Articles and Posts

I learned at XP that Banco Bradesco is using gen AI coding agents to perform Function Point Analysis to arrive at some more quantitative basis for assessing the size of a vertical slice of work (not the effort). Definitely yoinking that into my own work. The “new” software engineering is definitely giving off some 80’s SWE swagger. Thank you Dalí Freire, Ph.D. et al! #XP2026

I don’t care whether you think the future of software is humans, humans and gen AI, or autonomous agents. Because what I do about it is the same regardless.

As of now, the approaches co-exist. Which to use depends on the nature of the possible harms and value inherent in the software to be written.

So right now, the job is to learn how to do each type of software creation well.​​​​​​​​​​​​​​​​

So, the evolution of work with Gen AI will take longer than we think because it entails change at all levels of human systems not just replacing humans workers. All barriers to change apply.

Including that aggressive moves to get ahead of the curve are motivated by short term financial incentives and drive actions that work against meaningful adoption.

Because a large portion of these leaders lack awareness of the actually utility of Gen AI and are misapplying it or simply talking a good game about it.

Which, actually, makes them more replaceable with Gen AI than the workers they want to toss into unemployment.

Got it.

On the plane back from São Paulo. Distinctly US American crowd. No more “desculpe” only “sorry”.

I’m at #xp2026 and this very much a theme of conversation. The questioning is interesting. Certainty is not.

Disruptive change happens. But people, organizations, cultures resist change. And the premise that human systems and humans themselves are rational actors is dubious.

That means that the chances thing will drastically change because it makes sense for them to is dubious. And the assumption that it will make sense for them to drastically change is also up for grabs.

I’m more certain that we don’t know how things will be different as I am that they will be different.

What we need to do is try to be exceptional. Try to act rationally. Try to temper our fear, greed, and certainty. And just know that if we keep learning, keep contributing, challenge ourselves to act ethically, and navigate the infuriating fear, greed, and certainty of others that we’re doing what we can.

And also, people need to learn how to edit the drafts they generate with llms. That said self promotion whether well or poorly executed is the medium. It’s the oil in the paint of a linkedin post.

Reflecting on Jutta’s talk, I want to take my practice to the next level by acknowledging the polarity of leveraging human learning and use optimization to both master the tooling and reduce the impact on the planet.

Automate aspects of an existing structured process with a knowledge base 👍

Automate without an existing process or knowledge base 🫤🎲🎲

I’ve posted my presentation slides for my workshop at XP-Conf including instructions on how to run the coding session with my PDCA claude skill yourself.

https://lnkd.in/ev3WNeE5

#xp2026

Can anyone refer me to a project that is attempting smart routing to different models (including potentially local models) based upon the nature of the prompt? Or is everyone doing the model selection by hand prompt by prompt?

I’d like to find a way to be more cost effective with token usage, like the way claude code itself situationally routes requests to haiku or sonnet.

Both of my workshop sessions complete in the first half of the first day at XP Conf Brazil 2026. Now I can just meet people and learn. I’ll post my presentation tonight.

After live coding with agents in front of some very good developers and absorbing talks from researchers on what that work might become, I ended the day watching Hernan Wilkinson’s make the case that passion for learning and craft pursued for their own sakes creates beauty. #xp2026

Data is pretty clear. AI amplifies existing weaknesses as readily as existing strengths.