Hi there,
One of my senior developers told me something that hit me hard.
He said he feels sad that he is writing less code. He still codes. He uses AI, including agent mode, more often, and it helps his productivity. But he misses the feeling of writing the code himself like before.
That is a big signal.
You do a lot to make work enjoyable for your team. You think about the right challenges, growth opportunities, and learning. Then the AI shift happens, and suddenly, a part of the work they used to enjoy is different.
Some developers can articulate this clearly. Many cannot. They just feel that work is not as satisfying as before.
Part of your job is to notice this. To name it. And to talk about it openly with the team.
What's changing for you as a manager
AI is changing how we write software every day.
Some teams only use it for small things like drafting pull request descriptions or basic documentation. Some use AI inside the IDE with tools like Copilot. Some go further and run full agent workflows that plan and write code with very little human typing.
In the middle of all this, your role as a manager is changing too.
It is no longer just about choosing tools or switching frameworks. This is not like "we used Angular, now we use React."
It is a deeper shift in how your team works, how they feel about their work, and how you define a good day.
Your impact goes beyond picking which AI tool to adopt. Here is what your team actually needs from you.
Three things your team needs from you
1. Teaching the new skills
When I first used AI in agent mode, I made the same mistake many people make.
I did not give enough context. I did not set clear constraints. The agent produced code that did not match what I wanted. Sometimes it even introduced bugs. It was frustrating.
Prompting sounds simple, but it is still a new skill for many developers, especially when using AI to write code.
Your job is to help your team learn:
How to think ahead and plan the change before asking AI to write code
How to give enough context and constraints
How to review and refine the AI output instead of accepting it blindly
You also need to adjust workflows so AI-generated code still fits your standards:
Keep AI-generated changes small enough to be reviewed by a human
Keep tests, code review, and CI requirements in place
Make sure common best practices still hold, such as DRY and clear abstractions
AI can get you working code faster. But if that code is messy, you are just trading speed today for debt tomorrow.
2. Protecting junior developers
There is another risk. It is easy for junior developers to lean on AI too much.
If a new graduate writes most of their code by prompting AI, they may never build the mental models they need.
It is like a first-year student learning math and using a calculator for every step. They can get the right answer. They do not learn how or why.
To help juniors become strong developers, you may need to:
Set some tasks where AI is not allowed
Ask them to explain AI-generated code in their own words
Give them problems where they must design the solution first, then only use AI for small pieces
The goal is not to block AI. It is to make sure they still build the foundations they will need later.
3. Managing the productivity narrative
The goal is still the same: shipping valuable software.
But what is a "productive day" when AI can write code 2 times faster?
If leadership expects a simple "2x output," you may end up with unrealistic deadlines, pressure on the team, and quiet frustration.
On the other hand, if you ignore the impact of AI completely, you miss a real shift.
You need a more nuanced view of productivity. Track it across four dimensions:
Velocity: Features shipped per sprint
Quality: Production incidents and rollback rate
Review burden: Hours spent in code review
Maintainability: Tech debt tickets opened vs closed
AI amplifies existing skill gaps. Strong developers become 2x faster at shipping quality code. Weak developers become 2x faster at creating technical debt.
That means you may see more code volume than before, harder performance evaluation, review bottlenecks, and hidden technical debt.
One of your roles is to manage expectations from leadership, especially the business and non-technical stakeholders. They read headlines that say "AI is 10x productivity" and may expect the same from your team.
You need to explain:
Where AI really speeds things up in your context
Where the gains are smaller
Where AI may shift effort from coding to design, review, and testing
You are not saying "AI does not help." You are saying "Here is where it helps, here is where it does not, and here is how we will measure it honestly."
Where is your team on the AI maturity curve?
Different teams will sit at different stages of AI adoption. That is fine. What matters is being clear about where you are and what comes next.
Here is a simple four-level model you can use:
Level 0: No AI usage
No AI tools are used in day-to-day development. This can be by choice or by policy.
Level 1: Basic assistance
Developers use AI for light tasks such as documentation and pull request descriptions.
Level 2: Integrated assistance
AI is part of the IDE or workflow. Developers use tools like Copilot to complete code, but humans still drive design and structure.
Level 3: Agent workflows
Teams use more autonomous AI agents that can plan and execute larger pieces of work with human oversight.
There is no single "right" level.
A team can stay at Level 1 or 2 for a long time if that matches their risk profile, domain, or regulations. What matters is:
You know where you are
You know why
You have a view of what the next step would look like, if you chose to take it
Productivity Tip
We often need to summarize text: emails, articles, or news.
The tool I find most helpful is ChatGPT Atlas. It works across any webpage.
Beyond summarization, you can ask questions about the content or use it
as a note-taking tool while you read.
It is simple, accessible, and saves time across many scenarios.
Your Feedback
I am writing this for you.
Hit reply and tell me: what part of this was most useful? What would you like me to go deeper on in a future issue?
If you found this helpful, forward it to a manager dealing with AI adoption on their team.
Hamid