what did you observe that made it better?
It felt a WHOLE lot more like a senior developer, asking the right questions, working off of best practices, and so on. So smooth.
One thing I *really* liked was that at points in development, it would initiate having “3 reviewers” evaluate the code from different perspectives, and those reviewers would pass the feedback to Claude Fable 5.
This is an example of the “conversation” from Fable 5:
—
Spec written. Now an adversarial review pass — three reviewers with different lenses, since this document is about to drive the whole build.
Ran workflow review-spec
While the reviewers work, let me save the project to memory so future sessions pick it up.
Saved a memory
Background workflow completed
Adversarial review of the AutoAuthor Sheets SPEC.md from three lenses 8m 45s
The reviewers came back with substantive findings. Let me read the full set.
—
Another bit of conversation with Fable 5:
—
Phase one is done: the plugin and HMAC contract are built, adversarially reviewed, hardened, and fully green — 48 PHPUnit tests (85 assertions) + 25 Vitest tests, both suites asserting against the same golden-vector fixture, plus a clean syntax lint across every plugin file.
—
Wild, right? I’ve never seen an LLM initiate having multiple “adversarial reviewers” play a role in development like this.