Birgitta B

To vibe or not to vibe? I wrote a new memo about the constant little risk assessments I make during AI-assisted coding, thinking about probability and impact if AI gets it wrong, and if I will be able to detect that. https://martinfowler.com/articles/exploring-gen-ai/to-vibe-or-not-vibe.html

New GenAI memo: What role do developer skills play in AI-assisted coding?
https://martinfowler.com/articles/exploring-gen-ai.html#memo-13

After my mini rant about Vibe Coding yesterday, I'm following it up today with a list of examples of AI "missteps", illustrating how developer skills still very much matter in agent-assisted coding, and ideas how to mitigate that.

I wrote a guest post on The Pragmatic Engineer https://newsletter.pragmaticengineer.com/p/two-years-of-using-ai rounding up 2 years of using AI coding assistants - how they evolved; ways of working; impact I see on speed, quality and team flow; and some thoughts on the future

We recently ran an experiment to explore how far GenAI can currently be pushed toward autonomously developing high-quality, up-to-date software without human intervention, and gather observations about where it breaks down.

https://martinfowler.com/articles/pushing-ai-autonomy.html

I tried to make sense of spec-driven development by looking at 3 tools: Amazon's Kiro, GitHub's spec-kit, and the Tessl Framework
https://martinfowler.com/articles/exploring-gen-ai/sdd-3-tools.html

In the past few weeks, a new surge of "background" coding agents came out. I wrote down an example of using OpenAI's Codex. If you haven't had a chance to look closer at these agents yet, this will hopefully help you understand better what they do under the hood, and which agent category they fall into.

https://martinfowler.com/articles/exploring-gen-ai/autonomous-agents-codex-example.html

I recorded a conversation with Chris Ford about learnings from using computational sensors with a coding agent on an application, in particular static code analysis and test quality
https://www.youtube.com/watch?v=uLWOLmeHOSE

When I code with an AI agent, as soon as I feel like I’m losing control and don’t understand the many changes anymore, I revert to the last checkpoint where I felt comfortable. I wonder what that will be like for the delivery managers and POs of the future - what will THEY do when things around them change so fast and erratically that they feel like losing control?

One of the challenges with service templates is that once a team instantiated a service with a template, it’s tedious to feed template updates back to those services. I wonder if anchoring AI agents to a template or reference application could help make that easier? https://martinfowler.com/articles/exploring-gen-ai/anchoring-to-reference.html

Good scope management is the linchpin of agile software delivery. Generative AI is like a scope chaos monkey, generating more code, more story details, more requirements than necessary

I've seen a surge of discussions recently about what feels like "vibe coding extreme", about large AI-generated change sets that are impossible to review by humans, leading to speculation if we even still need to care about the code in the future. I still care, and expect to continue to care, especially if I have to be on call for it https://martinfowler.com/articles/exploring-gen-ai/i-still-care-about-the-code.html

Oh! Committing a vibe coding session to a "real" codebase is like reviewing and refactoring a huge pull request by a beginner programmer. Actively driving and steering an agentic coding session is like pairing, where you give and get just-in-time code review on every small step.

@iscarlosmolero@mastodon.social that’s what I do indeed when I lose control of AI, but what does a delivery manager do when everybody around them starts creating more throughput? I guess they can take the AI licensed away 😊

The way the term vibe coding was coined, became a meme, and then lost all meaning within just six or seven weeks is wild...

This is how I understood the original tweet by Andrej Karpathy:
- You don't look at the code.
- You just tell the AI what you want, maybe even using voice.
- You iterate via chat until it does what you meant.
- It's chaotic, fast, fun.
- "It's not too bad for throwaway weekend projects."

🧵