Tag
#engineering
11 posts
- The Work That Doesn't Demo AI-assisted The visible part of agentic coding is the build. The trustworthy part is what happens when the build gets messy.
- The Landing Phase AI-assisted As eforge moves toward public use, the hard part has shifted from generating code to landing work safely inside real development workflows.
- Eval: eforge harness backends AI-assisted Two eforge backends under identical model routing. Token cost replicates across runs. Behavioral metrics don't - and a post written after one run would have been wrong.
- The Open Source Alternative AI-assisted Factory AI ships Missions. I'd already built the open source version - from a different direction, with a different answer to what quality means.
- The Flinch AI-assisted Everyone flinches at the moment they're supposed to let go of the code. The skill that matters now is planning well enough to trust the handoff.
- What a Cheaper Model Actually Costs AI-assisted The policy that governs my primary dev workflow lives in an X reply. That's its own kind of vendor risk.
- The Hill AI-assisted Information asymmetry, the bitter lesson, and what stays durable when models keep getting smarter.
- The Handoff AI-assisted Abstracting away the build phase: learning to plan full time.
- What the Diff Already Knows AI-assisted An LLM reading a git diff cold can classify the nature of the work - feature, rework, refactor - with a granularity that file-level metrics can't touch.
- What the Numbers Show AI-assisted I've been tracking machine hours, session concurrency, and code rework across my agentic workflows. The most useful thing the data surfaced wasn't about the machines.
- From Assisted to Agentic AI-assisted AI tool adoption is nearly universal among engineering teams, but the workflows underneath haven't moved.