# The Four Levels of Agentic Product Adoption

> Ramli John · Thought Leadership

Five years after Product-Led Onboarding, the playbook has changed. Ramli John maps the four levels of AI-assisted adoption — AI that tells, guides, generates, and learns — and why most products are at Level 1 calling it Level 4.

*Guest post by Ramli John — Founder of Delight Path and author of* Product-Led Onboarding.

Five years ago, I wrote *Product-Led Onboarding*. The playbook was about removing friction, designing for the "aha" moment, and getting users to value as fast as possible. The tools were checklists, tooltips, empty states, and well-timed emails.

I don't think that playbook works anymore.

Claude, ChatGPT, and tools like them have fundamentally shifted what people expect from software. Users arrive at your product already conditioned to just describe what they want and have something happen. A checklist feels like homework. A tooltip feels like a speed bump. The bar has moved.

Through the Product Leaders Lab, I've spent the last two years in hundreds of conversations with Directors, Heads, and VPs of Product. One question keeps coming up: how do we actually use AI to accelerate product adoption? Not just add a chat widget. Not just automate a welcome email. Actually move the needle on time-to-value.

What I've noticed is that most teams are using AI for adoption. They're just using it at the wrong level.

Here's a way to think about it. Every click, every "here's how to do that, now go do it yourself" moment, is a tax on your user's attention and patience. Call it **click tax**, a term coined by Mickey Alon at Foldspace: the cumulative effort a user must spend between knowing what they want to do and actually doing it.

![The click tax: a user's intent has to pass through click, navigate, configure, apply and review before it reaches an outcome. Every step between intent and outcome is the tax.](/blog/the-four-levels-of-agentic-product-adoption/click-tax-path.png)

There are four levels of AI-assisted product adoption. Each one removes more friction than the last. The four levels are really a spectrum of click-tax reductions. As you move up, the user's job gets smaller. The AI's job gets bigger. At Level 1, the user carries everything. At Level 4, the user only has to say what they want.

Most products are stuck at Level 1. The companies pulling ahead are moving toward Level 4, where AI doesn't just answer questions or point users in the right direction. It does the work for them.

## Level 1: AI that tells

**What it is:** An AI chatbot connected to your help docs and knowledge base. The user asks a question. The AI answers it.

This is where most B2B SaaS products are right now. Connect a support chatbot to your documentation and you've got Level 1. It handles repetitive support questions, surfaces relevant articles, and reduces ticket volume. Not nothing.

![Intercom Fin can answer customer questions](/blog/the-four-levels-of-agentic-product-adoption/intercom-fin-chatbot.png)

**The problem:** It describes the path but doesn't walk the user down it.

A user who's confused about how to set up a workflow asks the chatbot. The chatbot explains the steps. The user then has to find the right screen, navigate there, and execute the action themselves. The knowledge gap closes. The friction doesn't.

This is what's happening when users immediately ask an AI chatbot "what can you do?" or start asking it to perform tasks. They've been conditioned by Claude, ChatGPT, and other tools to expect something more. A knowledge base in a chat window no longer meets that expectation.

**Click tax:** High. The user still has to find, navigate, and execute everything themselves.

## Level 2: AI that guides

**What it is:** Contextual AI that doesn't just answer. It launches the next step.

Instead of telling a user where to go, Level 2 AI opens the right screen, pre-fills the state, and puts the user one action away from done. The AI is action-adjacent. It bridges the gap between instruction and motion.

Optibus, a Foldspace customer, is a good example of this in a complex domain. For transportation planners navigating a sophisticated platform, the AI helps users orient themselves contextually, surfacing the right next step based on what they're trying to accomplish, not just answering what they asked.

![The Optibus agent alongside a timetable report, listing what it can do: answer questions, navigate the application, filter scheduling Gantt charts, and diagnose dataset imports.](/blog/the-four-levels-of-agentic-product-adoption/optibus-agent-timetable.png)

**Why this matters for adoption:** Navigation is a hidden adoption killer. Users often know what they want to accomplish. They just don't know where to go. Level 2 AI removes the navigation burden without taking over the decision-making.

**Click tax:** Medium. The user still executes, but the AI eliminates the wayfinding.

**What to watch for:** Level 2 still requires the user to follow the guidance and take action. If the steps are complex or the user loses momentum between the prompt and the destination, you're still losing people.

## Level 3: AI that generates

**What it is:** AI that creates something inside the product based on the user's description. The user doesn't navigate to a form and fill it out. They describe what they want, and the AI builds it.

![The generative path: user intent goes straight to a prompt and then to the outcome, with the click, navigate, configure, apply and review steps collapsed entirely.](/blog/the-four-levels-of-agentic-product-adoption/agentic-path.png)

Mixmax, a Foldspace customer, does this for sales teams. Instead of manually building outreach sequences step by step, a rep describes the campaign they need and Mixmax generates the email sequences and messaging directly. No blank canvas. No guessing at structure. The rep reviews, adjusts if needed, and launches.

The results back it up:

- **Week-one engagement went from 46% to 79%**, and it held past the first session, so it was a real behavior change, not a launch-day spike.
- **Users who engaged with the agent were 4.4x more likely to activate** than those who did not.
- **Time to paid dropped from 5 days to 2.6.**

Mixmax's growth lead walked through the full numbers, and the anti-bias tests behind them, in a workshop with Foldspace and ProductLed. The recap is here: [Turn Signups Into Revenue: The New AI Onboarding Playbook](/blog/turn-signups-into-revenue).

![The Mixmax agent rewriting a multi-stage outreach sequence, listing the personalization, clarity, and engagement changes it made before the rep applies them.](/blog/the-four-levels-of-agentic-product-adoption/mixmax-sequence-agent.png)

**Why this matters for adoption:** The blank state is one of the biggest adoption killers in B2B SaaS. Users sign up, see an empty dashboard, and freeze. Level 3 collapses that moment. The AI populates the product with something real before the user has had a chance to disengage.

**Click tax:** Low. The user describes intent. The AI handles creation. The user's only job is to review.

## Level 4: AI that learns

**What it is:** The agent doesn't just generate. It improves. Every interaction makes the output better for the next user.

This is where the compounding begins. The AI isn't just reducing friction in the moment. It's getting smarter about how to reduce it better next time. The product reshapes itself based on real user intent, not assumptions.

Level 4 is not simply the top rung of the same ladder. Levels 1 through 3 measure how much of the job the AI takes on. Level 4 measures whether the system improves. A product can sit at Level 3 forever and learn nothing.

Lovable is a clear example. Every website generated through the platform feeds back into the model. The next user gets a better output than the last one. Not because a designer improved the template, but because the system learned from what actually worked.

![Lovable improves with more usage: a chat thread on the left builds and refines a landing page rendered live on the right.](/blog/the-four-levels-of-agentic-product-adoption/lovable-generative-editor.png)

**Why this matters for adoption:** This is the closed loop that most product teams don't have. Right now, you're guessing at user intent based on clicks and drop-off points. At Level 4, you know what users are actually trying to accomplish, because they told the agent. That signal is far richer than any heatmap. **Clicks tell you where users stopped. Prompts tell you what they wanted.** You can iterate faster, deploy without a release cycle, and design for outcomes rather than assumptions.

**Click tax:** Near zero, and declining. The user's job shrinks with every interaction.

One caveat. A loop without evaluation is not a flywheel. If you are not measuring whether the output matched what the user actually asked for, you are just getting faster at delivering the wrong thing. Golden sets, shadow runs, and outcome-matching scores are what make the loop directional. That evaluation layer is the real cost of reaching Level 4, and it is the part most teams skip.

## Where most products are stuck

Level 1 is easy to ship. Connect a chat widget to your help center and you're done in a day. Level 2 requires integrating your AI layer with your product UI. Level 3 requires rethinking your empty states and first-run experiences. Level 4 requires rethinking what your product does.

Most teams treat Level 1 as "done with AI" and move on.

The result: users get answers but not outcomes. They know what to do. They still don't do it.

The question worth asking isn't "do we have AI in our product?" It's "how much click tax are we still charging our users?" And more importantly: what would it take to move up one level?

You don't have to go straight to Level 4. Moving from Level 1 to Level 2 is already a meaningful improvement. But you have to be honest about where you are.

Most products are at Level 1 calling it Level 4.

## The bigger picture

Five years ago, the best onboarding teams were obsessed with reducing steps. Fewer clicks to first value. Shorter time to the "aha" moment. That thinking was right, and it still is.

But the ceiling has moved. The question is no longer just "how do we reduce clicks?" It's "how do we eliminate the need for clicks altogether?"

That's what the shift to agentic adoption is really about. Not AI as a feature. AI as the thing that does the work your users were never going to do on their own.

The products that figure this out won't just have better onboarding metrics. They'll have a fundamentally different relationship with their users.

*Ramli John — Founder, Delight Path*