Where I Am on the Agentic Engineering Ladder
Bassim Eledath wrote a framework for levels of agentic engineering. I mapped my own work against it.
Bassim Eledath published a framework for levels of agentic engineering. Eight levels, from tab-complete to autonomous agent teams. Reading through it, I tried to place my own work honestly.
Level 4: Compounding Engineering
This is where most of my setup lives. I use GSD (Get Shit Done), a workflow framework for Claude Code that runs a discuss, research, plan, execute, verify cycle for every phase. It produces planning artifacts that feed future sessions. When something goes wrong, the fix gets encoded into a CLAUDE.md rule or a skill file so it hopefully doesn't happen the same way twice.
"Hopefully" because the compounding only works if the rules are good. Early on I had vague rules that Claude would interpret differently depending on context. Writing rules that actually stick across sessions took more iteration than I expected.
Level 5: Skills
I have custom skills for repeating work. Blog post generation, project entries, SEO checks. They're markdown files with instructions. No code.
The article's Level 5 talks about MCP servers, database access, CI pipeline integration. My skills are prompt files that tell Claude where to look and how to write. Useful, but simpler than wiring up external systems.
Level 6: Where It Gets Fuzzy
GSD handles the harness side of this. It spawns subagents for execution, runs a plan-checker before plans execute, and a verifier after they finish. E2e tests and build gates catch regressions automatically.
Qumio has cron jobs that run agent pipelines every morning without me touching anything. News gets fetched, summarized, saved, delivered. That feels like it's in the Level 6 territory of "automated feedback loops."
But it's not Level 7. Those cron jobs follow a fixed script. The agent doesn't decide what to work on. It runs the same prompt on a timer. Automated is not the same as autonomous.
What I Don't Have
Level 7 is background agents that plan and execute independently. I don't have that. The closest thing is GSD's executor subagents, but they follow a plan I approved. They don't decide scope or prioritize work.
Level 8 is multi-agent coordination without a central orchestrator. Not close.
I'd place myself at a solid Level 5, reaching into Level 6 in specific areas. The honest answer is that most of my day is still me typing /gsd:plan-phase and reviewing what comes back. The automation helps, but I'm still the bottleneck.