Why Intercom, Appcues, and Pendo Don't Work for AI Agent Products (And What Does)
Firstflow Team
What actually changes when your product is an agent?
The onboarding problem itself doesn't change. Users still need to discover features, understand what the product can do, and find value before they leave. That hasn't changed.
But everything else has. The interface is a conversation, not a dashboard. There are no buttons to highlight, no menus to annotate, no pages to walk someone through. The agent responds dynamically to every user differently. And the signal you need to improve it — what users liked, what broke, what confused them — lives inside those conversations, not in click-through analytics.
Traditional onboarding tools were built for a world with a fixed UI. Agents don't have one.
If you're new to agent onboarding, start with our post on why AI agent products need onboarding to understand the basics.
Why Intercom Doesn't Work
Intercom is one of the most capable customer communication platforms out there. For SaaS products with dashboards, it's a reasonable choice. For agent products, it breaks down in a few important ways.
The widget is not the conversation. Intercom's onboarding features — product tours, banners, checklists — live in a separate widget or overlay on top of your UI. If your product is a chat interface, you're stacking one chat widget on top of another. Users are already in a conversation with your agent. Intercom interrupts that with a separate experience they didn't ask for.
Tours don't translate to conversations. Intercom's product tours work by pointing at UI elements — "click this button," "look at this section." A chat window doesn't have those elements. You can't highlight a part of an AI response. You can't draw an arrow to where the user should type next. The tour format fundamentally doesn't map to conversational interfaces.
Structured input collection is not its strength. Collecting qualification data, preferences, or onboarding information mid-conversation requires either a form (which breaks the chat experience) or a rigid bot flow (which feels out of place in an AI-first product). Intercom was designed for human support agents routing tickets, not for structured in-chat data collection at scale.
Response feedback doesn't exist. There's no native mechanism in Intercom to let a user rate a specific agent response — thumbs up or down, with a reason — and route that signal to your product team. The closest equivalent is a CSAT survey sent after the conversation ends, which captures sentiment but loses the per-response granularity you actually need to improve the model.
Issue reporting is a support ticket. When a user hits a broken flow in Intercom, they open a support conversation. That conversation is separate from the original context, goes to a support queue, and loses the in-conversation signal that would make it actionable.
The cost problem. Intercom pricing is built around seats and monthly active users. At scale, the costs compound quickly — and you're paying for a platform designed for human support teams, not agent product infrastructure.
Why Appcues Doesn't Work
Appcues is purpose-built for product onboarding. If you're building a traditional SaaS product, it's a solid tool. But it has a fundamental assumption baked in: that your product has a fixed UI.
It requires a surface to attach to. Appcues works by targeting DOM elements — specific buttons, modals, nav items — and overlaying tooltips, modals, and hotspots on top of them. A chat interface doesn't expose those elements in a meaningful way. There's no "step 1: click the submit button" in a conversation with an AI agent. The flows Appcues is designed to create simply don't have a place to live.
No concept of conversation state. Appcues flows are triggered by page URLs and user properties. They don't know what the user just said to your agent, what the agent responded with, or where the user is in an ongoing conversation. An in-chat onboarding experience needs to be aware of conversational context to feel natural. Appcues isn't built for that.
No feedback collection per response. Like Intercom, Appcues has no mechanism for per-message feedback. It can survey users at the end of a flow, but not capture a thumbs down on a specific agent response with an optional reason.
No issue reporting. Appcues has no concept of in-context issue reporting. If something breaks, the user leaves — and your team finds out through other channels, if at all.
Why Pendo Doesn't Work
Pendo combines product analytics with in-app guidance. It's a strong tool for understanding how users move through a traditional product. For agent products, the same structural limitations apply.
Overlay-based guidance doesn't work in chat. Pendo's tooltips, lightboxes, and walkthroughs are designed to appear over existing UI. In an agent product where the interface is a conversation thread, there's no relevant UI to overlay on.
Analytics assumptions don't hold. Pendo's analytics model is built around page views, feature clicks, and session behavior. Agents don't produce that kind of event data. The meaningful signals are in the conversation — what the user asked, how the agent responded, whether the response was helpful. Pendo's analytics layer doesn't capture or interpret conversational data.
No response-level feedback. Same limitation as the others. Pendo can run NPS surveys and gather aggregate sentiment, but not per-response like/dislike with reason capture at the point of interaction.
What to Look For
Agent-ready onboarding infrastructure should solve these problems by design. Here's what actually matters:
1. In-chat delivery by default Every onboarding experience — capability introductions, surveys, announcements, skill unlocks — should be delivered inside the conversation, not on top of it. No widgets, no overlays, no context switches.
2. Conversation-aware flows Flows should be able to trigger based on what's happening in the conversation: what the user asked, what the agent said, which features have been introduced. Not just page URLs and user properties.
3. Structured input collection mid-conversation Step-by-step question flows, preference surveys, and qualification sequences that feel native to the chat experience — not forms embedded in a widget.
4. Per-response feedback Users can rate any agent response with a like or dislike, and optionally provide a reason. Signal captured at the moment of interaction, not in a post-conversation survey.
5. In-context issue reporting Users can flag a problem mid-conversation without leaving the chat. The report captures full conversation context automatically and routes to the right team.
6. Measurable Engagement, completion rates, drop-off, and feedback trends across every flow — so teams know what's working.
Comparison
| Feature | Intercom | Appcues | Pendo | Firstflow |
|---|---|---|---|---|
| In-chat delivery | No | No | No | Yes |
| Agent-aware flows | No | No | No | Yes |
| Structured input collection | Limited | No | No | Yes |
| Per-response feedback | No | No | No | Yes |
| In-context issue reporting | Separate widget | No | No | Yes |
| Progressive capability unlock | No | Limited | No | Yes |
| Pricing model | Per seat | Per MAU | Per MAU | Usage-based |
The Bottom Line
Intercom, Appcues, and Pendo are good tools. They work well for the products they were built for. But they were built for a world where the interface has fixed elements you can point at, where user behavior is measured in clicks and page views, and where onboarding is something you layer on top of the product after the fact.
Agent products are different. The interface is a conversation. The behavior is a message. The onboarding has to live inside the experience, not on top of it.
These tools weren't designed for autonomous conversational products. Firstflow is.
Get started with Firstflow · Read the docs
FAQ
Can I use Intercom alongside Firstflow? Yes. Firstflow handles in-chat onboarding, feedback, and issue reporting. If you're using Intercom for human support routing or external messaging, those functions are separate and can coexist. Firstflow handles the experience layer inside your agent product.
What if my product has both a dashboard and a chat interface? Traditional tools like Appcues or Pendo may still be useful for the dashboard parts of your product. Firstflow is specifically designed for the conversational parts — wherever the agent is the primary interface.
Do I need to rip out existing tooling to use Firstflow? No. Firstflow is additive. It sits alongside your agent as the experience layer for in-chat flows. You don't need to remove anything else to get started.