AI Blocked My Colleague for Three Days. It Blocked Me for Five Minutes

Photo by La-Rel Easter on Unsplash

Oh our tickets. They’re not (how should we say), great. Details missing would be one thing, but typically we have tickets which are simply titles. No acceptance criteria.

The thing is our tickets are self-written.

So what might go wrong when a technical lead writes a ticket using Claude? I think you’re not going to be surprised about this one.

An Incorrect Ticket

Working on one of our crucial systems, pushing out a new feature that will “revolutionize” the lives of our users.

I pick up the next ticket from the stack. As I start to complete the tasks I realise that one of our tickets is wrong. Not “slightly unclear” wrong. Not “could use a bit more detail” wrong.

Actually wrong.

The requirements had been generated with AI and somewhere between the prompt, the generated output, and the human. It didn’t take into account all of the edge cases that frankly were not niche. How system A would fit into system B was left as an open question, and there were multiple ways to figure that one out let me tell you.

Blocked

One developer declared themselves blocked due to the unclear ticket.

They pushed the issue back to the team responsible for the technical analysis and essentially stopped work until somebody worked it out for them.

I think that’s not the best, since no work could take place for an extended period of time due to this approach.

I did something different. I opened Claude.

Within a few minutes I had pointed it at the relevant code, asked it questions about the implementation, followed a few references through the codebase, identified the actual contract and carried on working. Claude could look through other systems and find the contract between them, so I knew what data needed to be passed through.

I still needed to confirm a few things with the humans over Slack, since the answers I’d improvised were not guaranteed to be correct.

Yet I unblocked myself. I managed to get things done, and get things shipped.

I’m now wondering if I did the right thing at all?

The Old Way

Traditionally software development has been built around information bottlenecks. If you don’t know something, you ask the expert and wait for the response.

If the expert is unavailable, you wait (and wait). If another team owns a service, you create a ticket. If the ticket is wrong, you send it back.

The system is made up of people protecting themselves from making a mistake, or being accountable. Over time this has become part of our software development culture, where people are comfortable being blocked and unable to progress in their work.

Is it surprising if this hasn’t really changed over time, even with the advent of AI and all that has occurred in the last few decades of software development?

Unless of course you thought we might be moving forwards and improving the world of software development. I guess I’m not surprised, when it comes down to it.

The Lack of AI Change

AI doesn’t give you answers (at least it should not). What it does do is reduce the cost of investigation. It should also be limiting the reasons why we are blocked, and the problems that a circle of developers blocking each other causes.

Ten years ago understanding an unfamiliar service might involve reading documentation, finding the right repository, identifying the correct owner, arranging a meeting and then waiting for somebody to become available.

Not anymore. You get an AI to “help” you and things are not nearly as difficult as they used to be.

The risk is turning every uncertainty into a dependency.

At some point software developers stop being problem solvers and become ticket routers.

That’s something that none of us signed up to be. And AI can’t solve the incompetence problem for any of us.

Conclusion

I can’t shake the feeling that software development rewards momentum.

If I can spend five minutes investigating before declaring myself blocked, shouldn’t I? Shouldn’t I be trying to make our software development process faster and easier for the whole team?

After all perhaps the real lesson isn’t about AI at all.

Perhaps it’s that we’ve become so used to waiting for answers that we’ve forgotten how to look for them ourselves.

AI just made looking easier. Yet it didn’t remove the core problem at the heart of it all.

About The Author

Professional Software Developer “The Secret Developer” can be found on Twitter @TheSDeveloper and regularly publishes articles through Medium.com

The Secret Developer is the core problem.