Tools and Strategies for Agentic Development (Into the Hopper Podcast)

It's been a busy week. I've been hard at work building an internal agent/LLM platform at Workhelix. But I was able to step away to have an introspective conversation with Tim Hopper on his Into the Hopper Podcast. It was a great opportunity to reflect on my use of spec-driven development, Steve Yegge's Beads for tracking tasks while using coding agents, and my work and research into managing agent memory. I took some time this weekend to share some of the code snippets and ideas I use every day in the Code and Tools section of this site.

Tim and I discussed how AI has changed our work patterns. We often hear that coding agents and GenAI are instilling bad habits in software engineers. To my surprise, I've actually become a much more intentional planner with agents. Pre-agents, I would be told I had to plan out every step thoroughly, yet beyond the architecture, I hesitated to over-specify my projects.

With agentic development, that's no longer the case. Agents have pushed me to become a firm believer in spec-driven development—the idea that larger projects should have a thorough architecture doc, a tasks doc fully breaking down different epics into atomic tasks, and thorough session logging. This shift has made me a much more intentional coder, and I believe my work has benefited from it.

Here's an example of my spec-driven workflow using Beads. See more in Code and Tools.

Bash
Copy
1# Create tasks with dependencies
2bd create "T1: Project Structure" -t task -p 0 -d "Set up package structure" --json
3bd create "T2: Core Client" -t task -p 1 -d "Implement client class" --json
4bd dep add <T2-id> <T1-id> --type blocks
5
6# Find ready work and claim it
7bd ready
8bd update <id> --status in_progress
9
10# Close and sync
11bd close <id> --reason implemented
12bd sync -m "Close T1: Project Structure"

One thing I'm curious about, with the advent of agentic coding, is how the complexity, maturity, or elegance of codebases is shifting. I haven't seen much research into this subject, but I intend to speak to other veteran software engineers to see if they've also improved some of their habits.

If any of these subjects interest you, I've embedded Tim's podcast below.