Short-form insights you can read in under 5 minutes.
For tech leaders trying to manage AI adoption who feel stuck in the nay sayer role…
The data has been out there from GitClear, Google Dora, and other research. Now it’s entering the business press. MIT Sloan might give exuberant mba types some proof points.
Now is a time for experimentation and pilots with measurement of quality of outputs and customer outcomes not just speed of activities.
Can AI help you write good code, not just more code?
Most teams are using AI tools without structured practice leading to a day of reckoning as the inherent qualities of agent generated code leads to maintenance and support issues.
I’m proposing a human-supervised Plan-Do-Check-Act approach based on the Agile practices we already know work to help us wrangle AI away from technical debt generator towards a tool for more sustainable development:
Plan: Use AI to analyze codebases and prevent architectural drift
Do: Generate code with structured prompts and active oversight
Check: Validate against an explicit definition of done
Act: Retrospect to continuously improve human/AI interactions
I discuss my working agreements, prompt templates, and measurement strategies designed to keep humans engaged, empowered, and accountable.
Can we make AI work for our codebases instead of against them?
Read details of the framework 👉: https://lnkd.in/dVV6xt2N
#AgileDevelopment #AIProductivity #CodeQuality #SustainableDevelopment
Frontier models convince you they are correct and that you are brilliant for the same reason a dog conveys that it loves you. Adaptation and training. The main difference is that a dog can be sincere. An AI model at the most fundamental level doesn’t care.
I embrace skepticism around facts but recognize the potential of frontier models to enhance research. By using tools like Claude, I can validate and summarize information effectively. My approach combines AI assistance with thorough understanding and editing, ensuring my voice shines through. It's crucial to teach others how to leverage these tools responsibly, rather than prohibiting their use.
I believe successful legacy modernization is crucial for growth. As systems age, they demand costly maintenance, hindering innovation. The strangler fig pattern offers a safer approach by gradually replacing old systems. With AI-powered analysis, we can create a roadmap that helps architects extract the right components efficiently, ensuring we deliver value sooner and minimize risks.
Code generation without capable oversight by an experienced software developer is like letting go of a disk saw. Just because the rotating blade pulls forward doesn’t mean it understands what you need and how to do it safely.
Dear tech executive, If you don’t know what you want or can’t articulate it coherently then speed isn’t your problem. Assuming AI code generation solves your problem is dead wrong.
“My engineering team consistently delivers the wrong thing.”
That’s a failure of definition, not implementation. Leadership either doesn’t understand the opportunity well enough to communicate it, or there isn’t one to begin with.
“Fail fast, learn cheap” works if you have something to learn from the experiments. If you can’t clearly articulate what you want, you won’t extract useful lessons from what gets built. Just building the wrong things faster isn’t the cure for that disease.
I recently encountered a situation where my AI code assistant created a third way to validate emails, leading to confusion and inconsistency. I realized that while AI can establish patterns, it often deviates, resulting in a maintenance nightmare. To avoid this, I now instruct the agent to analyze existing code and ensure consistency before making changes. This proactive approach saves time and effort in the long run.
Another helpful eyechart of the AI landscape (joke)
I’ve seen firsthand how developers often spend more time deciphering legacy code than actually writing new code. With research showing that 58% of our time goes to program comprehension, it’s clear that understanding complex systems is crucial. Our approach uses advanced tracing to document dependencies, enabling us to rewrite legacy applications efficiently while minimizing risks and costs. Let’s leverage AI to streamline this process.
Most teams have zero visibility into their actual AI development patterns.
My AI development patterns over 30 days:
📊 50 commits analyzed across 4 feature branches 🚨 46% were over 100 lines changed or added 📁 20% of commits were across 5+ files 🧪 58% contained changes to both test and code files
Even with a disciplined practice, I need to reduce the batch sizes of my code changes.
How to measure yours: https://lnkd.in/d9FnkqhU
#AI #SoftwareDevelopment #CodeQuality
Don’t just let the AI code - actively question its architectural choices, hold it accountable to existing designs, and challenge unnecessary complexity.
I just wrapped up a 3 hour coding session. I instructed the agent to break the work into planned steps and test drive each step. When it had completed a tested increment, we proceeded to the next. This resulted in 27 separate commits.
Even still, I made at least eight direct interventions where I stopped the agent as it generated code with questions like:
“Why mock file operations in integration tests? Use temp directories instead”
“We already built those convenience methods - what did we miss?”
“Why mocks over dependency injection?”
“Do your changes actually match the schema?”
The result was a cleaner architecture with dependency injection. Integration tests using actual file operations vs mocks. Schema compliance verified at each step. Code using existing design patterns instead of reinventing the wheel.
The most valuable interventions weren’t about syntax or bugs - they are architectural challenges that prevent technical debt and follow established patterns.
#SoftwareDevelopment #AI #CodeReview #CleanCode #TechnicalLeadership