Lean Inception in the Age of AI

Use AI to enrich the conversation, not to eliminate the conversation

Paulo Caroli

Artificial intelligence can already generate product visions, personas, user journeys, features, MVP Canvases, backlogs, code, and prototypes in minutes.

That is very useful.

But it also creates a risk: confusing a well-written artifact with a truly aligned team.

This guide complements the book Lean Inception: How to Align People and Build the Right Product. It was created to help facilitators, product leaders, and team members incorporate artificial intelligence into Lean Inception without losing what makes the workshop valuable: real conversations, explicit disagreements, shared decisions, and collective learning.

The goal is not to automate a Lean Inception.

It is to use AI to prepare the work better, challenge assumptions, explore alternatives, reduce operational tasks, and capture learning — without outsourcing judgment or responsibility for decisions.

Use AI to enrich the conversation, not to eliminate the conversation.

AI can organize, summarize, provoke, compare, and suggest.

People remain responsible for context, prioritization, trade-offs, alignment, and decisions.

Who this guide is for

• facilitators running in-person, remote, or hybrid Lean Inceptions;
• product leaders who want to incorporate AI into team practices;
• members of cross-functional teams who already know or have been exposed to Lean Inception;
• organizations building AI-enabled products or incorporating agents into their products and workflows.

The examples and prompts are starting points. Adapt them to your context, available data, and your organization’s policies.

This material does not replace the Lean Inception book, research with real people, facilitation, MVP validation, or common sense.

1. AI accelerates the work. Lean Inception improves decisions.

Artificial intelligence has changed the workflow of product teams.

Tasks that once took hours can now happen in minutes. We can start a discussion from a draft instead of a blank page, organize large volumes of information, create code, and reach prototypes much faster.

That acceleration is real.

But it does not automatically solve one of the main challenges of a Lean Inception:

How do people with different knowledge, perspectives, and responsibilities build shared understanding and decide what is worth validating first?

AI is very good at generating plausible answers and well-structured artifacts.

Lean Inception helps the team question those answers, confront them with context, reveal differences in understanding, and collectively own the decisions made.

The new bottleneck

AI is dramatically reducing the time between an intention and something concrete.

A Product Vision can be drafted in seconds. A prototype that once took days can be ready in hours. Code, documentation, and solution alternatives can be produced increasingly fast.

But that does not eliminate the hard part.

Deciding what to build, why to build it, what to leave out, and what we need to learn first still requires context, conversation, and judgment.

The faster we can build, the more important it becomes to choose well.

That is where Lean Inception becomes even more relevant.

It does not exist to accelerate artifact production. It helps the team build shared understanding, make choices, and define the smallest path to learning.

AI accelerates building. Lean Inception helps the team decide what deserves to be built.

Alignment is not an artifact

Products do not emerge from good answers alone.

They emerge from conversations, disagreements, negotiations, experiments, and adjustments.

Business, user experience, and technology perspectives see different risks and opportunities. The quality of a decision depends on the team’s ability to combine those perspectives.

AI can synthesize different opinions into an elegant sentence.

But an elegant sentence does not prove that people agree.

It may even hide a conflict the team needed to make visible.

Example: when the sentence was not the problem

In a Lean Inception for a financial institution, the team was working on a credit card initiative.

There were already presentations, approval documents, and an initial narrative. At first glance, everyone seemed aligned.

During the collaborative Product Vision activity, the conversation stalled.

Some people argued that the branch manager should have autonomy to define the initial credit limit. Others argued that the decision should be automatic, based on data and APIs.

AI could easily produce a sentence combining both points of view.

But the problem was not finding a better sentence.

The two positions represented different operating models, experiences, and strategies.

The activity made the disagreement visible and revealed that other people needed to participate in that decision.

The value of the activity is not only in the final artifact. Often, it is in the conversation required to get there.

2. What has not changed in Lean Inception

Lean Inception remains a collaborative workshop to align a group of people around a proposed solution and an MVP strategy.

AI changes preparation and the speed at which some things can be produced.

It does not change the purpose of the workshop.

Lean Inception still helps the team:

• build shared understanding about the product;
• make perspectives and assumptions visible;
• explore personas, user journeys, and features;
• analyze business, user experience, and technology perspectives;
• prioritize and sequence;
• define the MVP and what needs to be validated first.

What changes is the speed.

What does not change is the need for the team to talk, make choices, and reach a shared MVP strategy.

Keep the conversation visible

One of the strengths of Lean Inception is making information, choices, and disagreements visible to everyone.

This principle becomes even more important with AI.

When each participant talks privately with their own tool, we may create several private reasoning paths and share only the conclusions.

During the workshop, prefer uses where the input, the AI response, and especially the group’s review are visible.

AI-generated content should enter as material for discussion, not as authority.

A suggestion is a suggestion.

A hypothesis remains a hypothesis.

And a summary does not mean consensus.

Facilitation changes too

The facilitator remains responsible for creating structure, pace, and conditions for participation.

In the age of AI, an additional responsibility appears:

Do not let the ease of producing artifacts be confused with clarity or alignment.

Sometimes the best intervention will be to ask AI for help.

At other times, it will be to put AI aside and let people talk.

3. Before, during, and after Lean Inception

A simple way to think about AI use is to look at three moments: before, during, and after the Lean Inception.

Before: prepare a better conversation

Before the workshop, AI can significantly reduce operational work.

It can help organize documents, research, indicators, constraints, previous decisions, and open questions.

The goal is not to turn everything into a single version of the truth.

It is to arrive at the workshop with a better understanding of:

• what we know;
• what we believe;
• what has already been decided;
• what remains open;
• where contradictions exist;
• what the team needs to discuss.

From requesting an artifact to improving the conversation

Throughout this guide, you will find a few prompt examples.

The important difference is not in writing more sophisticated prompts.

It is in the intent.

A prompt can ask AI to produce the result.

Or it can ask AI to help the team think better about what needs to be discussed.

In the first case, we want a summary.

In the second, we want to arrive at the conversation better prepared.

That is the logic behind the prompts in this guide.

Who needs to be in the conversation?

AI can also help challenge the composition of the group.

• Who deeply understands the problem?
• Who understands the users?
• Who understands the technical and operational constraints?
• Who will be affected by the decision?
• Who needs to participate but is not in the room?

The tool can suggest missing perspectives.

The decision about who should participate still depends on the organization’s real context.

Agenda: adapt without eliminating conversation time

The first Lean Inceptions usually took place over a full week.

That already represented a major change. Before that, inception workshops could last much longer. Bringing business, user experience, and engineering perspectives together over a few days, focused on an MVP strategy, created significant acceleration.

Over the years, many teams experimented with other formats: fewer days, distributed sessions, remote meetings, and different combinations of activities.

That is natural.

Lean Inception does not need to be reproduced exactly as we practiced it a decade ago.

But there is a difference between adapting the agenda and eliminating the conversations that make Lean Inception work.

AI can help with that adaptation.

A huge prompt that generates every artifact in minutes does not turn a bad agenda into a good Lean Inception.

The agenda can change.

The space required for conversation still needs to exist.

What about the data?

Before sharing information, check your organization’s policies and the tool being used.

Strategic documents, interviews, contracts, personal data, regulated information, or architecture details may require additional care.

Anonymize when possible, work with authorized tools, and tell participants when AI is transcribing or analyzing the session.

During: use AI as a thought partner

During the Lean Inception, AI can play four useful roles:

• Prepare, by retrieving information or clarifying terms;
• Provoke, by challenging assumptions and bringing alternative perspectives;
• Organize, by grouping items and making differences clearer;
• Capture, by documenting decisions, questions, and pending topics.

The concern begins when AI starts shortening exactly the path the team needs to take to reach a decision.

AI can accelerate the conversation. It should not replace the path required to reach a decision.

A simple dynamic helps:

1. the team brings its context and perspectives;
2. AI is used to expand, challenge, or organize;
3. the group critiques the response;
4. people change, discard, or incorporate parts of it;
5. the final decision remains with the team.

For example:

We are not asking AI to close the conversation.

We are using AI to see the conversation we had more clearly.

And not every activity needs AI.

If the group is engaged and the tool would only add distraction, continue without it.

Supporting elements during Lean Inception

Parking Lot: from sticky notes to a group assistant

The Parking Lot has always been a simple way to capture important topics without allowing them to derail the current conversation.

Before, it might have been just an area of the canvas with a few sticky notes.

Now AI can take on this memory and organization work for the group.

It can capture topics, preserve the context in which they appeared, and help quickly retrieve what still needs to be discussed.

AI takes care of memory and organization.

The group remains responsible for deciding when to return to the topic, who needs to participate, and what to do with it.

Glossary and ubiquitous language: the connection between Lean Inception and DDD

A Lean Inception is also an important moment to align the language used by the group.

Words such as “customer,” “user,” “order,” “account,” “contract,” or “approval” may carry different meanings for business, user experience, and engineering.

That is why the glossary is not just documentation.

It helps the team make explicit a shared language about the domain.

There is a natural connection here with Domain-Driven Design (DDD) and the idea of ubiquitous language: using terms that make sense to people who know the domain and keeping those meanings consistent across the product and software.

This is especially relevant because Lean Inception is not used only when a product starts from scratch.

Many teams use the approach to evolve existing products, parts of large systems, and legacy environments.

In those cases, the glossary must also recognize and respect the domain language that already exists in the organization — in the business, processes, code, APIs, events, data, and domain models.

If the language is confused, AI may simply accelerate the spread of that confusion.

It can also help the team notice it:

AI can help preserve consistency.

But it does not own the domain language.

The faster we turn language into software, the more important it becomes to know exactly what those words mean.

After: accelerate without creating false certainty

After the Lean Inception, AI can help with consolidation.

Notes, photos, boards, and records can be organized quickly. Decisions can be documented. Pending topics and risks can be grouped.

The concern is not to let AI “improve” the material to the point of erasing uncertainty.

A decision that is still pending should remain pending.

A hypothesis does not become a fact because it is well written.

A disagreement does not disappear because the summary is elegant.

From MVP Canvas to prototype

AI is also dramatically reducing the time between an idea and a working prototype.

In one Lean Inception, two developers used AI agents after the Thursday session and arrived at the Showcase with something working, ready for an A/B test.

That changed the quality of the conversation with stakeholders. They could interact with something concrete.

But one thing did not change:

the time required to learn from real people.

AI can reduce the time between an idea and a prototype. It does not eliminate the time required to learn whether that idea actually works.

AI can also check consistency between the Product Vision, Feature Sequencer, MVP Canvas, and initial backlog.

But it does not need to turn that immediately into a complete backlog.

AI can accelerate the transition from Lean Inception to execution.

The goal remains to learn early — not to plan everything in advance.

4. AI in Lean Inception activities

There is no single correct way to use AI in each activity.

The most useful question is:

Where in this activity can AI help without taking away the conversation that makes the activity valuable for the team?

Kick-off

AI can help prepare the Kick-off by organizing context, decisions already made, constraints, and terms that need clarification.

But the Kick-off is not just information transfer.

It is the moment to make sure people understand why they are there and the context of the initiative.

If there are different interpretations of the workshop’s purpose, it is better to discover them now than to hide them behind a well-prepared presentation.

AI can prepare the conversation.

It should not create the impression that initial alignment has already happened.

Product Vision

AI can test clarity, identify vague words, or show different interpretations of the same statement.

But the activity remains a conversation about direction, audience, problem, and value.

When conflicting visions exist, do not ask AI to reconcile them too early.

The conflict may be exactly what the team needs to discuss.

The Product Is – Is Not – Does – Does Not Do

This activity helps the group make the product boundaries explicit.

AI can identify contradictions, vague terms, and items that appear incompatible.

It can, for example, ask:

But the boundaries need to be discussed by people.

The activity helps the team say no and align expectations.

A complete AI-generated table does not mean those boundaries are agreed.

Personas

AI can organize existing data and help separate behavioral patterns from hypotheses.

It can also create a very convincing persona even when there is almost no evidence.

That is the risk.

A well-written persona does not replace contact with real people.

User Journeys

AI can point out missing steps, exceptions, actors, emotions, dependencies, and friction points.

That can enrich the discussion.

But an overly clean journey can also hide real behavior.

Features Brainstorming

AI can expand the team’s repertoire and help reveal alternatives that did not appear in the first round.

But the goal is not to produce the largest possible list.

With AI, it becomes even easier to turn any possibility into a feature.

So use the tool to challenge the list itself:

AI can expand and organize possibilities.

The team decides which ones deserve to move forward.

Technical, UX, and Business Review

In this activity, different perspectives help make risks, questions, and uncertainty visible before Feature Sequencer and MVP decisions.

AI can help each person explore questions, surface dependencies, and investigate possibilities.

But the markings still need to represent the assessment of the people who understand the context.

AI tends to produce an answer even when there is little information. That can create a false sense of confidence exactly where the team should be signaling uncertainty.

A feature marked red or yellow is not a failure.

Often, it represents a conversation that still needs to happen.

So do not use AI to quickly turn questions into answers and convert everything to green.

Each person can use their tools to explore technical, user experience, or business questions.

But they should not delegate the final assessment.

There is another important concern: this is a Lean Inception, not a session to write a detailed specification for everything.

We do not need to resolve every uncertainty now.

We need to make uncertainty visible, understand which uncertainties really matter, and decide how to move forward.

Some questions will require research. Others, a technical spike. Others, a conversation with someone who is not in the room. Some can only be answered once we start validating the MVP.

The goal is not to turn every uncertainty into an answer. It is to understand the important uncertainties well enough to align the next steps.

Lean Inception is the starting point, not the final specification.

Feature Sequencer

The uncertainties made visible in the previous review help inform the Feature Sequencer.

AI can explore dependencies and present alternatives, but it should not erase the very risks we just made visible.

For example:

AI does not deliver “the answer.”

It creates material for a better conversation.

The team decides the sequence.

MVP Canvas

AI can quickly fill in the blocks of an MVP Canvas.

That does not mean it should.

The value of the Canvas is in helping the team make choices about the proposal, personas, journeys, features, expected results, and what needs to be validated.

AI can be especially useful for challenging an MVP that is growing:

At a time when building has become cheaper, slicing remains a strategic decision.

Showcase

AI can help organize the Showcase and accelerate the creation of what will be demonstrated.

But a better Showcase is not necessarily one with more complete slides or a more sophisticated prototype.

It is one that helps people understand:

• the direction chosen;
• the reasoning behind the MVP;
• what was deliberately left out;
• what we can already demonstrate;
• what we still do not know;
• what we want to learn next.

Use AI to make that story clearer.

Not to turn hypotheses into certainty.

Extension: when the product includes AI agents

When a Lean Inception deals with a product that uses agents, some additional questions may become relevant:

• Who requests the action?
• Who executes it?
• Who supervises it?
• What context does the agent receive?
• What can it decide on its own?
• Which actions require human confirmation?
• Who is accountable when it makes a mistake?
• How do we evaluate reliability, safety, and transparency?

These questions can be incorporated into existing activities.

But it is worth making explicit when we are introducing an extension to the approach.

AX — Agent Experience

It may also make sense to extend the Technical, UX, and Business Review with a fourth perspective: AX — Agent Experience.

If the agent is an important part of the product, ask:

• Is the environment in which it operates stable and predictable enough?
• Are the prompts, instructions, and context it receives well structured?
• Can it operate reliably in real-world situations?
• Where does it require human supervision or confirmation?

The idea is not to create another bureaucratic assessment. It is to make specific risks in the agent’s experience and operation visible before defining the MVP.

To go deeper into this adaptation, see Lean Inception for AI Agents.

An agent can be an important actor in a journey.

That does not mean every automated entity should become a persona.

5. What we should not delegate to AI

AI can summarize positions, present criteria, make trade-offs more visible, and explore alternatives.

But consensus, prioritization, and choice still belong to people.

We should also not use AI to make conflicts and uncertainties disappear. A convincing answer can artificially close a conversation that still needs to happen.

And there is something no simulation replaces: validation in the real world.

Synthetic personas, analyses, prototypes, and simulations can accelerate learning.

Evidence and real behavior still matter.

AI can recommend.

The team remains responsible for the decision.

Antipatterns to watch

In the book Teams Still Matter, I describe antipatterns that are emerging or being amplified by AI in product work. The five below have an especially direct relationship with Lean Inception.

AI-Generated Alignment

The Product Vision is well written. The MVP Canvas looks consistent. The presentation tells a clear story.

But people did not go through the conversations required to get there.

Disagreements remain hidden and the alignment is only cosmetic.

In Lean Inception, this may be the most dangerous antipattern.

Fake Discovery

Personas, user journeys, and analyses can become sophisticated very quickly.

But discovery is not about producing good artifacts.

It is about reducing uncertainty through learning.

When a team replaces contact with people and real evidence with plausible AI-generated outputs, discovery may look better while becoming more distant from reality.

Bloated MVP

The easier it becomes to build, the more tempting it is to add “just one more feature,” “just one more automation,” or “just one more integration.”

The MVP grows quietly.

And learning takes longer.

AI increases the importance of slicing discipline. It does not reduce it.

Artifacts Instead of Conversations

AI makes it easy to generate documents, tables, analyses, and presentations.

The problem begins when those artifacts replace conversations.

Lean Inception is not just information transfer.

It is the construction of shared understanding.

Apparent Maturity

A Product Vision may look mature. The personas, convincing. The MVP Canvas, complete.

But the textual or visual quality of an artifact does not prove that the team has learned enough to make those decisions.

If we are producing many more artifacts and discussing fewer decisions, AI may be avoiding the conversation instead of enriching it.

In the book, I also explore other antipatterns of product work in the age of AI. Here I selected the ones I consider most directly connected to alignment and MVP strategy.

6. How to create prompts that enrich the conversation

There is no magic prompt for Lean Inception.

What matters is the role we give AI.

A good prompt usually does one or more of these things:

• distinguishes facts from hypotheses;
• looks for ambiguity;
• makes disagreements explicit;
• challenges assumptions;
• presents alternatives;
• makes trade-offs explicit;
• suggests questions;
• shows what still needs evidence;
• avoids choosing for the team.

The difference is not only in the wording.

It is in the intent.

A prompt can ask for an answer.

Or it can help the team ask better questions before deciding.

A base prompt

Adapt it.

You do not need to include all these instructions in every prompt.

Want more examples?

I also prepared Lean Inception Prompts, with examples for different workshop activities.

Use them as starting points.

Especially when working with a team, treat AI responses as drafts, alternatives, or material for discussion — not as ready-made decisions.

Use AI to enrich the conversation, not to eliminate the conversation.

7. Checklist: is AI improving or avoiding the conversation?

And one final question:

If we removed the AI-generated artifact, would the team still be able to explain the decision, the trade-offs, and what it intends to learn?

If the answer is no, there may not be alignment yet.

Continue from here

We do not need to turn Lean Inception into an automated workshop to benefit from artificial intelligence.

Start small.

Choose one or two moments where AI can reduce operational work or help the team notice something that deserves discussion.

Then observe:

Did it improve the conversation or merely accelerate the work?

The tools will keep changing.

Responsibility for decisions remains with people.

AI accelerates the work. The team remains responsible for deciding what matters.

Want to go deeper?

• Explore the Lean Inception book.
• Go deeper into MVP thinking with the MVP Canvas.
• See the Lean Inception Prompts.
• Explore the AI Experiments.
• Read Teams Still Matter to go deeper into human direction, AI acceleration, and product work.

Want to bring this to your team?

If your organization wants to improve alignment, decision-making, and MVP definition in a context where building is becoming increasingly fast, explore our in-company Lean Inception training.

Or contact Caroli.org to discuss workshops, training, and applying Lean Inception in your organization.

Building became faster. Making good decisions became even more important.