BLOG 2: Big Epic? Don’t Start from a Blank Page – use AI to Find Vertical Slices

A big Epic on the backlog often looks simple. Until you try to break it down. That’s when the questions start:

• Where do we make the cut?
• What goes first?
• What can be delivered independently?
• What depends on what?
• What is user value, and what is just a technical step?

And very quickly the team is sitting in front of a blank board, trying to come up with the “right” decomposition.

This is where AI can be useful. Not because it knows your architecture better than your developers do. But because it can quickly offer several possible ways of looking at the same problem.

Don’t ask for “break the Epic into tasks”

That’s one of the easiest ways to end up with a list like this:

• front-end;
• back-end;
• database;
• API;
• testing.

It may look organized. But it may have no user value whatsoever.

A much more useful request is:

“Suggest possible vertical slices that represent a concrete user or business outcome.”

That way, AI has to think about the functionality from the perspective of outcomes, not just implementation.

Proposed prompt to get started

Role: You are an experienced Product Manager / Product Coach with experience in decomposing complex features. Propose an initial decomposition of the Epic as a basis for conversation and refinement with the team. The proposals are not a final specification.

Context:

Epic: [name]
User/target group: [describe]
Problem: [describe]
Expected value: [describe]
The Epic includes: [describe]
Dependencies/constraints: [enter]
Out of scope: [enter]

Task: Propose several possible vertical slices. The goal is for each slice to be the smallest possible coherent whole that enables a concrete user or business outcome.

Do not split the Epic only by:

• frontend;
• backend;
• database;
• API.

If an enabling/technical slice is necessary, mark it and explain why.

For each slice, provide:

1. name;
2. user/business outcome;
3. what it includes;
4. dependencies;
5. potential size;
6. open questions;
7. biggest risk;
8. up to 3 questions for refinement.

If a proposal is not a good vertical slice or is still too large, clearly mark it.

ATTENTION: AI shouldn’t tell you which slice is “correct”

This is important.

AI doesn’t know everything your team knows. It doesn’t necessarily know:

• the system architecture;
• technical debt;
• capacity;
• organizational dependencies;
• existing implementation decisions.

That’s why its output should be treated as material for a conversation and not as a specification.

Try it today

Pick one Epic that is currently sitting in your backlog. Don’t ask AI for User Stories.

Ask:

“Give me three possible ways to break this Epic down into vertical slices and explain the trade-offs of each approach.”

For example:

Approach A: smallest possible user slices
Approach B: slices optimized for early learning
Approach C: slices that take known dependencies into account

Then check the proposals with your team.

Suddenly you’re no longer discussing: “How do we break down this Epic?” but rather: “Which of these three approaches fits our context better, and why?”

That’s a much more productive conversation.

If you reject all of the AI’s proposals, the experiment was still a success. Because AI helped you get to the conversation faster.