The prototype is the argument. Production is the proof.
I'm usually the one on the team who takes the technical deep dive, builds the proof of concept, then sets the direction for the rest of the team to run with it. Leading small teams through that process, over 25 years, has meant computer vision in Adobe Creative Cloud, connected cars wired to Salesforce, robots that take orders from AI agents, and agentic tools teams use every day.
Try the Idea EngineerOne quick check. Each run spends real money on real models, so this keeps bots out.
Your concept shows up here.
| Step | Model | Start | Time | Result |
|---|
Whether it's a request or an idea of our own, I dig into the new technology first and figure out what it can really do.
Hardware, software, or both, built fast enough that the team can react to something real.
Walk people through how it works and train them up, so the project doesn't depend on me.
On a keynote stage, in a customer meeting, or in a tool people use every day. What doesn't hold up becomes a lab note.
Some of these started as a request and some as an idea we couldn't let go of. All of them ended up in front of real people, and some of them became product.
I'm on Salesforce's demo team, so a lot of what I build is made for a stage, and robots and cars show better than code. Every one of these runs on software built end to end, from the services and APIs to the data, cloud, and the interface people touch. Plenty of my work has no hardware at all, like the agentic internal tools just below.
From hand-coded moves to agent-dispatched, vision-guided robots by Dreamforce.
A concept car wired to Salesforce for CES, then a console simulator that travels.
A chain of custody demo whose ideas made it into Salesforce Health Cloud.
Light-up hair, a voice on his sweater, and a phone remote that reaches 2 km over LoRa.
Persistence-of-vision LED sculpture driven by a custom OpenGL effects pipeline.
Agentic and AI tools that teams inside Salesforce use every day. They're internal, so here's what they do and what they're built on instead of screenshots.
A multi-model AI pipeline (LLMs, diffusion, and vision models, plus fine-tuned LoRAs) that creative teams across Brand, Studio, Keynote, and Demo use to generate assets. Long-running jobs, sharing, access control, and retries are all handled.
Built for Salesforce's User Insights team in support of Global Marketing Messaging. It turns research on what keynotes tell customers and the public into synthesized analysis and recommendations, surfaced through dashboards and reports.
The experiments that broke. I keep them because they taught me more than the ones that didn't.
Apple turned down the first WIRED iPad edition under a new rule against interpreted code. We took a long weekend, rewrote the whole app in Objective-C, and it shipped.
Now I treat platform rules as an architecture risk, and keep model providers pluggable.
A gutted smart speaker turned digital puppet lost its 2.4GHz link once it was surrounded by cameras, operators, and their comms. I rebuilt it on LoRa at 915MHz and it worked on show day.
Now I test in the room the thing will live in, as early as I can.
Writing about what the demos taught me, mostly about where things actually break.
