Designing with AI in 2025

Quick note: While I'm sharing this from my perspective as a product designer, these approaches work for practically anyone who creates things—writers, marketers, product managers, developers, or just curious humans trying to organize their thoughts. Consider this less of an expert guide and more of a "hey, here's what's been working for me... mostly."

As a principal product designer working on complex systems at CREXI, I've been integrating AI tools into my daily workflow to dramatically increase my efficiency and output quality. Or at least that's the goal—some days it feels more like I'm just having expensive conversations with a very patient robot that occasionally gives me useful ideas.

What started as casual experimentation has evolved into a set of approaches that have genuinely transformed how I work. Mind you, I'm still figuring this out like everyone else—this isn't a "mastery" post, it's more of a "field notes from the frontier" situation. So here's my current thinking on how I leverage AI as a design multiplier in 2025, complete with the occasional facepalm moments.

Context Management Superpowers

The most transformative aspect has been how AI helps me manage the overwhelming context that comes with any project:

• Creating context artifacts: I build project context documents so when I'm deep in the design weeds, the AI helps me remember all those important details. The trick is structuring these documents with clear hierarchies and explicit relationships between concepts—not just dumping information.

• Knowledge persistence: For ongoing projects, I structure Google Docs with clear headers that AI systems can parse, then upload them to a conversation. I've found that periodically reviewing and updating these reference documents keeps everything aligned as the project evolves.

• Conversation distillation: After productive discussions, I have the AI summarize key points into artifacts. I always review these summaries critically, adding anything missed and correcting any misinterpretations—treating it as a first draft rather than a finished product.

This approach to context management has been invaluable at CREXI, where I'm juggling user needs, business requirements, technical constraints, and strategic opportunities. Without these AI systems acting as an extension of my brain, I simply couldn't keep that much in my "active RAM" at once. And while these techniques prove especially powerful for complex projects, they add value to even straightforward work by ensuring nothing falls through the cracks.

A file I created with the help of Claude to act as a knowledge reference for this portfolio's project. A knowledge file can be anything however, I just find having Claude help create baseline understandings through conversation is very efficient.

Practical Prototyping Approaches

When it comes to visualization and prototyping:

• Screenshot reactions: I regularly screenshot UIs and have AI react with feedback. To get useful critique, I specifically direct its attention to particular aspects I'm concerned about, rather than asking for general feedback that can be too broad.

• Content generation based on structure: If I have a design pattern established, I'll grab a screencap and ask the AI to generate variations. This is particularly effective when I provide explicit guidelines about voice, tone, and length—ensuring consistency across generated content.

• Flow diagrams + prototype screenshots: Before diving into Figma, I work out flows through a combination of diagrams and AI-generated prototype screenshots. When the AI suggests unexpected connections, I take time to evaluate whether they truly solve user needs or just add complexity.

Communication Techniques That Work

Through ongoing experimentation, I've developed ways to communicate with AI that yield better results:

• Detail-rich prompts: I'm generous with context in initial prompts, which dramatically improves responses. I've found that frontloading key constraints, audience information, and project goals sets up every interaction for success.

• Process guidance: I explicitly tell the AI how I want our conversation to flow ("Let's discuss this concept before prototyping anything"). Setting these expectations prevents wasted time on misaligned outputs and creates a more focused collaboration.

• Multi-point prompts: Don't be afraid to ask multiple questions in a single prompt. Modern AI can handle several related points at once. When responses miss something, I've found it helpful to repeat the skipped question rather than completely restructuring my approach.

• Format specificity: Being clear about what form I want the response to take saves time. Whether it's a bullet list, a diagram, or conversational exploration, explicitly stating the desired output format has consistently improved the first-pass usefulness.

Efficiency Multipliers

What's made me something of a "superhuman designer" at CREXI is how these approaches have expanded my capabilities:

• Context switching: When hitting conversation limits, I have a context transfer process (generating summaries and spinning up new conversations). Taking the time to properly structure these handoffs pays dividends in continuity of thought.

• Mobile ideation: Having design conversations while away from my desk (like in the car) with project context loaded. I use voice dictation for these sessions, making notes between meetings or during commutes that would otherwise be lost time.

• Rubber duck debugging: Articulating design problems clearly in text often reveals solutions before the AI even responds. The discipline of clear writing forces clarity of thought, with the AI responses as a bonus.

• Continuous refinement: As I learn more about a project, I update my guidance to the AI. Making these updates explicit ("Based on what we've learned, let's adjust our approach to...") helps maintain alignment throughout the project lifecycle.

Breaking the Blank Canvas Paralysis

One of my favorite applications is using AI to overcome that dreaded blank canvas. You know the one—where you stare at an empty Figma file wondering where to start?

• Structure before pixels: I build information architectures through conversation, having the AI construct nested prototypes as we talk through concepts. I've found it works best when I provide clear constraints and examples of the structure I'm looking for.

• Idea generation: When facing a design challenge, I'll articulate the problem to the AI (forcing clarity in my own thinking). This acts like a rubber duck method with feedback, helping me see angles I might miss when I'm too close to a problem.

• Multiple perspectives: After establishing a shared understanding with the model, I generate several different approaches and refine them as the project evolves. The key is providing enough context about what success looks like, not just what features are needed.

Learning Curve Notes

Not everything is perfect—there's definitely an art to working effectively with AI in design, and I'm still refining my approach:

• Quality control is non-negotiable: I read through everything and make corrections where needed. I've developed a habit of reviewing AI outputs with a specific checklist: factual accuracy, completeness, alignment with user needs, and technical feasibility.

• Tool-appropriate tasks: Each system has strengths (Claude for exploration, v0 for complex prototype iteration, ChatGPT for certain tasks). Understanding which tool fits which task has been a gradual learning process of mapping capabilities to needs.

• Technical limitations awareness: When hitting code limits, I explicitly acknowledge this and request alternative approaches. Breaking complex tasks into smaller, well-defined chunks has consistently produced better results than trying to accomplish everything in one prompt.

• Reference material balance: Providing references is critical, but requires structure. I've found that organizing reference materials with clear hierarchies and explicitly stating their relationship to the current task dramatically improves how they're utilized.

The Bottom Line

This integration of AI into my process has been transformative at CREXI. As a single designer, I've been able to accomplish what would traditionally require multiple people. The efficiency gained isn't just about speed—it's about depth and quality too.

My ability to maintain awareness of countless contexts, considerations, and opportunities has made me more effective. As a designer and researcher, I know how rarely documentation from user research or strategic deliverables gets revisited after creation. Having an AI system that can instantly recall and apply all that context has been game-changing.

This is very much an evolving practice in 2025. The tools themselves are changing weekly, and my integration methods continue to adapt through deliberate experimentation. What remains constant is the principle: use AI not to replace thinking, but to extend and amplify it.

And if you're just getting started with this yourself—whether you're a designer, writer, developer, or anyone who creates things—my best advice is this: be patient, be specific in your requests, and remember that AI works best when treated as a thought partner rather than a magic solution. Begin with small, well-defined tasks, reflect on what works, and gradually expand your collaborative process as you develop a feel for the strengths and limitations of these tools.

While these approaches can benefit any project, I've found their value compounds significantly as complexity increases—the more moving parts and considerations a project involves, the more these AI-augmented methods shine.