Agentic Experience · Case Study
Work in progress

Hitch — a new agent inside Pencil

An ongoing, Google-funded project with Pencil, the AI advertising platform used by global brands. I'm designing the end-to-end experience of Hitch, an agent that measures what AI assistants say about a brand, audits everything the brand publishes, and fixes the gaps — all inside a chat thread.

Agentic UX Funded by Google Scope → Flows → Edge Cases 2026 · Ongoing
01The brief

What I'm building

Shoppers increasingly ask AI assistants what to buy — and AI agents are starting to complete the purchase themselves. Pencil is building Hitch — a new agent for this moment — and Google is funding it as a co-development. I own the agentic experience design.

Client
PencilAI advertising platform for global brands and agencies.
Backing
Funded by GoogleA co-development through Google's gTech channel, delivering through 2026.
My role
Product Designer — agentic experienceScope, end-to-end flows, entry points, edge cases, interaction model.
Status
In progressBehaviour designed and agreed; the thread experience is being drawn now.

The agent runs one loop:

01

Measure

How often AI models mention and recommend the brand — a dated baseline every later score is measured against.

02

Audit

The six touchpoints AI reads — feed, product pages, markup, ads, creator videos — scored for consistency, with evidence on every score.

03

Resolve

Drafts the fixes; a human approves every one. Writes go out with rollback, the rest become briefs for whoever owns the channel.

Then it re-scores monthly against a holdout set — proof the fixes moved the number, not luck — and reports it as a heatmap leadership can read at a glance.

02The problem

Brands that contradict themselves lose the shortlist

Before an AI assistant recommends a product, it reads everything it can find: the ads, the creator videos, the retailer listing, the product page, the hidden markup. Those surfaces are produced by different teams on different cycles — so they disagree. A video says "lightly sweetened" when the feed says "no added sugar". The price is right in the feed and stale on the page.

The AI can't tell who's right, so it goes vague — or recommends the competitor whose story adds up. And the brand never finds out: no existing tool shows whether a brand's claims are consistent across everything AI reads. That cross-checking, plus proving the fixes lifted visibility, is the genuinely new part this project was funded to build.

Brands can't buy their way into AI answers. The only lever is making everything they publish tell the same story.

Work in progress Want to know how this project is going? The work is ongoing, and much of it has to stay under wraps for now. If you'd like to hear how it's progressing — the thinking, the process, what I'm learning along the way — do let me know. I'd be happy to discuss it.
Let's talk →