Skip to content

00 / AI integration

Design AI into the product you already have.

I decide with your team where AI belongs in the workflow, prototype it in code, and usability test output that changes from run to run.

01 / The problem

AI features fail in specific ways. Design for them.

  1. A wrong answer stated with confidence.
  2. No way to check where an answer came from.
  3. No point where a person can step in.

These are design problems before they are model problems.

02 / The work

The work, in order.

  1. 001 / Placement

    Where AI belongs in the workflow.

    Before any screen is drawn: which step of the existing workflow the AI takes on, and what it should and should not answer.

  2. 002 / Prototype

    Prototypes in code.

    A static mockup cannot show what a model does on the second try. The interactions that carry the most risk are built in code, so your team sees the behavior, not a drawing of it, before engineering commits.

  3. 003 / Test

    Usability testing of output that changes run to run.

    The same question can get a different answer tomorrow. Testing checks whether people can tell when to trust an answer and when to step in, across that variation.

03 / Proof

An assistant that answers from the customer’s own findings.

Amazon Q Security / Amazon Web Services

Amazon Q refused every security question, because a wrong answer in security is dangerous. I designed a chat extension that answers from the customer’s own GuardDuty findings.

Status
Design concept, shelved before the supporting API existed
Result
Interaction patterns added to the Amazon Q pattern library
Delivered for
Amazon Web Services
Amazon Q Security design concept. The assistant answers from the customer’s own findings, with each finding’s severity written out.

Delivered by Brad Shroyer as an Amazon Web Services employee. Shown as principal experience, not as a Veteran Design contract.

Amazon Q Security case study on bradshroyer.com

04 / Scope

No package. Scoped to your product.

Package
None. The scope is written for your product
First step
30-minute discovery call
Statement of work
Tailored, within 48 hours of the call
Also runs as
Part of an embedded product design engagement
On programs
AI usability testing for defense and federal program teamsFederal capabilities and entity data
AI tooling
AI tooling supports the work; I review and own the result. Its use is disclosed in writing in every statement of work

05 / Next step

Start with where you think AI belongs.

A 30-minute call. I will tell you whether I agree and whether this practice fits. If it does, a tailored statement of work follows within 48 hours.