Step 3 · Discovery

Two researchers, licensed to bring bad news

A competitive researcher and a user researcher run in parallel. The single most important thing about this stage: the user researcher found no organic demand for my idea, said so in the first line of the findings, and that survived to the final verdict.

What this stage wrote to disk

2 agents · 2 documents
research/user-research.mddemand evidence, affected segments, behavior chain, adoption friction126 lines
research/competitive.mdAI-assistant comparators, adjacent-category patterns, precedent for each sub-feature154 lines

The user researcher tags every piece of evidence by what it actually is, so a reader can weigh it:

[quote] real, sourced user voice with a link [roadmap] a signal about Anthropic's own plans [analog] evidence from a different product [research] a cited HCI finding [inf] the agent's own reasoning [gap] evidence looked for and not found

The honest headline

research/user-research.md · user-researcher

"For the specific ask — reorder/hide/default-landing on the mobile app's six-item drawer — I found no organic user-voice demand. Stated plainly: absent, in the sources I could reach."

What it did find is adjacent: the identical complaint shape, filed repeatedly by the same user segment against a sibling Claude surface. These are real GitHub issues it cited, with dates:

anthropics/claude-code #72126openfiled 2026-06-28

"Allow manually reordering sidebar groups in the desktop app… the auto-ordering buries the groups I use most."

anthropics/claude-code #70104openfiled 2026-06-22

"Allow reordering/pinning of session groups in the sidebar… a stable, user-controlled order makes the sidebar much easier to navigate."

anthropics/claude-code #25974closed · duplicatefiled 2026-02-15

Session pinning and reordering; closed as a duplicate, meaning a second independent report of the same need existed.

research/competitive.md · competitive-researcher

"The sharpest data point in this research isn't a competitor — it's Anthropic's own product family… That is the identical complaint the mobile feature idea is built on, filed by users of a sibling Claude surface, unresolved as of this writing."

The two findings that shaped everything downstream

Customization is a minority behavior, except for exactly this segment

The user researcher cited Jared Spool's Microsoft Word study: "less than 5% of users had changed any settings at all," but "programmers and designers were the primary exception, frequently customizing 40-80% of available options." That split maps onto this feature's target segment, and it became the PRD's discovery requirement: an affordance nobody finds helps nobody.

No AI assistant has shipped this, and one sub-feature has real backlash precedent

ChatGPT's community has asked for sidebar reordering since January 2024, unshipped. And for "set the default landing screen," the competitive researcher found no positive precedent anywhere, plus documented backlash when products force a default-view change (X/Twitter's "For You," the M365 Copilot redesign). That finding triggered a kill condition in the next step.