Articles and Posts

Exposing my code to peers who are more talented craftspeople than I am doesn’t take every ounce of perspective and ego resilience I have. Just much of it… At the end of the day though, given the choice between failing on my own or someone else’s terms, I’d rather just not fail.

Generative AI will help you arrive at different answers to a problem depending on the reasoning path you guide it down. So, don’t dismiss a bot — or a human — for giving you bad answers before first challenging yourself to ask more thought provoking questions.

I rely on generative AI as a reasoning partner and code generation agent. I’m working on creative, interesting, and valuable problems and I wouldn’t be making consistent progress without it.

Here’s some advice.

1. Ask the model to provide terse answers. Give it permission to focus on what’s important.

2. Hold off producing code. Dwell in analysis before jumping to solutions.

3. Talk through what your trying to do. The thread will improve as you ask more insightful questions.

4. Co-create test coverage. Test expectations result in better code and you’ll move faster when you can readily detect and troubleshoot breaking changes.

Make the effort to build machine context, human understanding, and a testing contract and you’ll be flying.

A concern about Anthropic claude is that it was trained, as with other frontier models like open ai, on proprietary content without affirmative consent from the creators. Anthropic is beginning to work with rights holders which is good to see. https://lnkd.in/gM89WcQ8

In my career, I’ve consistently sought collaborative peers with a learning mindset, autonomy, relative financial security, and time with my daughter (now grown).

The tradeoffs are unknowable — ego wants to believe there’ve been some — but the intention has been clear.

My advice for using generative AI lines up with how I’d approach building with quality - period. And that’s the point; flex to the tools in ways that make you faster but double down on what makes you good.

Tried a new prompting tactic:

“3 of 41 tests need analysis. We should analyze before making code changes to avoid regressions. Please help diagnose issues from console output. Refrain from suggesting code until agreed what to do about it.”

Claude in concise mode proposed a series of six tests which I ran supplying back the output with questions until we isolated the specific issue in the code.

Craft derives from three values:

Efficacy - Does what we build meet a need. Create benefit.

Quality - Does it minimize harm - including waste of resources, capital, and human labor.

Beauty - Does interacting with it bring joy.

These three both sustain and vie with each other.

Stats say junior developers gain the most productivity from generative ai.

This is a terrible mistake.

Veteran devs need to use these tools in ways that help them — not the ways they are hyped.

Then they need to spend more time mentoring.

So, we all know what we’re doing.

Reading on responsible AI, I came across this form of ethics washing: presenting a basic legal obligation as a principled commitment to mask real priorities. Something like, “We value user privacy — this is why we enable users to access and delete their data at any time.” 😑