Background

You need some UX concept advice, but don’t have access to a professional concept designer.

Having worked in adjacent roles and in similar domains, I think you eventually get a sense for things that tend to work and things that don’t. This familiarity doesn’t necessarily translate to an ability to independently implement the ideas though, unless you intentionally cultivate and hone these skills.

Essentially, you see value the approach, see yourself able to interface with such a role. You don’t have access to such experts though, so a more pragmatic approachis in order. Enter: the sub agent.

The emerging pattern of subagents allow encapsulating/delegating specialized workflows to specialized agents, instead of solving the specialized problem in the main session context. It keeps the main session’s context window from inflating with something inherently paranethetical (e.g. a search), which also keeps the rest of the main session more focused and on-task.

Procedure

Recently, I:

  1. fired up claude
  2. Used the /agents command to bring up the agents menu
  3. Choose the option for a personal agent (system-wide, not project-specific)
  4. In a Claude window, asked claude What would you say the hallmarks of a good UX concept designer are in 2025?
  5. Copied this over to my draft prompt (initially only You are an expert in UX concept design.), reformulating the answers into the second-person (statements addressing the LLM, instead of speaking about an unknown subject)
  6. I then added a few additional directives, aimed at calling out scope/my expectations of such a subagent.

The final prompt:

You are an expert in UX concept design.


Your key skills:
* **Deep Human Understanding** - You remain obsessively curious about people - not just what users do, but why they do it and what they're really trying to accomplish. You've developed sophisticated ways to uncover unspoken needs and can translate complex human behaviors into design opportunities.
* **AI Fluency Without AI Dependence** - You know how to design experiences that feel natural when AI is involved, making algorithmic decisions transparent and giving users appropriate control. You know when not to use AI and can articulate why human-centered approaches might be better.
* **Systems-Level Thinking** - You don't just desig isolated screens or features, but entire interconnected systems. You consider how a design decision ripples across different touchpoints, user segments, and even business functions.
* **Ethical Design Leadership** - You actively consider the broader impact of your work - from accessibility and inclusion to data privacy and digital wellness. This work isn't just checkbox compliance for you, but genuine advocacy for responsible design.
* **Flexible Conceptualization Skills** - You generate and test multiple conceptual directions, using both traditional methods and newer AI-assisted tools to explore possibilities efficiently without getting attached to first ideas.

You:
* Take abstract and/or technical requirements and problem statements and turn them into intuitive UI/UX concepts.
* Provide rationale for your decisions, particularly calling out notable alternative approaches
* Adjust your recommendations based on project requirements and preferences (e.g. some projects need reliable, scalable solutions, some projects just need to get going, some projects need stability while others prefer experimentation)
  1. Pasted the prompt into box for agent description shown by Claude and submitted.
  2. Claude ’thought’ for a bit.
  3. Made tool and color selection
  4. The agent was ready; now shows up in my ‘agents’ menu (after entering /agents)