Case studies, Kore.ai, AI and enterprise SaaS

Kore.ai Agent Marketplace: the PLG front door for enterprise AI

The discovery and deployment layer of the Kore.ai Agent Platform: 200+ agent templates, integrations with 150+ business apps.

Overview

The Kore.ai Agent Marketplace is the discovery and deployment layer of the Kore.ai Agent Platform, the place where enterprise teams go to find, evaluate, and install pre-built AI agents across Customer Service, Sales, HR, and IT. With 200+ agent templates and integrations with 150+ business apps, it was built to make enterprise AI feel immediately accessible rather than months away.

The marketplace was built from the ground up: no existing design language, no prior UX framework, just a product vision and a set of engineering constraints. That covered persona research, information architecture, agent card design, installation flows, and the component library that shipped with the product.

The challenge

Before the marketplace existed, deploying an AI agent at an enterprise required weeks of configuration by a developer. Non-technical users could not get started without IT involvement. The problem was not awareness, teams knew they wanted AI. The problem was the gap between wanting it and having it running.

Building a custom agent from scratch took weeks of engineering effort
Non-technical users (HR, Sales, Customer Service) had no self-serve path to AI deployment
IT admins needed enterprise-grade configurability, but the same UI had to work for a business analyst
No shared language existed for what an "AI agent" could or should do in their context

User research

The real friction was not finding agents, it was trusting them enough to install. Four distinct user types were mapped: the IT admin, focused on security, permissions and deployment governance; the solution architect, who wants to inspect tool dependencies and edge case handling; the product owner, evaluating ROI and fit; and the business user, non-technical, who needs a clear value prop and a guided install path.

Discovery was expected to be the hardest problem. It was not. Users found agents quickly using search and category filters. The real drop-off happened before install: users could not tell whether an agent would work for their specific context.

"What tools does this actually need?" was the most common pre-install question
Users wanted to see integrations before committing, not after
Template descriptions were too generic to drive confidence
Power users wanted to inspect agent logic; casual users needed a one-liner

Design strategy

Preview before commitment. Every agent card and detail view had to answer "will this work for me?" before the user reached the install button: surface integration requirements upfront, show use-case context rather than feature lists, provide a template preview, and make tool dependencies scannable in under 10 seconds.

Progressive complexity. The same interface had to serve a business analyst and a solution architect, so rather than two experiences there is one with progressive depth: a default view with one-click install and sensible defaults, expandable detail for tool configuration and permission scoping, and an advanced mode for full agent logic inspection.

Agent card system

The agent card is the atomic unit of the entire marketplace. Every card had to communicate function, context, and compatibility at a glance, without overwhelming users or dumbing down for power users.

Category tag, a scannable domain label (HR, Sales, IT)
Agent type badge, Agentic App vs Agent App
One-liner value prop, what the agent does, in plain language
Integration chips, the apps it connects to, surfaced immediately
Use-case tags, who it is for and when to use it

Template preview

A template preview experience where users and sales teams could demonstrate an agent before installation. The preview surfaced the agent’s use case, required integrations, sample workflows and reasoning behavior, helping customers understand value before touching production data. It turned the marketplace from a static catalog into a guided evaluation surface, and resolved the single biggest drop-off point discovered in research.

Installation flow

The install flow was the highest-stakes UX in the product. It needed to collapse weeks of developer configuration into a guided 15-minute experience, without hiding complexity that power users needed.

Step 1: select deployment target, AI for Work, Service, or Process
Step 2: choose scope, from a single discrete task to a multi-step workflow
Step 3: connect tools, Gmail, Elasticsearch, LinkedIn, Salesforce, one-click install per integration

Private marketplace

For enterprise customers, the public marketplace was only part of the journey. IT and platform teams needed a way to create their own internal marketplace on top of Kore.ai’s agent ecosystem.

Manage templates: publish, unpublish, and version internal agent templates
Featured templates: curate a spotlight row for high-priority internal agents
Branding: company name, logo, banner and primary brand color
Security and compliance controls in the same admin sidebar as the public marketplace settings

Design system

With no pre-existing component library, the marketplace design system was built alongside the product: agent cards with primary and compact variants, category chips, integration badges, a three-step install stepper, a diagnostic panel, tailored empty states and a preview modal. After the marketplace shipped, the agent card component system was adopted across the broader Kore.ai platform.

Reflection

What worked: the confidence-before-commitment framing, progressive complexity, and treating the agent card as an atomic unit that scaled from the marketplace grid to platform-wide adoption.

Design challenges: four distinct personas with one interface, building and shipping a design system simultaneously, enterprise UX constraints that conflicted with simplicity goals, and no user testing budget.

What we learned

Trust is the core design constraint in enterprise AI. Users will not install what they do not understand, no matter how capable it is. The most important design work often happens before any pixels: the framing, the persona clarity, the hierarchy of problems. In enterprise software the job of design is not to make things look good, it is to make complex things feel inevitable.

Outcome

Multi-day developer configuration collapsed into a guided 15-minute experience
Enterprise teams could self-serve agent deployment without opening a support ticket
The agent card component system was adopted platform-wide
Template preview and integration transparency resolved the primary drop-off point

What we did

Product Design, Design Systems, User Research