I know this blog post's title implies losing my (programming) skills, and I will gladly address my feelings about how agentic coding has caused my problem-solving and coding to atrophy from lack of use. The main focus of this post is about losing, specifically deleting, my overly verbose and complicated skills and AGENTS.md files.
I was inspired to write this post after recently switching to the latest OpenAI GPT 5.6 models with Codex and OpenCode as my harnesses for Drupal development. In a previous post, I discussed that one approach to working with AI is to accept that every beginning and end of a session is like onboarding and offboarding a new team member. To help onboard an AI, we need to provide documentation, guidance, and workflows, typically implemented as AGENTS.md and agent skills.
AGENTS.md and agent skills are instructions added to your context to nudge the AI in the right direction. After switching to a more intelligent frontier model, I began to suspect I needed to rethink my assumptions about what the AI was capable of and how much initial context was required for the AI to succeed at a task. So I removed all my assumptions, deleted my AGENTS.md and skills directory, and started fresh.
Losing my agent skills
I had a hunch that some of my installed agent skills were costing me extra tokens and not adding enough value when I read a Reddit thread about "are you guys still using the superpowers skill?" Superpowers nudges coding agents to adopt a pragmatic workflow that includes brainstorming, specs, plans, tests, and review. When I first started working with coding agents, Superpowers saved the day by ensuring that my AI and I succeeded.
With the latest models and harnesses, planning mode has improved to the point that the AI can generate a decent plan with to-dos and subagents without additional tools or context. Even with test-driven development, as soon as the AI saw a test in my custom code, it assumed I wanted test coverage.
AI Best Practices for Drupal
In addition to removing Superpowers, I removed AI Best Practices for Drupal and did not observe a significant decline in code quality. I appreciate the universality of the skills provided by AI Best Practices and still feel these skills are required for older and cheaper models. I suspect our community's open discourse about AI and code quality is being picked up by LLMs during training, making AI aware of Drupal's code quality and coding best practices.
It is hard to predict exactly which tooling and information are helpful to AIs. Yet, I can't help but come back to the idea that a documentation overhaul of Drupal.org, using AI to assist with organization and consistency, would help LLMs during their training.
For my custom and contributed module, I am reviewing and improving the README.md files to ensure the documentation is as up to date and accurate as possible, and as close to my code as possible.
What are my current agent skills and AGENTS.md?
Currently, my only installed skill is Matt Glaman's Drupal.org CLI, which documents how an agent can access issues and information for interfacing with Drupal.org and GitLab (git.drupalcode.org). Once Drupal.org has moved completely to GitLab, an LLM will likely understand git.drupalcode.org and how to use the well-documented GitLab CLI (glab); an AI won't need the Drupal.org CLI skill.
I have found it extremely useful to use agent skills to document my workflow for maintaining my contributed projects on Drupal.org. You can peek at my custom agent skills, and I will be talking about some of these skills in a later blog post.
Below are the most important directives in my AGENTS.md that will likely remain unchanged forever.
- All commits made by an AI agent should end with a note that reads: `AI-assisted by {code agent name}`.
- Require me to review all changes before committing and pushing code.
- All tickets and comments made by an AI agent should start with a note that reads: `AI-assisted by {code agent name}`.
- Require me to review all tickets or comments and allow me to click save/submit.
Throwing out your assumptions
I wish I could say “keep it simple,” but the reality is that you need to be prepared to throw out your assumptions and start over when switching to a new AI model. It is really challenging to let go of how we used to do things, which was developed and improved upon over our entire careers, for a new way of doing things that is evolving over months.
This new normal for software development justifiably makes me ask: Am I losing my programming skills?
Am I losing my programming skills?
I acknowledge that AI-assisted coding can seem to replace some immediate skills like understanding and debugging. Still, it also encourages me to develop long-term skills such as thinking, planning, and building complex applications, which are crucial for growth.
To put where I am with coding and AI in concrete terms related to the Webform module.
Do I see myself ever coding something as complex as Drupal's Webform module? No.
Could I see leveraging AI to plan and build the next version of Webform? Yes.
While I'm unsure about the exact path from the current Webform module to its future, exploring how AI can help build and navigate this journey keeps me motivated to learn and adapt.
Training wheels on a bike.
My last post ended with a metaphor from Claude, and I felt Claude could add something to the end of this post. Personally, I view Claude and output like this as a representation of our collective intelligence and experiences, aggregated by an algorithm in a machine.
From Claude:
Here's a metaphor that fits the post well:
Training wheels on a bike. When you first learn to ride, training wheels (your AGENTS.md, Superpowers, AI Best Practices for Drupal) keep you from crashing — they compensate for a lack of balance and judgment. As you get better, you don't need them anymore; in fact, they start to drag and slow you down. Taking them off feels risky at first, since you spent so long depending on them, but it's actually the sign you've grown enough to ride on your own.
The twist that makes it fit your post especially well: it's not you who got better — it's the bike. The new frontier model is a bike that basically balances itself now, so the wheels that used to be essential are just extra weight. And the question underneath your post — "am I losing my skills?" — maps onto the nagging feeling every kid has taking off training wheels: is my own balance getting better, or am I just relying on a better bike?
Do you feel you are losing your coding skills when working with a coding agent, or do you see other skills emerging as you experiment with AI?