The One Intelligence
Framework
What does it take to make a device ecosystem feel like a single, coherent mind? A deep dive into AI design heuristics, output evaluation, and generative workflow architecture — applied to Amazon's fragmented creative reality.
Before AI workflow, I was a UX and interaction designer — most notably at United Airlines, where I worked on large-scale digital systems serving millions of users. That work trained me to think in systems, evaluate designs against user mental models, and document decisions that engineers and non-designers could actually act on. Those instincts are what I brought here: if a prompt framework is the “interface,” it needs the same heuristic rigor and quality evaluation that any product design decision does.
A world-class ecosystem.
A fragmented creative reality.
Amazon’s device portfolio — Echo, Fire, Ring, Alexa — is one of the most sophisticated consumer technology ecosystems ever built. But behind the scenes, each product line was managed by separate teams and supported by different agencies. Every group had its own visual logic, tone, and interpretation of the brand. The result: a brand that looked and felt different depending on where you encountered it — on a shelf at Target, in a paid social ad, at a trade show, on the homepage.
“The goal was never to make these products feel like they work together. The goal was to make them feel like they were never apart.”
Fragmentation isn’t a brand problem.
It’s a workflow problem.
Style guides don’t scale across agencies — they get interpreted and drift. The real issue was upstream. I recognized this from UX systems work: you can document a design system perfectly, but if it doesn’t integrate into the actual workflow of the people building the product, it gets ignored at implementation. The shift: engineer the input, not police the output.
I didn’t design a campaign.
I engineered an intelligence layer.
Mapped every touchpoint using product design audit methods — treating each channel as a “screen” in a larger user journey. Identified where inconsistency broke the user’s mental model of the brand: retail, social, experiential, web.
Translated brand language, visual identity, and product hierarchy into a structured set of design heuristics — analogous to a design token system — that any generative AI tool could draw from to produce consistent output. Covered:
Built a quality evaluation rubric to assess AI-generated outputs — the same way a design review assesses whether a shipped screen meets product standards. Defined pass/fail criteria, drift signals (e.g. wrong product hierarchy in visual composition), and edge case handling for launches and co-branded moments. Designed to be used by external agency teams without direct creative oversight.
Each channel became an input variable in the workflow system — analogous to designing responsive components that maintain design integrity across breakpoints.
Working with Czarnowski’s production team, I built and tested a multi-tool AI workflow across formats. Provided heuristic documentation, evaluated outputs at each iteration, and gave structured feedback — functioning as design quality arbiter between the framework and final production output.
What this work
actually shows.
Design heuristics are the unit of work that makes AI scalable. The decisions that matter — how a brand speaks, how products relate, what a user feels at each touchpoint — are design decisions. AI is the tool that makes them reproducible. My UX background is what makes them rigorous.
Evaluation is not an afterthought. The rubric I built to assess generative output quality is as important as the framework that produced it. Without a systematic way to identify drift, failures, and edge cases, the system degrades over time.
“I don’t just use AI tools — I build the systems that make them work together. The output is consistency. The work is in the architecture.”