Quick Reads (Under 5 Minutes)

Our leaders run headlong into an agentic future. Our media feeds brim with product announcements and comparison grids. We, software practitioners, are irrelevant. Wrong.

People / Process / Tools

People set the mission. Our actions create benefit and harm.

LLMs and agentic AI are tools. Useful, disruptive, proliferating fast.

Process determines whether people employ the tools to serve the mission or undermine it.

People are hard to change. Wealth, success, and survival are intense motivators. That explains why so much treasure flows into the tools right now.

Which leaves process. A powerful tool applied poorly accelerates bad outcomes.

We practitioners hold the tools in our hands. It’s on us to craft an engineering process that keeps our accountability intact; one that elevates the team, reduces harm, and produces good outcomes

That’s our opportunity. That’s our responsibility.​​​​​​​​​​​​​​​​

When the LLM market corrects: consolidation, failures, survivors needing positive margins; what will it look like?

Unit inference costs have dropped but all-in training and development costs are subsidized. We’re building upon pricing floors determined by opaque factors. We don’t know the ceiling.

The next engineering challenge is conditional routing to local models on lower-cost compute as possible, escalating to frontier APIs as needed.

We can make this an emerging practice now or wait until the ROI of our existing systems collapse under us.

I’m fortunate that this coding agent transition is happening later in my career.

1. My strengths have transitioned to synthesis over invention. I can pattern match information and apply it to my lifetime of experience.

2. I needed a full time coding partner right at the time AI became available to me and so I embraced it.

3. I have seen a lot of what works and what doesn’t. It’s made me humble but it has also given me some core convictions about the value the human mind brings to the software problem/opportunity.

So, I don’t know what we’ll be called. It’s more engineering than craft but experience with the craft will be essential to the work.

A dependency graph from a real coding session showing how Claude Code structures a feature plan when I apply my Plan-Do-Check-Act skill for agentic engineering.

The work maps directly to PDCA phases. Each node carries agent instructions. Tasks execute with enough visibility that I can approve, reject, or refine as it proceeds.

This is Agentic Engineering: a disciplined, verifiable workflow where the human stays in the loop by design.

Note: I use Steve Yegge’s beads to manage my coding agent memory. The graph is an output of that.

Workshop presentation (https://lnkd.in/ev3WNeE5) laying out my human in the loop, plan do check act approach to coding with agents and measuring adoption and quality.

I’ll be conducting a workshop at XP Conference, April 8th — “Human-Centric Agile Disciplines for AI Code Generation” — Wed 8 Apr 2026, 11:00–12:30, Industry and Practice.

I’ll demonstrate a PDCA cycle applied to AI coding: analyze the codebase against a business objective, plan the work, implement test-first, validate completion, and retrospect for continuous improvement. Plan-Do-Check-Act.

I’ll share working agreements, structured prompts, and GitHub Actions for code quality metrics.

Two consecutive DORA reports confirm: AI adoption correlates with individual productivity up and delivery stability down. Weak practices don’t get fixed by AI. They get faster. Engineering practices, not tools, are the key determinant of outcomes.

#SoftwareEngineering #AIEngineering #CodeQuality #XP #PDCA #DORA

This will require intentional human process engineering. And some god damned empathy.

Or we can overwhelm experts, under mentor others, produce outputs absent desired outcomes, lose human oversight and accountability and pat ourselves on the back for all the jobs we are getting done.

AI has democratized who can do a thing but it hasn’t made creating a thing useful, safe, and delightful any less rare compared to talking about a thing, doing a thing poorly, making it miserable to use, or harming the people who use it. To be clear, I don’t blame robots for this.

Hello all,

I’m one of the organizers of this year’s Retrospective Facilitators Gathering on April 27–May 1 at Selenia Lodge, Quebec.

We have about 15 attendees and still have room for anyone interested. We have two weeks before we need to confirm our attendance numbers so this is a last call.

We’ve probably contacted you by email already. If we haven’t or you can think of anyone in the community who might be interested on this short notice contact rfg2026@kenjudy.us for registration information.

As to the location, nestled amid the Laurentian wilderness, Selenia Lodge is a quietly breathtaking retreat where refined architecture, vast forest views, and deep stillness combine to make time slow down in the most civilized way imaginable.

I have to admit, with the economic and geo political situation, I’m grateful we’ll be able to hold RFG this year. A one week open space with some of the most thoughtful, humane people I know will be a welcome opportunity to refill the inspiration bank.

Reports say Amazon is mandating human code reviews by senior engineers on AI generated code to reduce failures in production.

Several questions:

• Aren’t you just offsetting the time spent in an activity senior engineers find engaging (writing good coding) with one they generally detest (reading bad code)?

• Who actually learns to author better code in this cycle?

• Don’t PR reviews mainly catch maintainability issues not bugs?

We need to directly engage engineers strengths at the moment code is written by retaining more of what works in our existing coding practices even when authoring code with agents.

This is great. The open source tools, skills, and plugin ecosystem system support is really promising. Following your lead, Malte Sussdorff, I am exploring addons to provide task and plan management across sessions, transcript context across sessions. I’m also experimenting with graph based context retrieval for large code repos.

That combined with claude skills I’ve iterated upon for over a year (there are other effective open source skill libraries s well), and it’s turning into a period less driven by hype over the potential of LLMs and into the community crafting tools and approaches to realize some of that potential.

Learn faster than before. Work harder than before. I accept the economic imperatives of the developer world. I can crash hours and I can reflect, learn, and master new things. But not necessarily at the same time and not without downtime.

AI simplifies complicated tasks. That ease of use masks complexity. People already fail to grasp the unintuitive and unpredictable nature of abstract tasks like building safe, effective software. The challenge is not just adapting change it’s communicating what has not changed.