Fable Critiqued My Workflow, Then Ran It and Changed Its Mind — Plumbline in practice
With the launch of Fable and seeing its potential, I showed it my agentic coding workflow, Plumbline, and asked it what it thought. After taking a look, it gave me this as part of a response: "Plumbline has two kinds of rails mixed together: rails that compensate for model weakness, and rails that create accountability. The second kind — committed tests per criterion, evidence capture, fresh-eyes inspection, deviation logs — are the product, and they age well. The first kind — micro-steps, 'do not read ahead,' move-by-move Build instructions — were written for a model that loses the plot without them. I don't. Those rails now cost more than they protect." It said that it doesn't lose the plot?! That has seemed to hold true in my testing. This took me back. I responded with "How do you know?" and pointed out that this workflow will probably be used with other models. It humbled itself a bit afterward and agreed it would be helpful. I saw this as an opportunity. Let's have it do its thing with some coding I had to do, then follow my workflow. Ask it again how it felt afterward. I'll put a picture of the full response, but here's the highlight: "The friction is real but proportionate at this size. Checkbox edits, deviation notes, and per-slice commits added maybe 15% overhead; in exchange the audit trail let the inspector verify claims without trusting me. Below this feature size I'd feel the ceremony; above it I'd want it more." It changed its mind. It added overhead, but it found things that it might not have found. The lesson isn't that frontier models are weak. It's that capability and accountability are different problems — and no amount of the first one solves the second. Files and folders aren't a crutch. They're the receipt. If interested you can get the workflow here https://github.com/BytesFromToby/plumbline