Articles and Posts

With everyone from Anthropic Claude best practice to AWS AI DLC publishing their own AI software development lifecycles, I still haven’t found a frame more insightful than the classic plan-do-check-act (Demming/Shewhart) cycle in use over one hundred years and core to TPS, lean and agile.

1. Plan what you’re going to do.

2. Do it.

3. Check that you actually did what you intended.

4. Learn from and act upon what went poorly or well in the session.

There are Gen AI specific nuances and techniques in each phase such as broad analysis in plan, test driving in do, refactoring in check, micro retrospection in act. But the cycle stands.

In fact, the new minted frameworks that rename these phases consistently group commit(s) with PR after the “do” phase and omit retrospection altogether.

How many times do we have to learn small batch sizes and continuous learning are ESSENTIAL to an adaptive, high output system?

And why can’t we build upon and iterate. Why can’t we provide proper attribution for the brilliant contributions of others in our industry over decades.

#GenAI #TDD #Retrospection #KentBeck #DianaLarsen

The algorithm has shifted. I am seeing more posts of people who use gen ai, are not in an apocalypse while seeing many dangers, and have found that human creativity and practical expertise remain essential to designing and building complex things.

We were always there by the way. We should follow each other while we can because the algorithms will start circulating some other point of view soon out of simple novelty.

Remember when your feed is whiplashed from one nonesense to another that somewhere there is a nuanced conversation about how to use these tools to create something useful, and what we might do to mitigate the risks in using them.

Claude code solving a bug in a single turn. Copilot engaging in a round trip of pointless troubleshooting.

Humans learn and adapt. The Gen AI agent adapts when the human uses that learning to refine its instructions, prompts and context. The Gen AI agent helps guide that process.

I’ve modified my micro retrospection instructions for my PDCA coding process. All the prompts and the skill are on github at kenjudy/pdca-framework.

See full post with Retro prompt: https://lnkd.in/edPM7rRv

#retrospection #rfg2026

I create and build things with other people.

My impact through my work, whether positive or negative, is at an ordinary human scale, mostly upon my family and my co-workers.

That’s the sum of the career I’ve built. It often brings me joy, if not always contentment.

Prof. Cornelia C. Walther is asking us, as leaders, builders, or users, to accept responsibility for our actions with Gen AI today and for how we ourselves will be shaped by those interactions over time. Will we craft and use these tools in ways that deepen or erode our natural intelligence.

Like Jutta Eckstein, she is investigating what we are doing to ourselves with this technology. Both are helping us understand our moral agency. Both avoid the hype cycle without being dismissive of the technology.

Joshua Kerievsky 🇺🇦 thank you for the clarity. I am working on a git actions to help detect code degradation as teams adopt Agentic coding tools. This lays out a significant improvement I can make to it.

So, here’s my bit of pointless AI prognostication. I mean why not take part in the tinpot futurism of the day.

I see even greater value in individuals who both connect work to a useful outcome and perform that work themselves. Mastering both sides of that equation will be valuable; not visionaries waving their hands at robots, well-rounded people who make good things well.

I also see even more value in people who generously mentor and sponsor other people. Learning on the job needs to be an intentional act rooted in values even more than it currently is because it will be even more of a trade off against short term output.

Vision + craft + empathy = sustained growth

I expect society to largely fail to accept this until necessity forces our hand.

On a true vacation with my wife and adult daughter in Ireland.

I know it is a privilege to work for a living and have paid time off time for this kind of experience. An opportunity that may be shrinking as it becomes acceptable for employers to walk back PTO policies they once considered a competitive advantage.

Everything wrong with AI hype in one linkedin comment (below).

“Moving from a $1.5M annual OpEx to a $500/month infrastructure”. GenAI agents wholesale replacing human teams is not validated in any scenario where software has to accomplish real work of consequence for users.

“Remove the friction of headcount management.” Staggering lack of empathy. It also ignores how labor and capital trade power over time.

“Architecting Decisional Integrity into the engineering loop.” What. Are. These. Words?!

The current reality is that working with agents is an emerging practice. The tools are evolving. People are learning. The failure modes are real. Human expertise to navigate them matters.

Speed with direction and discipline scales value.

Speed without direction or discipline scales waste and someone, somewhere, someday will pay the bill.