In the Netherlands, if you install solar panels and have extra electricity, you don't sell it back to the grid - you pay them to take it. This is capitalism at its most extractive. Even when you've done something for your own benefit (reduced electricity bills) and the collective's (reduced reliance on fossil fuels), it somehow costs you.

AI is similarly extractive. It's also addictive. My partner compared me to an addict this week - fair assessment. I was hyped up on vibe coding another random idea, impatient and unable to focus during the wait for MOAR TOKENS, prioritizing dopamine over the more legitimate items on my todo list. There was an article in HBR about how AI makes people work more intensely, not less. I felt that.

One of the challenges of the hype is that we have a hard time accurately assessing our own productivity. A 2025 study of OSS developers found they thought they were 20% more productive with AI, but were actually 19% less productive. Probably also true of me. The tool's purpose is fulfilled, yet the scope creep continues. Now it has graphics and I'm contemplating turning it into a small desktop application. Am I more productive, or less?

Something that fascinates me is the gap between individual and collective productivity. How often have we read some AI augmented, or straight up generated "work" and found it a waste of our own time to consume? One person's productivity is another's slop.

But also, interestingly, one person's productivity can be another's prompt. My partner works in accessibility - his output is no longer code but clear specifications. He got feedback from a team about how they're using AI to be productive on the implementation work, but the input is everything. His work. Their prompt.

Coming back to the idea that one person's productivity is another's slop - I think we're politely calling this "taste". I would prefer to use the word judgement. Taste is aesthetic preference. Judgement is knowing what actually solves the problem, what's good enough, when to ship. I find AI fascinating as a multiplier of competence. Some of the most competent people I know have been unleashed. Every incompetent person I encounter (especially when I'm busy or stressed), I would be happy to replace with a chatbot. However incompetence equipped with a chatbot is much worse than those who haven't found it yet, because it increases the volume you need to review.

Grant me output, and the judgement to tell the difference.

The challenge of the moment is that all this individual productivity meets an organization and turns to dust. The best documented model of the software factory is just three people. Even in what was considered a small team - 20-50 people - how do you take the lessons of three people and apply it? Now a team is not the product of a set of people, but the product of all those people and their own individual set of multipliers.

The challenge for managers: how do you coach someone to do better on a skill you don't have? I've been hearing people who have left coding say they think this shift doesn't apply to them. I believe it does. Instilling judgement is not an easy task. If the team makeup changes, the job changes. The team makeup has changed in ways we currently don't have a clear model for. That means the job is harder than ever.

I empathize with those who are not yet at this conclusion - I only reached it by having time to think and going back to being more of an IC on my own projects. But having done so, even for a short time, it feels inescapable.

Part of the problem is all the bragging, versus helping others learn. In a race where our industry is upending and many of us feel under threat, perhaps sharing feels like at best a distraction and at worst a threat. Or at least something that should be buried within 24 pages and put behind a paywall. (I really wish every AI thinkfluencer would learn the prompt "make this more succinct" and use it liberally.)

I'm not running a large team right now. I'm building from the ground up, which is easier, and using what I've learned to project onto something I do know how to do well in the old model. So it's easy for me to have opinions. But if I were responsible for a team right now, here's what I would be doing:

Remove bottlenecks. Speed up CI, improve test coverage - The data from CircleCI is clear on impact.

Look for classes of problems that can be solved systematically. Implementation of something that is clearly specified - like the accessibility problems my partner works on, particular kinds of instrumentation, or planned migrations.

Look at ways to automate collective work. Build an agent that takes a first pass on test breakage. Build an agent that looks at your hotfixes or incidents, identifies defect patterns, and provides options for improving. When they fail, feed that information back in and help it improve.

Consider the level of identity threat within your team and look for ways to inoculate people against it. Help them embrace an identity (solve problems, ship things) that is bigger than the skill of writing code. Foster a learning culture - a study from Cat Hicks' lab showed it helps people's resilience to AI skill threat.

I started with solar panels for a reason. AI is extractive - our code trained these models we now pay for. The dopamine hits are real and can reach addictive quality. One person's productivity is not necessarily a net positive in the world.

But here's the thing about extractive systems: you can resist them, or you can figure out how to operate within them while minimizing harm. Industries shift. We adapt or find somewhere else to be.

It's hard to navigate from in the midst of it. I feel grateful to have the freedom to experiment right now, to build from scratch with minimal dependencies.

It's also more fun than I expected. Things are possible that weren't before. Jean and I have our own platform for DRI Your Career that we can iterate on as we expand our ideas. I can have a feedback loop when working alone - for whatever I'm working on - which saves me from unproductive time sinks and the feeling that I'd be bothering someone to ask for help. A different kind of collaboration, but still useful, and available whenever I need it. Timeframes are shorter. I can try more ideas. I'm finding a level of joy in creation I haven't felt in years. The capitalism is still extractive, but I'm adapting. More to come.

If you liked this, you might also like my blog post: The Ground Decisions

What I've been reading

High Agency - The concept is useful (are you happening to life, or is life happening to you?) but the delivery is painfully bro-ey. All male examples, "third world jail" might be an evocative shorthand but is very problematic. Skip to the Wilbur example for the actual insight.

The Claude C Compiler: What It Reveals About the Future of Software - Chris Lattner (the Swift compiler guy) on why an AI-generated compiler is such a milestone. Key insight: AI excels at assembling known techniques and reproducing accumulated consensus, but struggles with open-ended generalization. The value in engineers shifts to composition and novel problems. The definition of "boilerplate" just got a lot more expansive. Particularly loved this:

"Documentation, clear interfaces, and explicit design intent are now operational leverage, not optional overhead."

A Month With OpenAI's Codex - I've been following Steven's Codex adventures on Mastodon and really appreciated this post tying it all together.

"This feels like a first foray into what software development will look like for the rest of my life. Transitioning from the instrument player to the conductor of the orchestra. Incredibly exciting, and deeply terrifying."

Mistakes engineers make in large established codebases - If you've ever joined a large codebase and felt lost, this breaks down the actual mistakes engineers make. Concrete, helpful advice.

The Five Types of Network Connections You Need at Work - Great framework for thinking about your professional network beyond "networking."

Just saying hi - As someone fortunate to be on Q's (the author's) Todoist, I always appreciate the pings. A nice post about the value of reaching out and a system for doing so consistently.

What I'm doing

• Our EM Survival Guide starts next week - it's not too late to join us! The job is harder than ever, and this course is about how to actually do it - get outcomes, make space to think, upskill your team. It's the resource I was desperate to give new managers when I was managing a large team.

• We're wrapping up the first cohort of DRI Your Career which has been so fun. We will run it again in April, and registration is open. We sold out last time, so please don't wait too long if you're interested!

• Vibe coding - wrote up my raccoon quiz adventures and now building something new in social-brain.

• Using a spreadsheet + Claude to track both my impact and how I feel about it - Getting a lot of value from this system for helping me focus and be kinder to myself about how I work across multiple work streams.

• I'm on the Swift Con program committee and submissions are open