Field report methodology
The trends report uses anonymous community survey responses, normalized into broad buckets so the public read is useful without exposing small cohorts or individual setups.
Responses come in anonymously through the public form, which stays open year-round. The survey covers role and company size, the stack (editor, agent and its autonomy, primary model, model access, MCP servers, rules files, and the skills and plugins people run — autocompleted against the field guide), spend and who pays for it, the biggest blocker, and how much of the work AI now does.
Submissions are moderated before they count publicly. Obvious spam, duplicates, and unusable rows are rejected; only verified rows reach the report.
Any bucket with fewer than five people is suppressed or pooled into 'Other', so no small cohort or individual setup is ever exposed. We publish aggregate patterns, never raw rows.
Budgets are normalized to USD at submission using a fixed exchange-rate snapshot (currently May 2026), then converted back at display time when readers switch currency. The charts emphasize medians, interquartile ranges, and ranked shares — more durable than exact point estimates in a growing sample — and every cut honors the same five-response floor.
Collection is continuous, but the report is published on a rhythm: a quarterly flagship edition (each response is stamped with the quarter it was collected in) plus a lighter monthly index pulse recomputed from a trailing window. A frozen core of questions keeps editions comparable, so quarter-over-quarter movement is real rather than an artifact of a changed question. Thin cuts widen their window before they’re shown.
This is a self-selected sample — the builders in ANALOG’s orbit, not a census — so read it as directional. Authority comes from consistency and transparency over editions, not from claiming to represent everyone. Revalidation is tied to moderation events, so public charts update after new rows are accepted.
Back to the report