Many developers ask for AI coding agents to help them code, expecting a significant increase in productivity. Nevertheless, the outcome in terms of velocity and deliverable quality can be highly variable and often disappointing. Powerful AI such as Gemini Pro or Claude code, are very capable tools for coding and reasoning (Chain of Thoughts). However, very few agile teams truly understand how to fully leverage them. Those who succeed increase both their velocity and the quality of their deliverables. This is not Vibe Coding, it is augmented coding and so much more.

The top 3 Areas to focus on

This article will check out 3 areas where AI can really boost your team:
- Architecture & Planning: Using AI to nail the initial design and planning phase.
- Long-Term Velocity and Quality: Seeing how AI can keep product quality high, so you can deliver faster and consistently.
- Fluid Workflows: Looking at ways to make sure your team’s operations are super smooth and efficient with AI on board.

Architecture & Planning

Using AI to Nail the Initial Design and Planning Phase.

The most underutilised, yet highly impactful AI application involves leveraging reasoning with large language models like Gemini or Claude (Technically called Chain of Thoughts). Their powerful reasoning features enable them to function as an additional team member, providing crucial insights and technical information for software development and clarification.

When to Use AI Reasoning

Reasoning should be used during events or workshops such as sprint planning, PI planning, product backlog refinement, architecture workshops, journey mapping, and story mapping.

How to Trigger AI Reasoning

To trigger the AI and initiate planning or reasoning, use specific keywords such as think-hard, think, or plan. If you are using Claude code use the plan mode with the ultrathink keyword

Here are key considerations:

Ensure that reasoning capabilities are activated for the current model. Some LLM agents use a switch and additional keywords.
Utilise a dedicated agent for planning, architecting, and reasoning.
Instruct the agent to adhere to your architectural best practices (e.g., Domain-Driven Design – DDD). A GEM can be used with Gemini Pro for fine-tuning with this very purpose.
Save the resulting plan or architecture in a Gemini.md file (if using Gemini pro), or in a Claude.md file (if using Claude).

Long-Term Velocity and Quality

Seeing how AI can keep product quality high, so you can deliver faster and consistently.

There is No High Velocity without Quality

Technical debt is the main culprit in terms of velocity loss and quality loss. Therefore, a top-quality product must be achieved in order to have sustainable high-velocity project teams

DoR

The Definition of Ready of any user story can be discussed with the AI itself for improvement, and the AI should be instructed to respect it via a Gemini.md or Claude.md file to check if the specifications are ready for implementation. For instance, the AI can check if acceptance criteria are available and automatically prepare tests for user stories.

DoD

The Definition of Done is a key element to produce a working software. By instructing it in the Gemini.md or the Claude.md your agent will always make sure that all the prerequisites for a working software are met before going to production.

This is particularly effective for software quality when the Definition of Done incorporates requirements like unit, integration, and scenario tests. A domain where AI excels and significantly saves time for developers.

Agentic Reviewer

Another very useful feature with agentic AI is the ability to run parallel agents as extra reviewers. Those reviewers agents can run locally, and help dramatically improve the quality even before pushing the code on the remote repository for a second review performed by a real developer.

It will therefore save time and resources because the code quality will already be improved before pushing it on the remote repository

Fluid Workflows

Backlog Maintenance

Using strong AI reasoning is key for sharp Backlog refinement and really getting into the weeds with task breakdown. Artificial Intelligence tools are great for making tasks like sorting through and cleaning up a huge product backlog much easier. Gemini, Claude or any other agentic LLM can be connected to tools like JIRA via an MCP (Model Context Protocol) server to automate the process.

Profiling and Optimisation of the Technical Workflow

Why not use AI for Runtime Profiling and Optimisation? It could really speed up Continuous Integration validation. Implementing loggers and analysing performance data is usually tedious and repetitive for us, but it’s ideal for an AI.

Enforce Best Practices with setup Markdown files

To keep developments modular and sustainable across one or more teams, make sure everyone knows and uses the common best practices like SOLID, KISS, AGNI, TDD, BDD, and DDD. You could put these instructions in files like Gemini.md and Claude.md.

Conclusion

Forget thinking of AI just as a coding helper; it’s a total game-changer, acting as the ultimate co-pilot for any Agile team. By smartly dropping powerful AI models like Gemini and Claude into three key spots, you can seriously level up your team’s game from average to absolutely exceptional.

The three secrets to leveraging AI in Agile Teams:

Smarter Planning and Architecture: Integrate AI into Sprint Planning and backlog refinement for superior reasoning. This yields better architectural choices, clearer building options, and proactive risk detection.
Killer Quality and Speed, Always: Defeat technical debt by training AI on your team’s Definition of Ready (DoR) and Definition of Done (DoD). The AI acts as an automated second reviewer by running parallel quality checks and automatically preparing unit/integration tests, ensuring long-term speed and top quality.
Smooth, Effortless Workflow: Use AI to simplify both mundane and complex tasks. Detailed Tasks: AI refines the backlog by breaking down tasks into small, manageable pieces to mitigate risks. Performance Boost: AI handles runtime profiling and optimisation, accelerating Continuous Integration (CI) validation. Enforce Rules: Embed best practices (SOLID, KISS, DDD) into AI-readable files to maintain modular, efficient development effortlessly.

Ready to transform your team’s performance?

Start by defining your AI agent’s architectural best practices today and save it in plan.md file.
Write your best practices (DoR, DoD, TDD, …) in a setup markdown file.

Read Further

For a deep dive into the technicalities:

Also read: From the Agile Promise to Industrial Bureaucracy: Farewell to a Betrayed Philosophy

How to Successfully Leverage AI to Bolster Agile Team Workflows

Author

Share