Newshot

What is a team in the age of AI?

Original text(via HN AI)

I’ve been thinking about this question quite a lot lately. Partly because it’s a question I increasingly get asked.

We’ve built software engineering teams around complementary skills, shared knowledge, and the ability to solve problems together.

But what happens when one engineer can work with five, ten, or even twenty AI agents?

Does a team of ten engineers still make sense? Could three engineers achieve the same output? Would they make equally good decisions?

I have no answers. Just questions.

a black and white photo of a plane in the sky
a black and white photo of a plane in the sky

Why do we need teams?

Having led engineering teams for years, I know that much of their value is difficult to measure.

We measure velocity, delivery, quality, and increasingly, how much more an individual can accomplish with AI.

But how do we measure the value of a colleague challenging a decision before it becomes a costly mistake? Or someone remembering why an architectural choice was made three years ago? (This need happens way too often).

What about the engineer who notices that a teammate is struggling? Or the conversation over coffee that changes how we approach a problem?

These things never appear in productivity metrics. Yet they shape the quality of our work.

We know that a team is more than the sum of its individuals.

Good teams disagree, bring different experiences, challenge assumptions, and build on each other’s ideas.

Sometimes the most valuable contribution someone makes is convincing the rest of the team that they’re wrong.

When one person becomes many

Research is beginning to challenge our assumptions about teamwork.

In The Cybernetic Teammate, researchers studying 791 professionals at Procter & Gamble found that individuals working with AI could produce solutions comparable in quality to those of two-person teams without AI.

Some of the benefits we associate with collaboration can be achieved through technology. And as models advance more and more can be achieved.

But solving an innovation task together is different from maintaining a software system for years. (We could even challenge how good is AI at innovation).

Software engineering involves decisions whose consequences may only become visible months later. Someone needs to understand those decisions, take responsibility for them, and help others understand them.

We can delegate more implementation. But what happens to the shared understanding that used to develop while doing the work?

And if everyone consults similar models, will we challenge each other’s thinking as often?

I don’t know.

The human side of engineering

There is another dimension that concerns me even more. People want to belong.

We want our work to matter to someone. We want appreciation, recognition, and the occasional disagreement with a colleague we respect.

We want to celebrate achievements together.

I fear the day when the only feedback I get is from Claude.

It might tell me that my solution is elegant, identify a mistake, or suggest a better approach. But there’s something different about hearing from a colleague who understands the effort behind the work.

Trust develops through shared experiences. Through difficult decisions, mistakes, late-night incidents, and moments when someone steps in to help.

I’ve always believed that trust is one of the most important things a leader can build.

I wonder how that trust develops when engineers increasingly work independently, surrounded by agents rather than colleagues.

What happens to the next generation?

There’s also the question of learning. Most experienced engineers became experienced by making mistakes, asking questions, reviewing code, and working alongside people who knew more than they did.

Some of those activities are now easier to delegate.

AI can be an extraordinary teacher. It can explain unfamiliar systems, answer questions without judgement, and make experimentation cheaper. But learning also requires developing judgement.

How do you learn to recognise a poor architectural decision if you’ve rarely had to make one yourself?

How do you become a good reviewer without first experiencing the consequences of your own mistakes?

Perhaps the next generation will learn faster than we did. Perhaps they’ll develop entirely different skills.

But I think we should be careful about removing the experiences through which expertise develops.

What kind of teams will we build?

Engineering teams will change. Some will become smaller. Individuals will take on responsibilities that previously required several specialists. The boundaries between roles do not exist.

But I hesitate when I hear that a handful of engineers with agents can simply replace a team.

Output is only one part of what makes a team effective. There’s continuity, accountability, mentorship, disagreement, and the ability to make good decisions under uncertainty.

And there is something deeply human about building something together.

Perhaps we’ll discover that smaller teams work better. Perhaps greater individual productivity will allow us to pursue more ambitious projects with the same number of people.

What concerns me is that we might make these decisions based on what is easiest to measure.

We can count tasks completed, pull requests merged, and hours saved. It’s much harder to measure the value of belonging, trust, or a conversation that prevented a mistake.

The biggest change in software engineering is still to come. It will affect how we organise the people responsible for building and maintaining our systems.

And as we rethink what an engineer can accomplish alone, I hope we also ask ourselves what makes working together worthwhile.

Thanks for reading The Velocity Curve! Subscribe for free to receive new posts and support my work.

The text above was extracted automatically from the source page and may miss images, tables or code blocks. The original is authoritative.

Published by HN AI
This site only summarises and aggregates; copyright stays with the original author. Headlines and summaries follow the source — where they differ, the original wins.
Read the original ↗