From Scrum to AI-Native Delivery: What Changes, and What Should We Watch?

[#1]

What happens to Scrum when the team is no longer the only thing producing the work?

This is not just a theoretical question for me. I expect to experience this transition firsthand.

Many teams are moving from a Scrum-centered operating model, often scaled through frameworks such as SAFe, toward a more AI-native way of working.

But I don't think this is mainly about AI writing code faster.

The bigger question is: what happens to the delivery model when AI becomes an active participant in the process?

AI can increasingly generate code, tests, documentation, prototypes and technical alternatives.

As a result, the bottleneck may move away from implementation and toward:

• clearer product intent
• better specifications
• faster validation
• architecture and decision-making
• security and governance

This raises an interesting question:

Is the user story still the primary unit of work?

Or are we moving toward something closer to specification-driven delivery?

The better the specification, the more useful AI becomes.

The worse the specification, the faster AI helps us build the wrong thing.

Team structure may change as well. Smaller teams could become much more capable with AI, but simply reducing team size creates risks: knowledge concentration, overload on senior people, weaker reviews and less resilience.

If AI increases the amount of work one person can initiate, the bottleneck may simply move from execution to review, architecture and decision-making.

This also makes me question traditional Agile metrics.

If AI can dramatically increase the amount of code a team produces, does velocity still tell us much?

And this is where I am particularly curious about Jira.

If the flow of work evolves from:

Epic → Story → Task → Sub-task

toward something closer to:

Intent → Specification → AI-assisted execution → Validation → Learning

should the way we model work in Jira evolve as well?

Will Jira remain primarily a system for tracking work, or could it become an orchestration layer between people, AI agents, specifications, code, tests and outcomes?

I don't think Scrum is obsolete. Transparency, feedback, collaboration and adaptation may become even more important.

But perhaps the emphasis is changing:

Less ritual. More clarity.

Less focus on output. More focus on validated outcomes.

Less coordination of human effort alone. More orchestration of human and AI capabilities.

For me, the biggest risk is treating AI-native delivery as simply a tooling upgrade.

AI can accelerate delivery, but it can also accelerate unclear requirements, poor decisions and weak quality gates.

So I am curious to hear from others:

Are you adapting Scrum for an AI-native environment, moving beyond Scrum, or building a hybrid model?

What should replace, or complement, story points and velocity?

And perhaps the question I am most interested in:

How should Jira evolve to support this new way of working?

[Nigel Budd]

Terrific questions, a lot to unpack! I'll give my opinions, but please don't hold me to my answers in a years time, it's likely to have all changed by then!

Q: Or are we moving toward something closer to specification-driven delivery?

A: Weren't we doing this already?  A developers job should be to take a customer requirement (in user-story format for example), and to turn that into something that can be coded, and importantly defined in a testable way.  The only change I see here is that the actual coding part of the process will be agentic.

Q: Team structure may change as well.

A: I think the Team size may well remain the same because of the reasons you gave, but I could see less need for Team of Teams delivery, if a Team can product the work of 5 or 10 times developers a single team might be all you need.

Q: bottleneck may simply move from execution to review, architecture and decision-making.

A: bottlenecks are always changing.  The agentic AI supported developer is doing more governance work than actual coding, the bottlenecks are now likely to be deployment, or even in the product management process.  The need for DevOps, automated testing, effective monitoring is more important now than it has ever been.

Q: does velocity still tell us much?
A: I think it has in the core purpose of velocity which is to answer the question of how much 'stuff' can a team deliver in the next sprint.  I don't really care if velocity goes up, that doesn't tell me much, what I care is how much value the team delivers and increasing that can't be a bad thing.

Q: toward something closer to: Intent → Specification → AI-assisted execution → Validation → Learning
A: I really loved this challenge...but I think you might be confusing hierarchy with workflow.  

Epics->Stories = Decomposing a large thing that the customer wants into smaller parts so the team can deliver it incrementally.  I think that should still work in an agentic dev world.  (although Vibe coding could well change that dramatically!)

Intent->Spec->etc   I think that describes a pretty much standard dev workflow as I said in my first answer, all that's changed is the "AI-assisted".  

Q: Will Jira remain primarily a system for tracking work, or could it become an orchestration layer between people, AI agents, specifications, code, tests and outcomes?

A: Isn't Jira an orchestration system already?  AI agents are just a new kind of resource, the other things were always there.

Q: Scrum Less ritual. More clarity.  Less focus on output. More focus on validated outcomes.  Less coordination of human effort alone. More orchestration of human and AI capabilities.

A:
Less Ritual, more clarity.   Should always be part of Scrum.
Less Focus on output, more focus on outcomes.   Should always be part of Scrum.
More orchestration.  I don't think Scrum should care who did the work, whether it was automated or not, the outcome and customer value is what's important.

Q: biggest risk is treating AI-native delivery as simply a tooling upgrade.

A: 100%

Q: Are you adapting Scrum for an AI-native environment, moving beyond Scrum, or building a hybrid model?
A: I think Scrum is robust enough to work in this environment, I am interested to hear what others think though.

Q: What should replace, or complement, story points and velocity?

A: They still have their place, value is what's most important, and let's not forget quality.

Q: How should Jira evolve to support this new way of working?

A: It already is, Jira Agents, Rovo Dev, as well as the cost and quality tracking features in DX and Focus too.

[Andrea Mura]

Thanks for such a thoughtful reply, I really appreciate the time you took to break down each point. I actually think we agree more than we disagree.

Perhaps we're looking at the problem from two different levels. You're asking whether Scrum itself remains a solid framework, and I tend to agree that its core principles, transparency, feedback, adaptation and delivering value remain just as relevant.

The question I was trying to explore is slightly different: does AI change where the human effort is invested within that framework?

On specification-driven delivery, I don't mean that specifications are anything new. Good teams have always transformed customer requirements into something implementable and testable. My thought is that AI may amplify the importance of that activity. As implementation becomes increasingly automated, defining intent, constraints, acceptance criteria and validation may become a larger part of the engineering effort than it is today.

I also liked your point about Jira already being an orchestration platform. I think that's fair. Perhaps the more interesting question is whether what Jira orchestrates evolves over time. Today, the focus is mainly on issues, workflows and people. Tomorrow, it may also need to coordinate AI agents, generated artifacts, validation gates and governance alongside traditional work items.

Your observation about bottlenecks really resonated with me as well. AI doesn't remove constraints, it shifts them. If implementation becomes dramatically cheaper, the real bottlenecks may increasingly be product decisions, architecture, validation, governance and deployment. In that sense, we're making a very similar point.

The question I'm still exploring is this: if AI eventually generates a significant share of the implementation, what becomes the primary engineering responsibility of the team? My intuition is that the biggest change won't be in Scrum itself, but in how teams define, validate and govern the work that AI helps deliver.

Really enjoyed your perspective, this is exactly the kind of discussion I was hoping to start.

[Nigel Budd]

We do agree a lot, if you are in Amsterdam in October lets have a beer :-)

I do believe the human effort will change dramatically, and very rapidly.  Already developers are doing much less coding, the dev is now performing a governance role, describing the need, often describing how they want the work performed, and then checking the output afterwards.  

Because the developer isn't actually writing the code, I think there is now a hybrid dev-QA role evolving, especially if the AI (or a different model) is creating test automation to validate the output.

This in a way is a deskilled role from a traditional software engineer/quality engineer, the new core skills in my opinion are being able to translate what's needed into instructions for the AI to follow, attention to detail, complete lack of trust in AI (so you don't get fooled into believing it all the time) and an ability to multi-task like crazy.

These are parenting skills really, and dealing with an AI platform is a lot like dealing with a toddler.

Devs can also be closer to the business or customer in this new model, reducing handoffs and mis-translations, this could be a huge change in the skill-set required by a modern developer.

you asked: What becomes the primary engineering responsibility of the team?

I don't think that changes, the job is still to deliver customer value, some business and consumer challenges using technology.  How they achieve this is being revolutionised at the moment, and it's incredibly exciting.  I believe tools like Jira are now even more essential to orchestrate the mix of human + AI endeavours, to provide transparency, change management, traceability, and make sure that the quality doesn't degrage as the amount of work being delivered increases.

 It would be so fun to look back at this conversation in a years time to see how our predictions worked out.