Redesigning the Agent Library
to increase agent usage in the org.

Overview: As AI agent workflows scaled across our customer organizations, employees couldn't find the right agent in a flat, unfiltered grid, and automations had no home. I conceptualized and designed the Agent Grid as an org-wide surface: curation, access control, chat-first entry, and dedicated automations.
Company: StackAI
Feature: Organization Agent Grid
Year: Q4 2025
Impact: Engagement grew 3.4× in the first month, from ~1,400–1,900 to 6,200+ monthly users with 80–105k chat page views from end users.

PROCESS

User feedback and must-have requirements for the agent grid

Grid redesign, curation and personalization controls

Chat-first exploration, conceptual brainstorm session

Automations model, trigger-based workflow discovery and configuration

1. What was broken

As workflows scaled, the grid became a flat list with no filters or structure. Automations had no home either, so they shared the same cards as on-demand agents. Synthesized enterprise feedback into a short brief: filters, tags, branding, and access control were must-haves; the rest could wait.

2. Rebuilding the grid people browse every day

Redesigned cards, hierarchy, labels, and sort so the library feels curated and owned by the customer, not a flat dump of workflows.

3. What if chat was the front door?

Looked at how Lindy, Langdock, and Gumloop lead with conversation, then sketched what that shift would mean for StackAI's published agents surface.

4. Automations needed their own home

Trigger-based workflows don't behave like on-demand agents. Mapped how they should be discovered, configured, and monitored on a dedicated surface instead of the same cards.

SOLUTION

Chat as Entry Point. Most employees start with intent, not a catalog. After benchmarking chat-first tools like Lindy and Langdock, we made conversation the default surface so people can ask, run, and return to work without hunting for the right card. The agent grid stays below as a browsable catalog and shortcut layer for favorites and discovery.

Chat-first entry point with suggested prompts and favourite agents below

Chat composer with tools and knowledge menu open

Organization Agent Grid. The old grid was a flat, unfiltered dump, so people could not find the right agent as workflows scaled. We rebuilt it as an org-owned library with department structure, visibility controls, filters, favorites, and sort, the enterprise must-haves from customer feedback, so curation and access control live in the product instead of tribal knowledge.

Organization agent grid with favourites, filters, and agent cards

Favourite agent cards with labels, descriptions, and integrations

Dedicated Automations Screen. Trigger-based workflows do not behave like on-demand chat agents, yet they used to share the same cards. That mixed model hid what was running and made setup hard to trust. A dedicated surface lets people discover automations, configure and test triggers, then monitor and debug failures without competing with the day-to-day agent grid.

Automation details screen with workflow graph and setup controls

Automation workflow detail with AI agent prompt and setup controls

INSIGHTS

The grid is becoming a platform

What started as a simple app launcher is evolving into an org-wide AI operating surface, chat, agents, and automations in one place, each with a distinct interaction model.

Chat and agents are complementary, not competing

The chat-first model doesn't replace the grid; it gives users a faster on-ramp when they don't know which agent to pick.

Automations need their own UX contract

Background workflows have different states, failure modes, and ownership responsibilities than on-demand agents. Forcing them into the same card pattern was the wrong abstraction.

The grid is a trust surface

If it looks like a third-party tool, enterprise employees won't adopt it. White-labeling and curation are table stakes, not polish.