01
We mapped where the hours actually went
Two weeks of time-tracking with the editors, not a workshop with their managers. The bottleneck was not writing or editing — it was the twenty minutes of reformatting after every single piece.
Verse
- Client
- Verse
- Discipline
- AI Workflow
- Year
- 2024
- Services
- AI Workflows, Internal Tools
03Verse
Verse came to us with a request we hear a lot now: make the editorial team faster with AI. They also came with a fear we hear just as often, mostly unspoken — that the project's real purpose was to make the editorial team smaller.
The problem
Publishing an article took fourteen steps across five tools. Roughly nine of those steps were mechanical: reformatting, resizing, tagging, cross-posting, checking links. The remaining five were judgement, and judgement was what the editors were hired for.
We said at the kickoff that we would not automate the five. That constraint shaped everything, and it was the reason the team engaged with the project honestly rather than defensively.
What we did
02
We automated the mechanical nine
Three assisted steps replaced fourteen manual ones. Every automated output lands as a draft an editor approves, never as a published artefact. The human stays in the loop by design, not by policy.
03
We built the guardrails before the features
Evaluation sets, regression checks on tone, and a rollback path. An assistant that is wrong 4% of the time is unusable in publishing unless you can find the 4% before your readers do.
Where it landed
Hours per published article fell 71%. Output roughly quadrupled without a change in headcount.
Zero editors were replaced — the number the team cared about most, and the reason they now ask us for more of it rather than less.
Results
Hours per article
Editors replaced
Workflow research
Aktays
Assistant design
Aktays
Automation build
Aktays
Evaluation framework
Aktays + Verse editorial
Next project
Kavun
A food brand with an appetite for white space