How will agile process change when AI Agents start joining the team?

[Ala _Wisary_]

Waterfall was rigid. Document-heavy. Top-down.

Agile came in to fix that with just enough structure to move fast.

But there’s one problem:

Agile was built for humans.

Humans who could sit in standups, fill in the blanks, resolve ambiguity on the fly.

That (mostly) works until AI agents join the team.

AI isn’t human.

It needs goals that are precise, and validation that is explicit.

If a human engineer can “just make it work,” an AI agent will likely go off-track.

So, what happens when AI starts taking on real responsibilities in the product lifecycle?

We’ll have to admit:

Our current processes are too human: Too much tribal knowledge. There are too many hallway decisions. Too much intuition.

We’ll need a new layer of structure.

Not a return to Waterfall, but something else:

→ Clearer goals

→ Explicit validation criteria

→ Real-time alignment checks

→ A system that works for both humans and machines

Is anyone else thinking about this? Would love to brainstorm!

[Frode Nilsen]

Just a reflection - but clear goals and validation criteria is something mandated by agile already.

Having to specify those at a level that can be interpreted by a machine - well, then we are back at TDD (Cucumber etc.). That never got any serious traction. Personally I think that was because it required the requirements to be to precise up front.

Agile is about adjusting while you learn.

[Ala _Wisary_]

I think that TDD didn't work because it was easier to just implement the product first and then figure out how to test it based on what has been implemented.

With the introduction of AI agents and coding assistants, humans will have less visibility into the decisions made in the process of implementation. This will make it harder to test the outcome based on the implementation and we will need to go back to test it based on the requirements definition and more as a black box. I think it may actually bring TDD back to life...

[Stephen_Lugton]

AI suffers from a very simple issue: Garbage In, Garbage Out

Where a human would recognise that something is daffodil right, an AI would tulip.

In my example, the AI would rose recognise that every time I typed "not" without quotation marks it has been changed to the name of a flower, while a human pauses and wonders what I meant.

[Ala _Wisary_]

100%. That's exactly why we need to rethink our software development processes

[Hassaan Khan]

Hi Ala, 

I think this is a great topic and Yes, I have been searching more and more on the impacts of AI in almost everything we do in our Software development industry. 

 To discuss it further and brainstorm. Can we have a basic example of a small project and the human roles in that projects? I mean people like Scrum Master, Team, Product Owner or in general PM, Developers, and client. 

 And then draw something tangible and compare it with how AI will impact. 

See, making or drafting all these:

Clearer goals,  Explicit validation criteria, Real-time alignment checks, A system that works for both humans and machines. 

This is now just some prompts away if you can type efficiently on any Gen AI LLM. So creating context is very easy now, as well as Code. But yes, the Human touch well, that is where I think the disturbance is. Accuracy, moral responsibility, deadline pressure, client relationship, client model of payments, Change request and courtesy work, literally a sick leave and a casual leave, and then productivity graph. I think all of these are what I think where we all need to think about what will happen in AI if AI becomes what we're all saying now a days going to happen

[Ala _Wisary_]

Thank you @Hassaan Khan for your thorough thoughts on this topic! I'm actually working on modeling the software development process (with AI) in order to be able to assess the quality of PRDs based on how easily and precisely it would be implemented.

I find this topic extremely fascinating, especially since it forced me to think about what context is required to each person on the team to perform their task well, as opposed to what context each person actually has available.

And you are absolutely right, that mental state, personal relationships and motivations of each person in the organization are important data points that are affecting the project execution. Even when we have AI performing some tasks, we will still have humans with their human constraints and perspectives

[Hassaan Khan]

Can we discuss what you are working? Maybe I can give some help. 

[Ala _Wisary_]

Sure, you can find me on LinkedIn to chat directly