You are reading a paid post from my newsletter — The Art of Doing Technical Program Management. I write with an aim to demystify the art of technical program management and deliver proven real-world strategies, practical advise and tips, and actionable frameworks to help you level up as a Technical Program Manager.
Upgrade To Paid - 20% OFF Annual Plans
If you already are a subscriber and found this essay helpful, please share it with your colleagues and network. I am sure others might find it useful as well.
A few weeks ago I started building an AI agent to generate a morning briefing. The idea sounded straightforward. Every morning it would read the documents I cared about, summarize my Slack channels, look at my meetings, and tell me what deserved my attention.
As I started wiring everything together, I created a spreadsheet listing every input the agent would need. Every recurring meeting, Slack channel, document, dashboard, JIRA project. This is sage advice everyone has about building agents. Halfway through building it, I stopped.
The spreadsheet wasn’t revealing what my AI agent needed. It was revealing how complicated my work had quietly become.
There were too many meetings.
Too many Slack channels.
Too many places where information could live.
It suddenly hit me. I didn’t have an AI problem. I had an organizational design problem. Classic problems, same old same old. Instead of making the agent more sophisticated, I started asking different questions.
Which meetings could disappear?
Which Slack channels should be consolidated?
Where should the single source of truth actually live?
What were the expectations for where different kinds of information belonged?
The AI project ended up improving the system long before the AI agent was finished. That experience changed the way I think about AI. Not because it taught me a better prompt. Because it reminded me that good judgment beats good prompts.
AI is getting remarkably good at execution. It can summarize meetings, draft status updates, build project plans, write documentation, and analyze risks faster than ever before. But none of those answer the question that matters most: Should we be doing this in the first place? That’s a judgment call. And I think that’s where TPMs become even more valuable - making good judgement.
💡Judgment is one of those words everyone uses these days but rarely does have an agreed upon universal definition. For a TPM, I don’t think judgment is simply making the “right” decision. I’ve started thinking about good judgment as having four elements:
What action should be taken.
When is the right time to take it.
Where the conversation belongs.
What context matters before acting.
Miss any one of those elements and even a technically correct decision can become the wrong one. Sometimes the right answer is a Slack message. Sometimes it’s a five-minute conversation. Sometimes it’s escalating to a VP. And sometimes… the best decision is not taking action yet.
AI doesn’t eliminate the need for judgment. It changes where judgment gets applied. The pattern I’ve started noticing is this.
Someone asks, “Can AI do this?” and the immediate response is to build an agent, write a prompt, or automate the workflow.
Sometimes that’s exactly the right answer. But I think the better question is: Should AI do this?
More often than not, that’s a judgment call, not a technical one. Looking back at my morning briefing project, I realized something. The AI agent wasn’t the solution. It was the flashlight. It exposed complexity I had slowly become accustomed to.
Instead of helping me automate a messy system, it helped me see that the system itself needed redesigning. Sometimes AI isn’t the solution. Sometimes it’s the flashlight. It exposes unnecessary complexity that good judgment can eliminate.
One thing that worries me about an AI-first world - learning through mistakes. It has become incredibly easy to outsource not just the work… but the thinking behind the work. Judgment isn’t downloaded after reading a book. It’s earned through repetition. Every escalation you handled well. Every roadmap discussion where you pushed back. Every meeting you scheduled at the wrong time. Every mistake that taught you something. Those experiences slowly calibrate your internal compass. That’s how judgment develops.
The danger isn’t that AI replaces our work. It’s that AI replaces our practice. If AI makes every difficult judgment call for you, you might become more productive today… while becoming less capable tomorrow.
💭 One of the most underrated parts of becoming a great TPM is making mistakes, reflecting on them, and carrying those lessons into the next project. AI can accelerate learning. It can’t accumulate wisdom for you.
That brings me to something I’ve been thinking about recently. We talk a lot about technical debt. I think AI introduces another kind of debt: Judgment debt.
Judgment debt accumulates when AI repeatedly makes decisions you should have learned to make yourself. The tricky part is that it feels like progress. Your emails are polished. Your project plans are better. Your stakeholder updates are clearer. Everything looks more efficient.
But judgment debt doesn’t show up while AI is available. It shows up when a senior leader unexpectedly asks, “What do you think we should do?”. There’s no time to engineer the perfect prompt. That’s when your accumulated judgment becomes visible.
The good news is that judgment is a skill. Like every other skill, it can be developed deliberately. Honing it will take time but there are things you can start doing today that will help you on your journey to get better at good judgement.
🔐 For paid subscribers, here’s the 2-step framework I use to sharpen judgment as a TPM and something you can start today.
1. Ask Better Questions Before You Ask AI
Whenever someone asks, “Can AI do this?” Pause and ask three questions first:
Comments
Nothing yet. Say the first thing.
Sign in to join the conversation.