Building an Agentic Flows Inventory in Airtable
Upgrade your design system inventory with agentic flows
👋 Get weekly insights, tools, and templates to help you build and scale design systems. More: Design Tokens Mastery Course / Podcast / My Linkedin
As designers, we know context matters most. I created an Agentic Flows table (part of the design system inventory) in Airtable — to track how AI shows up in the user experience.
Why I love it:
Spotting new opportunities where AI could improve or simplify existing flows
Challenging assumptions like:
Are we forcing AI into places it doesn’t belong?
Are we missing obvious automation wins?
Helping us balance impact, complexity, and cost
What’s in my Agentic Flows table?
I track every user-facing AI interaction, including:
Error states (e.g., when AI fails or times out)
Editable AI-generated content
Trust messaging
You need to think about your flows and interactions
For each flow, document:
AI Role: Is it optimizing, suggesting, explaining, etc.
Figma Link: Keeps everything connected to the design.
Notes: Design logic, edge cases, UX considerations.
Status: In design, review, implementation, etc.
Components Used: Linked to the component library to track reuse
And a bonus for those who want to go deeper:
Collaborate closely with product or engineering so you can add fields like:
Model Required: GPT-3.5, GPT-4, Claude, etc. — helps estimate performance, latency, and feasibility.
Cost Estimate: Based on token usage and frequency — essential for balancing UX ambition with business constraints.
Btw, it is impossible to predict exact costs. 🫣
Why mapping matters
AI isn't just a feature — and it's not going away.
But before you commit to building, ask yourself:
Are we leaning too hard on AI to solve every flow?
Not everything needs a smart suggestion. Sometimes, a clear button or a simple form is better UX.
Should this flow be redefined with AI — or left to humans? Some tasks are too nuanced, sensitive, or expensive to automate.
Are we defaulting to “AI thinks…” when we should just ask the user?
Transparency is key — think tooltips, footnotes, and fallback options.
How expensive is this to run? There’s a big difference between a helpful $20/month suggestion and a background agent quietly burning $1,000/month.
Mapping flows first helps you design intentionally — so you know where AI belongs and where it doesn’t.
👉 Try It Yourself
Start small:
Map what already exists.
Use Airtable (or wherever you have your inventory).
Iterate in Figma and then link your Figma flows.
Track models, cost, and reuse.
Build prototypes.
Track behaviour
Iterate again. 🙃
It’ll bring clarity and visibility, save time, and make every AI decision more intentional.
By the way, if you missed the agentic flows overview posts, here are AI UX Patterns parts 1 and 2, plus Automation in design systems.
Enjoy. Stay tuned. ✌️
💎 Community Gems
Design system annotations, part 1: How accessibility gets left out of components
by GitHub
🔗 Link
Servicenow, Horizon design system new version with AI guidelines
At ServiceNow, our mission is to use the transformative power of AI within workflows, driving enhanced performance and world-class experiences.
🔗 Link
Gemini 2.5 Pro Preview: even better coding performance
🔗 Link
Conference is happening in 12 days 🔥
🔗 Link
Great insights