How AI Workflows for Digital Product Design has changed over time. Each entry preserves an earlier published version; the live article stays canonical.
2026-09-20 version
The version published before the 2026 audit. That audit corrected the portability section's claim that Claude Code does not read AGENTS.md natively (it does, from v2.1.277, unless a CLAUDE.md sits in the project or above it), refreshed three gstack appendix rows that had drifted (/autoplan, /cso, /pair-agent), dropped a wrong year from the A2UI mention, and added TypeSafe’s Jev, the first System One model, to the four places the playbook already made room for a cheap, typed verdict: the loop-engineering verifier, the model-routing tip, the eval rubric in the metrics-driven appendix, and the generative UI card, where Vercel’s AI SDK now exposes the call as experimental_evaluate. The same day’s editorial pass repaired the run-on sentences left by an earlier em-dash removal, retired the “this is new” framings that had aged (loop engineering, context engineering, auto memory, AGENTS.md), unified the Jev latency figure, moved the branch-name examples off a 2024 date, made three prompting cards harness-neutral, normalized spelling to US English, and cut the duplicated screenshot and reframing advice back to a cross-reference. Later the same day the gstack appendix’s 26-row skills table was cut to the four skills the section uses, with a pointer to the README for the rest, and the brand.md section gained Vercel’s measured design.md result.
2026-09-08 version
The version published before the 2026 audit. That audit corrected the gstack appendix's one-line description of the /browse skill. gstack v1.81.0.0 changed which browser the skill drives: it now drives the Aside browser first and falls back to its own bundled Chromium, so calling /browse “the headless browser” no longer describes what a reader gets by default.
2026-07-13 version
The version published before the 2026 audit. That audit corrected two capability claims against the Claude Code documentation: Agent Teams is experimental and disabled unless you set an environment variable, which the frontier appendix presented as an available mode; and auto-memory is no longer a research preview but is on by default, which changes how much weight the passage puts on it.
2026-07-09 version
The version published before the 2026 audit. That audit corrected the Agent Teams description (Claude Code has no built-in council or voting mode; the frontier appendix's “council mode where multiple agents debate and vote” became a critique you compose from teammates, and the unverified “7x tokens” figure was softened), named the Linux Foundation's Agentic AI Foundation as the steward of AGENTS.md and MCP, and added Google's A2UI protocol and Vercel AI SDK 7 to the generative-UI section.
