In August I spent most of my time inside AI coding agents — the tools, the workflows, and the questions underneath them. A common thread was getting practical: how a minimal CLI agent behaves, how to script and automate one in CI, and how to structure the work an agent does with subagents and spec-driven skills.
Two smaller threads ran alongside. One is the real cost of going API-first with AI and how quickly a workflow can end up locked to a single provider. The other is more low-key but worth checking in on: where Markdown actually stands as a documentation format.
1 AI coding agents, hands-on
These two are about the tools themselves — what a minimal, hackable coding CLI is like, and how a more established one holds up once you start scripting and automating it.
- Pi Coding Agent Review: Minimal, Hackable AI Coding CLI
- OpenCode CLI in Practice: Workflows, Automation, and Pitfalls
2 Structuring agent work
This pair steps back from the specific tool and asks how to organize the work an agent does — isolated context and model routing with subagents, and a skills framework that pushes you toward spec-driven, test-first development.
- Claude Code Subagents: Setup, Config, and When to Use Them
- Superpowers Quickstart: Install, Workflow, and Tryout
3 The cost of API-first AI
This one is about the economics underneath the tooling. Data gravity is the quiet force that pulls an AI workflow toward a single provider, and it builds up in stages you don’t notice until you’re already dependent. The post walks through the lock-in mechanism and a checklist for staying portable.
4 Markdown and documentation
A quieter thread this month, but a useful one to keep an eye on. This comparison lays out how CommonMark, GitHub Flavored Markdown, and Pandoc Markdown differ across tables, footnotes, math, and metadata — and what that means for portability.
If one of these articles is useful to someone building AI systems, backend systems, infrastructure, or technical knowledge workflows, please forward this email or share the link with them.
Thanks for reading, Rost