👋 Hi, I’m Jeff. I write weekly about lessons I’ve learned leading product teams and working to lead well in chaotic environments, drawing on 25+ years of building.
1:1 exec coaching: I’m working with a small number of senior product leaders one-on-one as a coach. If that’s you or someone you know, ping me.
⏱ Read time: 2 minutes
I’m writing from Paris, where I’ll be until the end of the month with equal parts work, relaxation, and exploring. Two things I’ve been noticing.
———
First, coffee culture and how digital shows up. We’ve been exploring coffee and café culture on the road.
We’re noticing two distinct modes:
1: Coffee shops or cafés with a single giant community table in the back and lots of people with laptops and small teams working in small groups or even dialing into conference calls. All over coffee. It’s almost like a coffee shop as WeWork model. This scenario is mostly Amsterdam.
2: Coffee shops or cafés with explicit “no laptops” signs on any flat surface. Everyone is reading a book or, gasp, talking. We’ve noticed this mostly in Paris or Amsterdam where the coffee is especially high end.
It’s worth watching how everyday people here treat their “third place.” Charlene wrote about this if you’d like to go deeper.
Either way, the coffee has been excellent.
———
Second, train rides and long coffees mean lots of time to read. I came across this New York Times article by a reporter who covers white-collar work. Two themes stood out.
AI compresses execution, not decision-making. Tasks like coding, writing decks, and drafting strategy are getting done faster or automated entirely. But decisions, alignment, and prioritization still require humans. As execution gets easier, the hard human work—judgment, empathy, alignment—will rise in value.
Social skills are becoming a real moat. Persuasion, storytelling, relationship-building, and trust matter more as the barrier to getting work done drops. Winners may be generalists who can influence, lead, and tell stories. Building is getting easier. Direction and narrative are the differentiators.
As I experiment with tools and automation, I keep coming back to the human conversations. The tools are helping me do things faster, better, but I still need to be good at dealing with people. Not only is it not going away. It’s becoming even more of the whole game.
What are you seeing?
I think an important realization for software development and AI is that AI compresses development, not code comprehension. A developer and YouTube influencer I follow once said that _the_ thing that made him tremendously valuable to one of his former employers is that he could sit down with stakeholders and iterate versions of software prototypes so fast that he often significantly changed the direction of a project before it even became an official design proposal. (My comment: Isn't that that supposed to be how agile development works?) AI has the ability to let more people do that exact thing -- iterate code at greater speed.
There is another half to the process that the developer acknowledged: with the mental map embodied in a prototype, he, as a software architect, could help drive the project to completion by understanding what parts of the prototype are incomplete, what parts need to be reworked, and what is needed for the application to function at scale. This is where AI is less helpful, because it only understands what has seen in its training data and the contents of the prompt. In fact, it can hinder (or significantly alter) this part of the process because it can generate thousands of lines of code quickly. For humans to comprehend those lines, they need to longer amounts of time reviewing the code to ensure that implementation choices the AI has made are acceptable. This reduces velocity. To do otherwise can turn a product code base into something more like legacy code -- parts of the system written in the past, that work (or at least pass tests) but no current employee fully understands.
There are those who say that this changes how software engineers should be trained. To keep the code comprehensible, developers need to become better design spec writers. Perhaps design spec readers is the better term, because AI can be prompted to write or modify existing specs before actually generating code. Engineers make sure that the AI shapes code correctly by critiquing and modifying the specs. This steps away from the "code as source of truth" approach that has prevailed in industry for a long time. While the best code should always embody that principle, a good spec makes an effective cheat sheet to understand how a particular 200 lines of code should work in the context of a much bigger picture.
The other part of it is testing. AI written code can be created so quickly that the bare minimum test for its acceptability is that it passes any tests bundled with it. Evaluating the scope and completeness of those tests is something human engineers will need to do.