About
How I got here.
Websites first, since the late nineties. Then maker, then creative director, then group, then a 70+ person department. In 2025 the job changed shape, and the change was orientation rather than size.
Most people doing this work arrive as technologists, who can build anything and are guessing at what’s worth building, or as strategists, who can frame the opportunity and can’t test it. I came at it as a creative director who never stopped tinkering, so I’d rather build the thing than have an opinion about it. Forty-odd experiments in, that is still the fastest way I know to find where a capability actually stops.
Running creative was vertical: one department, one set of standards, one place to push. AI transformation is horizontal. It reaches every function in the agency and now into Saatchi & Saatchi Health and Razorfish Health, and nobody adopts a new way of working because they were told to. That takes a different kind of creativity.
Want to see some of my creative work? Watch the reel. It’s behind a password.
Selected work
Three that made it into the work.
Each one got past the prototype stage and into someone’s actual process. Where the work was for a client, I describe how it works and nothing else.
-
How do you teach taste to a machine?
A few hundred forced choices become fourteen evidence-backed preference objects. New work then gets reviewed against them.
Read the case study
-
Building a shared decision engine
Audience strategy, channel logic and business rules in one view, bringing five different capabilities together before anything gets made.
Read the case study
-
Same facts, different context.
An assignment carried through three simple questions, from assignment to production, AI keeping things on track the whole time.
Read the case study
Experiments
Things I made to find something out.
Most were small. Several were an afternoon. A few became the work above.
- Shared canvas for team decisions
- Assignment to production-ready content
- Turns choices into a taste profile
- Surfaces rare signals, not loud ones
- One content system, many clients
- Brand planning with scenario comparison
- Asteroids, but you solve math
- Builds assets from approved parts
- Patient conversation into strategy openings
- Strategy through to live activation
- Claims policy enforced by the build
- Grades copy through switchable audience rubrics
- Builds and validates synthetic personas
- Turns a protocol into a brochure
- Monster Quest: A side-scrolling adventure
- Routes work through review gates
- Generates banner sets from one layout
- Tests which findings survive pressure
- Forced choices reveal what you prefer
- Captures pages into a review packet
- Checks a persona against real evidence
- A graphic novel, page by page
- Diffs two documents side by side
- Scores plan options on fixed criteria
- Turns live sites into design tokens
- Mines caregiver forums for unmet needs
- Coordinates a sixty-day adoption sprint
- Fast prototyping outside the main stack
- The site you are reading now
- Searchable catalog of prompts and tools
- Voice answers grounded in one document
- Reads a corpus for hidden patterns
- Traces a claim back to source
- Ask about the text you select
- Scroll-driven animation, just to try it
- Answers only from one approved source
- Pulls the final frame from video
- Keeps the deciding with a person
- Breaks long content into modular parts
- Reads what visitors said out loud
-
Deciding
How does a team actually agree on something?
-
Finding
What’s true in here that nobody has noticed?
-
Making
How does the work actually get out the door?
-
Curiosity
What pulls at me when there’s no brief attached?
Two of these are live. Go break them.
- Knowledge Exploder Paste in any body of text and it maps the themes, the outliers, and the things almost nobody is saying. Bring your own, or load the sample. Opens in a new tab
- SignalField MC Three message ideas, scored on six language signals, then ranked against a made-up audience. Rebuild that audience fifty times and see whether the winner stays the winner. Every input is invented, which is the point. Opens in a new tab
Writing
Some things I've written lately.
These are some of the ideas I keep returning to, and the conversations that come with them.
-
How much thought does this work deserve?
Chopsticks or Chopin -
The hard problem isn’t generation. It’s preserving intent.
Marge vs. the Monorail -
Data without workflow is a warehouse. Workflow without data is automation running on assumptions.
Output is where AI work starts -
Every email they sent was clearly written with AI. He knew immediately.
The second read tax -
We’re repeating the recycling mistake. This time with work.
AI accountability and the recycling mistake -
Technology allows us to gain velocity. It may change the process, but the soul of the work remains.
Velocity and the soul of the work
Talk to the site
Ask it something hard.
A voice assistant runs on this page. It reads a handful of files I wrote by hand, it won’t pretend to be me, and there’s a written list of things it refuses to discuss.
Needs a microphone. See if you can get it to slip.
Or read the seven files behind it, and why the API key never reaches your browser
Contact
Let's talk.
This goes straight to my inbox. If you are hiring, a line about the shape of the role is more useful to me than a job description.