How One Team Cut Planning Time by 70% Using AI That Reads Communication

Peter owns a fast-growing German software scale-up. By developing in an agile way, his team can handle the high level of dynamics in their domain. But, as with many agile teams, this comes at a cost: upcoming tasks need to be specified, prioritized, and planned almost in real time to keep the sprint cycle from breaking down.

Peter describes the situation like this:

Planning is a huge effort. Our team spends two hours in sprint refinement to specify the upcoming work. Before that meeting, our product owner is busy for the entire two weeks, jumping from one business meeting to the next to collect requirements and estimate priorities. However, during the refinement meeting, we still often end up discussing business details involving the project manager, scrum master, and even team members who also have direct contact with our customers. Often we need to extend the meeting to four hours and still hardly have enough time to refine the tickets on a technical level.

And technical refinement is becoming more important than ever.

In a world where developers are increasingly supported by AI agents, their role is shifting. They are no longer just programmers. More and more, they are becoming technical analysts who define instructions for supporting agents, review the generated code, and specify test cases. As coding itself becomes faster, thorough specification turns into the real bottleneck. The result: developers move faster and faster, pulling more and more poorly refined tickets from the backlog. That often leads to misunderstood customer needs and flawed architectural decisions.

Tim, the team's lead developer, is responsible for keeping the codebase clean and aligned with customer needs. He describes the challenge like this:

Agents help us move a lot faster, but our job supervising them is incredibly important. Agents can easily adapt thousands of lines of code, touch dozens of files, and implement architectural changes in only one minute. If you don’t know exactly what you are doing and why, this becomes extremely dangerous—especially in a team with seven developers.

As a result, technical specification now takes a lot of time. A developer implementing a change often spends most of that time thoroughly discussing the required adjustments with the team—even beyond the actual refinement meeting.

And even though the team tries hard to do this well, things still go wrong. Major changes sometimes make it into production before the product owner realizes that the new software increment is different from what was actually agreed with the customer.

The reason is simple: the discussion with the customer never made it into the requirements.

More careful meetings and heavier ticket hygiene helped at the margins, but they did not scale: every jump in development speed widened the gap between what customers had actually said and what showed up in the backlog.

At this point, it became clear that we needed to tackle this with AI. When developers become unbelievably fast with AI, but business still operates slowly, that simply cannot work.

The key insight was this: tickets need to specify themselves in real time.

How it works

Whenever someone talks to customers and something comes up that should affect the developers’ ticket specification, that information is now captured automatically.

When developers discuss how to implement something, the ticket specification is updated as well. Even the test cases described in the ticket can evolve automatically.

The team now uses agents that listen to communication. Whenever something relevant is discussed, the agents propose an update to the ticket specification.

Now, during sprint refinement, there is no longer a big discussion about business requirements. The specification in the tickets is kept up to date in real time—and we can see which communication, and by whom, is associated with each ticket.

Example workflow

1. The product owner talks to the business about a new topic → the AI proposes creating a task with a rough description.
2. The product owner discusses the required work in multiple meetings and emails with different business stakeholders → every new piece of information automatically ends up in the ticket.
3. The product owner discusses the ticket with the project manager and scrum master → both raise valid points and send follow-up emails to the business. The answers also flow into the ticket automatically.
4. A developer picks up the ticket → he has a question that affects the architectural setup. He first discusses it with his team → the AI updates the ticket.
5. He reaches out to the business → the ticket is updated again.

Previously, a lot of this new information never made it into the ticket. As a result, a lot of bad outcomes were shipped simply because development had become much faster than the business side.

Now, the business side has learned that it can also use AI to become faster and more precise.

Coding with agents only works well if your business is ready for it.

Related Reading

• aicoo – Connecting Communication with Project Management Artifacts
• From Chaos to Clarity: Why Project Communication Must Become Structured Data

Try it for free

See for yourself how aicoo can keep your tasks aligned with ongoing communication—without turning refinement into a replay of every stakeholder conversation.

Open the tool