What Changes with AI … and What Keeps Us Grounded
The following best practice principles are still relevant (and should even be more valuable) in the era of AI and automation. This content also covers methods of leveraging AI to greatly improve productivity and automate previous designer > developer process bottlenecks.
Staying Grounded in Core Principles
This section covers best practices, principles, and OKRs that can be used across products and teams - and provides orientation at the platform and project levels.
Design Opportunities with AI
AI provides a number of opportunities to improve user research and measurement, as well as opportunities to simplify interactions and workflows with assistants and agents.
Figma, Claude, & MCP
We have an opportunity with AI tools like Claude Design and MCP servers to greatly improve processes, including the hand-off, design system management, and practical pattern development.
Core Principles
Product, Business, and Engineering Alignment
Design the Meaningful
The following are simple guidelines that I use as a reference (and to help the team) understand user needs, clarify outcomes, and measure quality.
Design critique tends to fail when it has no frame of reference. So I give it one.
Ground the Team Before Design Begins
I take into account the objective, the system around it, its constraints and tradeoffs, and a scalable, incremental path forward — grounded in evidence and shared context, so we solve the right problem.
From platform-level journey maps, information architecture, accessibility checklists, terminology references, and more, we can maintain a system mindset and more easily see connection points, dependencies, and overlap across products.
Deeper Questions to Answer
The following list is a living artifact of key questions for designers, product managers, engineers, and executive stakeholders.
The usability criteria is only meant to be a high-level reference of the most impactful categories.
Understand the “Why“
The PM heard "I want mobile" from customers, worked with engineers to make the desktop app responsive to mobile, checked the box, and moved on.
I requested that we touch base with customers to dig into the “why“ behind this - which was related to receiving network alerts on mobile.
So, I had the team play out the scenario. With a mobile site, customers would never hear the alert coming in (no bell ringing in their pocket).
A native app would accomplish this, but timing was a factor.
So, I recommended a trade-off of allowing users to leverage existing native apps (Outlook, Slack, etc.) for notifications. That will free up real estate on the desktop experience - to improve patterns and interactions there - where most users spend most of their time.
We can make a dedicated native app a future goal in the roadmap.
Solving one problem (with design thinking) solved several.
Improve the Experience and Monetization
As the sole designer, working across 4 PMs and 20+ Engineers, I helped bring a tool to market that allows DemandGen and MarketingOps users the ability to stitch together multiple marketing campaigns into one coherent strategy (view, edit, and automate in one place).
The monetization model stands to be an important money maker for the company since every action a user takes uses "credits" that customers purchase.
Because it's an automation feature (running daily), it stands to make the company more money over time (more credits consumed).
The tool itself, however, is offered free - removing barriers to adoption. Customers have the flexibility to start small and scale as needed.
Solve the Right Problem
I nudged the team toward a change in product strategy to fit what large enterprise customers actually wanted for GDPR workflows - full automation - based on a series of user research sessions.
PMs were wanting to take the "sticky app" approach.
Going down the wrong path would have meant low adoption (lost opportunities) and turning back from the path of a complex new feature would have been very difficult later.
Helping PMs see this from another perspective (giving customers what they actually want and need and improving adoption) would present a major value ad to the cloud platform - and the team's contribution could be SMEs for data privacy.
Secondary value-ads could have been offered, such as ServiceNow templates.
Designing with AI
Discovery, measurement, assistants, and agents
AI-Assisted Research, Discovery and Measurement
Let AI do the repetitive and time-consuming footwork for discovery - turning one-off research projects into standing signals and strong starting points.
Gemini or Zoom meeting transcriptions for user research sessions
ChatGPT for finding categorized themes from user feedback
Claude and ChatGPT for company, persona, and market research
Claude Design for generating 2-3 design options for early design ideation
ChatGPT/Claude for summarizing CSV data from win/loss reports, customer support logs, Slack conversations on CSM channels, etc.
Agents Change the Paradigm
The human sets goals, approves high-impact changes, and measures outcomes. Agents do the rest.
In some business processes, the goal can be embedded and the workflow completely event-driven; only flagging high impact items for human approval.
When designing for multi-agent orchestration, are we considering … ?
If we’re matching the right model to the right task
How permissions, state, and error handling affect error-free operation
When humans should be in the loop
Agent lifecycle management (versioning, ownership, rollbacks, etc.)
Business outcome measurement
Operational visibility and cost awareness
Drift management
Assistants and Agents Working Together
This is an example of a design for a Netflix observability tool.
Common, repetitive workflows were mapped ahead of time to provide the AI (in JSON files), the deterministic data needed to run investigations. Users choose a use case from a prompt library in an assistant (along with account ID, and time window) and the agent runs the actual investigation.
Ground the AI with Deterministic Data
The following are the investigation workflows used as deterministic data for the automated investigation project mentioned above - mapped with Claude initially - and then validated with users.
Process Improvements with AI and MCP
Design System Governance, Patterns, and Improved Hand-offs
The Bottlenecks of Translation and Multiple Reviews
The current process of managing design system implementation and updates, handling designer-developer handoffs with multiple questions and reviews, and guideline creation has always been time-consuming.
We labeled it as a ‘means to an end‘ and repeated the same laborious process for years. We now have an opportunity to improve several processes at once - with one new simple format.
MCP as the Governance Layer
The Source of Truth
If we could leverage the power of AI design tools like Claude and a MCP server (acting as the governance and design system management layer) we could go from completely probabilistic AI to AI that’s more grounded in deterministic rules.
Designers and developers would be working off of the same source of truth.
A Fast, Streamlined Process
If the process were redesigned with 3 core parts, we may be able to implement multiple improvements at once.
The design (Figma file, screenshot, or prompt in Claude Design)
The MCP Server (maintains and understands how to apply the design system and can process any designs sent to it with the correct tokens, components, layouts, behaviors, accessibility, etc.)
The Staging Server (The endpoint of the process for designers. With an engineer’s help, designers can go from idea to production-ready React pages very fast).
The team would review on a real, production-ready page, instead of a Figma simulation. Managing the design system would be done directly on the MCP Server, instead of the traditional Figma > Storybook > Production flow. Figma won't disappear. Its role will shrink from "where the product is designed" to "where humans resolve ambiguity that agents can't efficiently resolve."
Moving Out of the Weeds
The time saved could be spent creating reusable design patterns at a higher level than components only. Guidelines at the component (atom or molecule) level are typically designer-created documentation for other designers.
Patterns become much more valuable when they’re created in a format that matches the way we all think.
Design for Where We Live
Higher-level patterns match the way we all think and work.
Product (Thinking in features and flows)
Engineers (Thinking in pages, flows, and seeing components as a means)
End Users (Thinking in terms of tasks and flows)
Designers (Matching the mindset of the user with tasks and flows)
Platform Level Patterns
An example of a use case (or platform-level) pattern would be a simple wizard that the design team wants to present in a modal. That one pattern is not a component. It’s a collection of pages and components, organized into a flow.
Modal window
Layout spacing
Form fields and labels
Typography for title and subtitles
Progress Bar
Buttons
If patterns were approached from this perspective, they would be much more understandable, practical in terms of how we think about solutions, could provide more deterministic data for the MCP Server, and may help the team move faster without compromising quality.
Start Where You Are
If there are enough skills to set up a pilot MCP Server - teams can partner at least one designer and developer to run a pilot project for one project for one release.
Review accuracy - and the additional work on the Engineers side. Despite the potential need to start slow, I believe teams will see immediate productivity enhancements.
For the embedded MCP design system, start with a pre-existing design system (eg. AirTable, Material Design, etc.) to test with. These frameworks are fully featured, vetted in many ways, will simplify setup, and - with needed branding additions - could become the production design system (post pilot).
Improving Both Strategy and Execution
Everything mentioned above comes back to 3 core areas:
A more informed strategy, grounded in data and communicated clearly to the team.
A streamlined AI process that automates repetitive work, provides automatic governance, and allows the team to shift more time toward strategy and validation.
Practical Patterns that provide a much richer and more practical match to the way we approach projects.