AI Integration,
Built To Run.
We build AI into sites and internal tools where it earns its cost, and we say so when it does not. Most of what gets sold as AI on a website is a chat bubble nobody opens.
How we actually work on this
Agents that do a job, on a schedule
We run a daily SEO audit agent against our own site and an autonomous optimiser against our own ad account, with hard-coded spending limits. We build this kind of thing for clients because we run it ourselves first and know where it breaks.
Retrieval over your own content
Answering from your documentation, catalogue or case history rather than from the open internet. This is the version that is worth paying for, and it lives or dies on how the content is structured.
Measuring AI assistant traffic
Referrals from assistants do not show up cleanly in a standard analytics setup, so most dashboards are quietly undercounting an increasingly important channel. We have written up how to instrument it.
Where we say no
If a feature only works when the model behaves, it is a demo, not a product. Anything we ship has a defined failure mode and a path that works when the model is unavailable.
The work behind the claim
Not a list of capabilities. The actual builds, the writing that shows the thinking, and the sectors we have done this in before.
Case study
Lamacraft
One parametric spec driving 210 parts, WebGL and render
Case study
OLVR
Three data streams reconciled into one product interface
The thinking, written up
Sectors we have shipped this in
Tech & SaaS
E-Commerce
Public CompaniesScoped so you can say no to half of it
Find the job
We look for a task that is repetitive, rule-shaped and currently done by a person. If we cannot find one, we say so.
Prototype against your data
A narrow version on real content, measured, before anyone commits to a platform or a monthly bill.
Ship with rails
Spend ceilings, a defined failure mode, and a path that still works when the model is slow or unavailable.
Vancouver-based, working across Canadian, US and European time zones. This work needs more conversation than a brochure site does, so we keep the client list short enough to have those conversations.
Straight answers
Is this just a chatbot?
Usually not, and usually it should not be. The work that pays for itself is generally an internal workflow — triage, drafting, monitoring — rather than a widget on the marketing site.
Where does our data go?
Wherever we agree it goes, documented before anything is built. If your content cannot leave your infrastructure, that constrains the design and we will say so up front rather than after.
What does it cost to run?
Model usage is metered, so we scope a monthly ceiling and build the rails to enforce it. We do this on our own systems and would not ship one without it.
Start with the teardown,
not the invoice.
Send us your URL and we'll send back what's slow, what search can't see, and what we'd change first. Two business days, free, no call required.