On September 22, 2026, Pendo announced Pendo for Agents. If you run onboarding or support for a B2B SaaS product and you have a Pendo contract, a product tour library, or an AI support agent in flight, the announcement changes the shape of the decision in front of you, even if you never buy it. This post explains what Pendo for Agents is, what it is not, and what it means for in-app onboarding specifically. Everything stated here about Pendo was checked against Pendo's own website, documentation and announcement on October 5, 2026, and is described as of that date. Pendo will keep changing the product, so treat this as a snapshot.

The short version: Pendo has decided not to be the agent. It has decided to be the thing the agent plugs into.

What is Pendo for Agents?

Pendo for Agents is a pair of capabilities that give AI agents built or bought by a company access to Pendo's product usage data, and that measure how those agents perform. As of October 2026, Pendo describes it as providing "product context and observability to the agents you build or buy." It consists of Agent Toolkit, which feeds behavioral context into an agent, and Agent Analytics, which reports on agent usage and outcomes. Pendo's own documentation lists Agent Toolkit as an open beta and Agent Analytics as generally available.

Pendo's launch post frames the problem it is solving as agents that work technically but fail commercially because they lack context. Pendo says it looked at 47 million agent conversations across roughly 500 companies and concluded that "agents have no idea who they're talking to." It calls the result "zombie agents": deployed, answering, and blind to who the user is and what they were doing before they asked.

That diagnosis is correct, and it is worth sitting with before looking at the fix. A support agent that can only see the chat transcript is guessing at the one thing that determines whether its answer lands: where the user actually is in the product and what they have already tried.

What Agent Toolkit does

Agent Toolkit connects product usage data to an agent so the agent can use it in a conversation and, in some cases, act on it. As of October 2026, Pendo's product page says the toolkit works with Fin, Claude, ChatGPT, Gemini, Salesforce, Notion, Decagon, Sierra, and custom-built agents. Pendo's support documentation lists five things the toolkit lets an agent do: surface relevant Pendo guides in response to a visitor's question, draw on the visitor's session context when responding, submit visitor feedback to Pendo, re-engage visitors who drop off from a workflow, and respond to friction signals such as rage clicks or errors.

The launch post adds detail on the context itself: session history of user actions, "frustration triggers: rage clicks, dead clicks, U-turns, and errors," workflow nudges for incomplete processes, and guide surfacing to relevant walkthroughs. Pendo says the toolkit is in open beta and free to existing Pendo users during that beta.

What Agent Analytics does

Agent Analytics measures the agent rather than informing it. Pendo's documentation says it tracks prompt volume, unique visitors and accounts, retention, and reactions over time, detects patterns in what visitors ask, surfaces problems, shows full conversation transcripts alongside session replay, and compares agent versions and configurations. The product page summarizes the metrics as "session volume, resolution rates, and task completion." The idea is that an agent should be judged by whether users came back and completed what they set out to do, not by token counts or latency.

What Pendo for Agents is not

Pendo for Agents is not an AI agent. As of October 2026, nothing on Pendo's product page, documentation or launch post describes Pendo supplying the conversational agent itself for customer-facing support or onboarding. The agent is something you build or buy elsewhere; Pendo supplies context to it and measurement of it. Pendo's own case example makes this explicit: the Teachable story in the press release is about Intercom's Fin, which Teachable's Kathleen Ross describes as being "powered by Pendo."

This matters because the phrase "Pendo now does AI agents" is already circulating, and it is not what the product is. If you need an agent in your product that onboards users, answers their questions and completes setup with them, Pendo for Agents does not give you one. It gives you a way to make an agent you acquire separately less blind.

It is also separate from Novus, which appears on Pendo's pricing page as of October 2026 as a free open beta aimed at "teams shipping with AI who want proactive product insights, delivered where they work," with continuous product monitoring, recommendations delivered in Slack, and auto-instrumentation through GitHub. Novus is about instrumentation and insight for product and engineering teams. It is not an onboarding or support agent either.

What is Pendo conceding with this launch?

Pendo for Agents concedes two things that teams running in-app onboarding have known for a while. First, that scripted guides, tours and resource centers do not resolve the questions users actually have, which is why an agent is now required at all. Second, that an agent without live product context is not good enough, which is why Pendo believes it has a role in the agent era. Both concessions are in Pendo's own framing: the toolkit exists to "surface guidance right when users need it" and to let agents "adapt onboarding in real time."

Look closely at the first of the toolkit's five capabilities: surfacing relevant Pendo guides in response to a visitor's question. The agent becomes a better router to the same guide library you already have. If that library is stale, incomplete, or built for a version of the UI that shipped two quarters ago, the agent now surfaces stale, incomplete, outdated guides more efficiently. Pendo for Agents does not change the maintenance problem that comes with authored content; we wrote about why that content goes stale in a separate post, and nothing here alters that economics.

The second concession is the more interesting one for onboarding. Pendo is saying that the valuable thing is knowing where the user is, what they clicked, where they rage-clicked, and which workflow they abandoned. That is correct. The question a buyer should ask is whether that knowledge is best delivered to the agent as a feed from an analytics vendor, or whether the agent should have it natively because it lives inside the product.

How does a bring-your-own-agent stack work in practice?

A Pendo for Agents deployment for onboarding involves at least three parties: Pendo, which holds the behavioral data and the guide library; an agent vendor, which holds the conversation, the knowledge base connection and the model; and your team, which wires the two together and keeps both configurations in sync. Each layer is configured separately, each has its own admin, and the behavior a user sees is the product of all three.

Here is what that looks like for a concrete onboarding moment. A new admin on a trial has connected one data source, started the second, and stopped for four minutes on the permissions screen. In a bring-your-own-agent stack, Pendo registers the dead clicks and the abandoned workflow. If the toolkit is wired to your agent, and your agent is configured to act on friction signals, the agent can open a conversation and pull the relevant Pendo guide. Whether the agent can then finish the permissions step for the user depends on what actions the agent vendor supports in your product, not on Pendo.

This architecture has real strengths. If you already pay for Pendo analytics and already run Fin, Decagon or Sierra for support, the toolkit is the cheapest way to make the agent you have less blind, and Agent Analytics gives you a measurement layer that the agent vendors do not all provide on their own. If your product team lives in Pendo dashboards, keeping the behavioral data there is reasonable.

It also has costs that do not appear on the launch page. Two configurations drift. The guide library still has to be maintained by hand. The friction signals the agent can respond to are the ones Pendo instruments, which means the agent's awareness is bounded by your tagging discipline. And the agent vendors Pendo names are, as of October 2026, primarily support agents: Decagon's and Sierra's own sites describe chat, voice, email and messaging channels, and neither lists an in-product onboarding surface, Slack or Teams on the homepages we checked this month (we compared the two in more depth in Decagon vs Sierra). Onboarding is being handled by routing a support agent toward a product-analytics feed, not by an agent built for onboarding.

How does an AI onboarding agent differ from Pendo plus a support agent?

An AI onboarding agent reads the screen and the account state directly, decides what each user needs next, answers from company knowledge, and takes permitted actions inside the product. Context is not piped in from a separate analytics vendor; it is what the agent is standing on. Worknet is built this way, and it is the architectural difference that matters most when comparing it with a Pendo for Agents stack.

Three practical differences follow from that.

One configuration instead of three

With Worknet, a success or onboarding lead writes the goal in plain English, for example "get every new admin to a connected data source and an invited teammate in the first session," and the agent builds and ships the guidance in the product. There is no guide library to author and no separate toolkit mapping between an analytics tool and an agent. When the UI changes, there is no walkthrough to re-record, because the agent reads the UI as it is.

Context that is native, not instrumented

Pendo's toolkit can tell an agent about the friction signals Pendo has been configured to capture. An agent that is in the product sees the permissions screen, the half-finished form, the error banner and the account's plan and role without a tagging plan behind it. Both approaches can act on a stalled user. Only one of them depends on someone having tagged that screen correctly last quarter.

Coverage beyond the app

Pendo for Agents is about agents inside your product's web experience. Worknet runs the same agent in-app and in Slack and Microsoft Teams, connected to Salesforce, Zendesk, HubSpot and other systems through API and MCP connections, so the user who asks a setup question in a shared Slack channel gets the same agent with the same context as the one who asks inside the product.

To be fair about where the comparison does not favor Worknet: if your core need is agent measurement across several agents you already run, Agent Analytics is a dedicated product for that and Worknet is not. If you want to keep Pendo as your analytics system of record, you can run Worknet alongside it; teams typically keep the analytics and let the agent take over guidance, onboarding and in-app support, which is also how we framed the migration path in our guide to moving from Pendo guides to an agent.

On cost, neither side gives you a number to compare. As of October 2026, Pendo's pricing page shows "Request pricing" for its Base, Core and Ultimate plans and lists Agent Analytics as an add-on, Agent Toolkit is free during its open beta, and Worknet is quote-based too. This post makes no claim about which stack is cheaper, because that depends on what you already own and what you are retiring.

What should onboarding and support leaders do about Pendo for Agents?

If you are mid-contract with Pendo and already run a support agent, test the toolkit during the open beta and judge it on one measure: how often the agent resolves an onboarding question without sending the user to a guide that then needs a human anyway. If you are evaluating what onboarding should look like for the next two years, treat the launch as confirmation that the category has moved from authored guides to agents, and decide whether you want to assemble the agent out of parts or deploy one that already has product context.

Four questions sort that out quickly.

First, who owns the onboarding experience day to day? If it is a product team that lives in Pendo and already has an agent vendor relationship, assembly is tractable. If it is a success or support team without engineering time, a single agent they configure in plain English is the realistic option.

Second, how often does your UI change? Every change is a maintenance event for a guide library, and Pendo for Agents routes to that library. An agent that reads the live screen does not inherit that cost.

Third, where do your users actually ask for help? If a meaningful share of onboarding questions arrive in Slack Connect channels, shared Teams channels or email, an in-app-only context feed covers less of the surface than it appears to.

Fourth, what does the agent need to be able to do, not just say? Finishing a setup step, inviting a teammate, or changing a permission with the user's consent is what turns an answer into a completed onboarding. Check whether that capability exists in the agent vendor you would pair with Pendo, because the toolkit itself supplies context and guide surfacing, not the actions.

Conclusion

Pendo for Agents is a clear statement from one of the digital adoption category's incumbent vendors that in-app onboarding is moving from scripted guides to agents, and that agents need product context to be worth deploying. Pendo's answer is to supply that context and measurement to an agent you source elsewhere. That is a coherent strategy for companies that already own Pendo and an agent, and a three-vendor assembly for everyone else. An AI onboarding agent that lives in the product, reads context natively and acts on it collapses that assembly into one thing a success team can run.

If you want to see what that looks like inside your own product, read how an AI adoption agent compares with Pendo or book a demo and bring the onboarding step your users stall on most.