Each document has a saved prompt with the same three parts: what good looks like (an example), what's banned, and what format to return.
Tool sprawl is real. Every new AI tool adds a login, a subscription and a place for context to get lost. I've cut back to two.
Client data needs rules. Some clients don't want their numbers in any model. Ask first, write it into the retainer.
The team has to want it. A process nobody follows is slower than the old one.
Boxing gym trial-to-member follow-up Experiment · placeholder Example project: an AI-drafted follow-up sequence for people who book a free trial and don't join, personalised from the trial booking form. Brief · Outcome · Stack · Limits ↓
Placeholder brief. Swap in a real gym client when ready. The structure (form answer in, three-email sequence out, owner approves before send) is the reusable part.
Placeholder outcome. Report conversion from trial to membership, before and after.
Trial form→Zapier→Claude: 3 emails→Owner approves→Send
Approval step is non-negotiable. Fully automated messages to real people is where trust gets lost fastest.
Panel talk series: episode pages from a recording Experiment · placeholder Example project: turn each recorded panel into a web page with summary, speaker bios, pull quotes and clips list, within an hour of the event ending. Brief · Outcome · Stack · Limits ↓
Placeholder brief for a talk-series client. Recording in, structured CMS item out.
Placeholder outcome. Time from event end to page live; page views per episode.
Audio→Transcript→Summary + quotes→CMS item
Transcripts of panels with cross-talk need speaker labels checked by a person. Quotes attributed to the wrong speaker are a real risk.
Experiment log
Short notes on things I tried. Kept, dropped, or mixed. Newest first.
Batch product copy from a spec CSV Keep
One prompt, whole range. Quality is fine at the 80th percentile and the hand edits are quick. Used on LX3.
AI-generated Webflow layouts Drop
Generated HTML doesn't map to Webflow's designer cleanly. Faster to build layout by hand and use AI for copy, CMS planning and embeds.
Proposal drafts from call transcripts Keep
Placeholder entry. The trick was giving it one great past proposal as the example and a list of phrases never to use.
How I work
Every project on this site follows the same loop. It's the same loop I'd set up for you.
1. Write the brief→2. AI first draft→3. Human edit→4. Ship→5. Log what broke
The brief is where the value is. The draft is cheap. The edit is where taste lives. The log is what makes the next project faster.
Work with me
Two ways in, depending on whether you want it done or want to learn to do it.
Hire me to implement it
For agencies and businesses with a physical space who want AI in their marketing and operations without a six-month transformation programme.
- Audit of where the hours go
- Two or three workflows rebuilt, with prompts and tools handed over
- Team trained on the loop above
Learn to do it yourself
Everything on this site is written so you can copy it. The updates go deeper: the prompts, the failures, and what I'd do differently.
- One email when a project or experiment is added
- No schedule, no filler
Updates by email
Thanks. Wire this form to your email tool before launch (Mailchimp, Beehiiv, Substack).
Booking link goes here (Calendly, Cal.com or similar). Until then, email me:
damoy@redhillslab.comAbout
I'm Damoy Robertson, Agency Lead at Red Hills Lab, a full-service marketing agency for businesses with a physical space and an experiential offer: boxing gyms, markets, panel talk series. Brand, web, content, social, strategy.
I think the businesses that get AI into their day-to-day workflow early will pull ahead of those who wait, and that most of the advice out there is written by people who don't run a business. This site is my working notes, in public.