I mistakenly halted my team‘s development by an AI agent. Here’s why.
Chasing agentic automation has me questioning the net-positive, and leads me to an unexpected conclusion.
Several weeks ago I built an agent to compile my team’s weekly status updates (see my article I got two hours back from the worst part of my job. Here’s where they went.),
I said that I saved 2 hours… and since then it instilled doubts about my leadership.
Let me explain. I made myself the sole operator of the agent - the guys on my team don’t have to write these updates now.
The great part: they got the context switch back, so I see it as a clear win. Yet somehow that’s the part that keeps nagging at me.
I mentioned coaching above. Well, that part of my brain can’t process whether I’d done my team a favor, or just took away a development opportunity.
But is it still manager’s job nowadays to coach and develop people? Even in the tech industry?
The manager job looks to me that we need to do everything we’ve been doing up until now plus put AI on top of that and to ‘work better’.
My non-negotiable rule: AI should sharpen my judgment, not give me more work.
When I was writing the weekly-reporter agent, I had one motivation: make room for the judgement, by having the agent take the work that was never judgment.
Underneath all of it, the job hasn’t moved. A manager is, foremost, elite at crafting relationships. Not only with the team, but with the people who benefit from the team’s work.
Would you buy a product from someone you don’t like? I know that I wouldn’t. Collaboration runs on the same rule.
Purely for the sake of the argument, let’s say that development is equivalent to a promotion.
A promotion is an outcome of the quality of those relationships. Say what you want: pure competence, the complexity of the deliverable, or the money-amount saved for the company are key signals for promotion, but one will never get it unless they form a strong bond with those, who are the decision makers. There’s no way around it.
Do you see where I’m getting?
Here’s what I actually coach toward:
With the juniors, it’s almost always the same gap - they can write the code, but they can’t yet tell you what it’s for. Getting them aware of the expected outcomes of what they produce is the thing that needs the most work;
With the seniors, the shape flips - they’re technically excellent and their work is visible across the org, but because they’ve never had to manage people, getting that work to land with non-engineers is the muscle they’ve never built.
Two different gaps, same underlying skill: taking your own work and making it matter to someone who wasn’t in the room with you.
My coaching question chain:
What do I want to do? → Why does it matter? → How will it manifest to others? → Why will it be appreciated?
The last two links are key: the relationship-formation part, the exact spot where an engineer has to stop thinking about their code and start thinking about the person on the receiving end.
The weekly status update is where they used to practice it. By resolving it for my directs, I deleted the last two links from their week.
This doesn’t let me sleep at night:
I have 7 peers, each of them has four to eight directs. Let’s say that our entire org comprises of 60 people. If every one of those managers runs the same agent and buys back two hours a week the way I did, the time saved across the org is enormous!
And every one of them could easily take on half of the deal - the time saved - without ever reaching the question of a lost development opportunity.
Automate the same two links away — across a 60-engineer org, and what happens to the next wave of seniors who were supposed to learn, somewhere, how their work lands on other humans?
I don’t have an answer for that. And it bothers me.
Here’s what I decided:
I’m definitely not getting rid of the agent. The information in a status report matters because leadership needs to receive it, not because my engineers need to spend an hour reformatting it every week.
However, the last two links of the chain are definitely not the busy work that is worth automating, quite the opposite. Those are real opportunities to form a relationship, real moments of impact that you can get through coaching sessions.
What I didn’t expect:
The agent still sharpened my judgment - it reminded me about what I actually love about this job! Coaching → and by that, helping my guys build lasting relationships with the people their work serves.
That’s the engine of my effectiveness.
By my own rule, that makes the agent a win. It did the thing AI for a leader is supposed to do - it made me think better, not just work less.
But a realization on its own changes nothing. Judgment only counts when it turns into a decision, and I have two in front of me.
What’s I will do next:
Either I reconfigure the agent so it still makes my engineers do the rep - it compiles the facts, but each of them still has to put words to who the work is for and why it will land, or - I stop pretending status reporting was ever the right home for that muscle, and I rebuild my coaching around something with real resistance in it.
I haven’t fully chosen. What I won’t do is let the agent make that choice for me by default.
Bottom line - the status report was just where the coaching used to happen by accident. The agent didn’t take that away. It handed me the judgment to do it on purpose.
If you manage people and you’ve started handing recurring work to an agent, the hours it saves you are the boring part.
The real question is what that work was quietly teaching whoever used to do it - and whether you’ll spend the time it frees up sharpening that, or just doing more.
Which one are you actually doing with yours?