Skip to main content Intro Thesis Assessment Three Forces The Opportunity Three Practices Principles Manifesto Start Now ☀︎ Light

Agile & AI — A Scroll for Product & Engineering Teams

Your team. Your process. The AI era.

How to evolve your agile practice, your developer experience, and your team rituals to make the most of what AI now makes possible together.

Read on Agile Workflows — Developer Experience — AI-Powered Development — Definition of Done — Continuous Improvement — Inspect & Adapt — Human + AI Craft — Agile Workflows — Developer Experience — AI-Powered Development — Definition of Done — Continuous Improvement — Inspect & Adapt — Human + AI Craft — The Core Idea You need agility to evolve your developer experience. And you need to evolve your DX to truly leverage AI in your digital builds.

These aren't separate initiatives — they're one living system. When your process, your tooling, and your team rituals grow together, your capacity to do great work with AI grows with them. The teams thriving right now aren't the ones with the most AI subscriptions. They're the ones who've made continuous learning a core habit.

Where Are You on the Journey?

Three questions. No wrong answers.

Work Tracking Does your team capture how AI was used on each ticket — the prompt, the tool, what worked? Yes Not yet Refinement Cadence Do you use refinement sessions to do AI-assisted pre-work before tickets enter a sprint? Yes Not yet Definition of Done Does your Definition of Done require AI-generated documentation with a human review sign-off? Yes Not yet The Foundation

Three forces. One system.

01 — Practice

Agile as a Discipline

Agile at its best is a permission structure — an invitation to inspect how you work and make it better. That spirit is exactly what lets teams absorb new tools, learn from experiments, and grow every sprint. It's not just a process. It's a practice.

Without it ↓ Static team

Same sprint structure for 3 years. Retros that produce no process changes. AI tools adopted ad-hoc, unevenly, with no shared learning.

Agile team

Every retro has an AI segment. Processes update to match new capabilities. Prompt libraries grow. The team compounds its edge every sprint.

02 — Mindset

Product Thinking

When engineers think like product owners — anchoring every decision to user impact and real value — AI becomes a lever for outcomes rather than a novelty. The best work happens when the whole team is asking "what should we build?" alongside "how do we build it?"

Without it ↓ Without product thinking

AI is used to generate clever code faster. Velocity increases. But the wrong things get built faster — and still don't move the metrics.

With product thinking

AI helps with discovery, not just delivery. Teams ask "what should we build?" as much as "how do we build it?" Speed is applied to the right problems.

03 — Enabler

Developer Experience

Great developer experience is the soil that everything else grows in. Fast feedback loops, clear conventions, and frictionless tooling don't just help developers — they make AI suggestions easier to validate, integrate, and build on. Invest in DX and your AI investment multiplies.

Without it ↓ Poor DX

Slow CI, inconsistent environments, unclear conventions. AI suggestions are harder to validate and integrate. Developers spend their saved AI time fighting the toolchain.

Great DX

Fast feedback loops mean AI-generated code gets validated in seconds. Clear conventions mean AI suggestions fit the codebase naturally. The multiplier multiplies.

The Opportunity

Agility is what unlocks everything AI can offer.

AI tooling evolves on a weekly cadence. Model capabilities shift. Best practices get rewritten. The good news is that agile teams are already built for exactly this kind of change.

The deeper gift of agile has always been its permission structure — the standing invitation to look at how you work and make it better. That invitation is more valuable than ever now. It's what lets your team experiment with new tools, learn from them, and build those learnings into how you operate.

Teams with a genuine inspect-and-adapt habit don't just absorb change. They get better because of it.

Click any node to explore the cycle The agile loop is not a metaphor. Each stage is a place where AI changes the work — if you let it. What to Try

Three practical changes. Any one of them counts.

Not theory. Not a roadmap item for next quarter. These are things you can bring up in your next planning session and have running before the sprint ends.

01 Work Tracking Change to your work-tracking system

Add an AI field to every ticket.

A simple field on every story or task — for the starting prompt, the tool used, what worked, what didn't. It doesn't take long to fill in, but over time it becomes something genuinely valuable: your team's shared memory of how AI actually shows up in your work.

When that knowledge is captured, your team can learn from it together. Which prompts are worth reusing? Which tools saved the most time? Where did the model go confidently wrong? Shared learning compounds. Invisible learning disappears.

Works today in Jira, Linear, Azure DevOps, or any tracker with custom fields. You can add this field before your next standup. Linear — Issue Detail ENG › Sprint 24 › ENG-4471 Allow social login on the /register route Assignee M. Reyes Status In Progress Sprint Sprint 24 — Oct 14 Estimate 3 pts Starting Dev Prompt You Human decision: what context does this developer need? You've read the ticket. You know the codebase. Now write the prompt that tells your developer exactly where to start — which file, which pattern, which constraints. Then capture it. → + Add Starting Dev Prompt This field doesn't exist yet — but it should. 02 Ceremony Cadence Change to your meeting cadences

Double your refinements. Use them for collaborative AI pre-work.

Refinement sessions have always been where a team makes vague things concrete. Now they can also be where you bring AI in as a collaborator — before anyone writes a single line of code. Use the extra sessions to explore tickets together with AI assistance: draft acceptance criteria, surface edge cases, sketch the technical approach.

Developers who arrive at their keyboards with direction and context do their best work. Refinement is how you give them that.

Try proposing one additional refinement per sprint, framed as collaborative AI pre-work. See what changes in sprint predictability within a month. Refinement Session — ENG-4488 ENG-4488 — Feature Implement dark mode toggle No estimate — Backlog User can switch between light and dark themes. Preference should persist. You Human insight: what does this ticket actually need? You know the product context, the user need, and the technical landscape. Before AI can help refine this ticket well, your team brings that knowledge to the session. That's the thinking that happens before you click. → Run AI Pre-work A vague ticket. See what AI surfaces when you bring it into refinement. Acceptance Criteria User can toggle dark/light mode from the top nav on any page Preference persists in localStorage across sessions System preference ( prefers-color-scheme ) is respected on first load All existing components respond to the theme variable without component-level overrides Edge Cases to Consider User has no system preference set — default to light Pages containing embedded iframes or third-party widgets (charts, maps) may not inherit the theme Print styles should remain light regardless of preference Implementation Note CSS custom properties approach recommended over class toggling for breadth of coverage. If on Next.js, consider next-themes to avoid flash-of-incorrect-theme on SSR. 03 Definition of Done Change to your Definition of Done

Let AI draft the docs. Let humans make them trustworthy.

Expand your Definition of Done to include AI-generated documentation — inline docs, READMEs, API references, architecture decision records. This is one of the highest-leverage things AI can do for your team: produce thorough, consistent documentation at a scale that simply wasn't practical before.

The human review step is what makes it reliable. AI drafts fast, but it can also be confidently wrong about details only your team knows. A quick review by someone who wasn't the author closes that gap — and produces docs that future developers can actually trust.

Add two criteria to your Definition of Done: AI docs updated, and human review confirmed. These work as a single gate, in sequence. Definition of Done — ENG-4471 Unit and integration tests passing Quality PR reviewed by a peer developer Process Acceptance criteria verified in staging Quality No console errors in production build Quality AI-generated documentation updated (inline docs, README, ADR if applicable) AI Criterion — New Human review of AI-generated docs signed off (not the author) AI Criterion — New You Human decision: what does "done" mean for your team? Your Definition of Done reflects your team's values and standards. Adding AI criteria is a deliberate choice — a signal that your team takes AI-generated work seriously enough to quality-gate it the same way you quality-gate everything else. → + Add AI Criteria to Definition of Done Mark as Done → Your current Definition of Done is missing something. The Mindset

Principles behind the practice.

On Experimentation

Try it before you institutionalise it.

The best AI practices emerge from genuine exploration, not top-down policy. Give your team space to experiment freely — then codify what genuinely helps. Systematising before exploring produces rules that nobody understands and everyone works around.

On Partnership

AI is a collaborator. You're still the expert.

Think of AI as a tireless contributor — a first-draft machine, a research partner, a rubber duck that never gets tired. It handles volume and breadth beautifully. Architecture, ethics, and user empathy are yours. The two together produce something neither can alone.

On Learning

Make your learnings visible to the whole team.

Reserve space in every retro for what AI taught you this sprint — what worked, what didn't, what surprised you. Unshared learning fades. Shared learning becomes part of how your team operates. The AI field on tickets is the start. The retro is where it compounds.

On Process

Your process is always a work in progress. That's the point.

Teams that feel empowered to propose changes — to their ceremonies, their Definition of Done, their workflows — find it much easier to absorb new capabilities as they emerge. You don't need permission to improve how you work. Agile already gives you that standing invitation.

On Tooling

Your DX choices are now a team decision, not a preference.

IDE plugins, AI assistants, local dev tooling — these compound across every developer-hour in your sprint. Choosing them thoughtfully, evaluating them together, and updating them intentionally is craft, not overhead.

On Quality

Human review isn't a bottleneck. It's the craft.

AI generates at a pace that can make review feel like friction. Reframe it: review is where your team's judgment, context, and care for users shows up. Speed without that care produces things fast. Speed with it produces things well.

The Spirit of It The teams doing the best work in the age of AI are the ones who kept their curiosity alive. They try things. They learn together. They keep getting better.

Continuously improve your processes — in the true spirit of agile — so that your team is always exploring what's new, experimenting with what's promising, and building the best of it into how you work.

Ready When You Are

Pick one thing. Try it this sprint.

The AI field. An extra refinement. An updated Definition of Done. Any one of these creates real momentum. You don't need to do all three at once — you just need to start somewhere, see what you learn, and go from there.

Connect with En Dash A Dev-Focused Scroll Your Team — The AI Era

Agile — Product Thinking — Developer Experience — AI