Revista Anfibia: A Slack chatbot for routine visual content
Automating the flyer queue — while keeping AI out of article imagery, after a map-error scar the year before.
Problem statement
Anfibia is more than a magazine — it also runs a podcast production company, a non-fiction theatre lab, a large education and training area, and a membership programme — and every branch needs visual content for social media: flyers, promotional images, event invitations. All of it lands on an art department of two to three people. Many requests were simple and repetitive, yet still required design skills, so orders and revision rounds piled up. The goal was framed squarely as efficiency: automate the routine requests to free the art team, not to replace anyone.
The solution
Anfibia's path is straightforward to retrace: an internal needs assessment (interviews with each area's leaders about what they needed and imagined doing with AI); a written brief, refined with their programme mentor; hiring an external developer — a staff tech reporter posted a call on Instagram, interviewed three software engineers, and selected one he had worked with before; an iterative build over roughly two to three months, then user testing; and finally two short developer-led training sessions for area leaders.
In use, about five designated users — one per area — sit in a shared Slack channel and submit image requests to a chatbot. The newsroom understands the system as ChatGPT wired into Slack, with n8n (a paid workflow-automation tool) running underneath. Every output travels, together with the requester's brief, to the art director, who must approve it before anything is used. The decisive scoping choice: the prototype handles only the easiest, lowest-risk content — workshop flyers and education- and membership-area communications — and is deliberately not used for imagery attached to articles.
Results
The tool is in active testing and the newsroom calls it plainly useful, estimating five to seven pieces of content a week that would otherwise have gone to the art department. Not everything generated gets used: some outputs contained mistakes or needed changes and were simply discarded. The main thing they would do differently is involve the whole social media team from day one. The project was designed from the art department's perspective, but the daily end users turned out to be the social media team, who only joined mid-project — so the feedback that mattered most arrived late.
Insights
The pilot design is itself a lesson: one representative user per area, a shared channel, mandatory art-director sign-off on every piece, and a start on the highest-volume, lowest-risk content before touching anything editorial. The boundary between promotional artwork and journalism imagery is, in this newsroom, a scar rather than a theory. The year before the prototype, on a high-volume publishing day, an AI-generated Instagram gallery accompanying an article on Patagonia and the climate included an image drawing on a map that labelled the Falklands/Malvinas as British — published without supervision. Argentinian readers reacted furiously; the team describes it as a traumatic episode. As a direct result, Anfibia cut back AI imagery for articles: AI is for creative and promotional artwork, not journalism visuals. The caution is also strategic: Anfibia publishes critically about AI, including a five-episode podcast series, and its followers watch what it does with the technology — careless AI content would cost it more than it would larger outlets.
CONTRADICTION FLAG — carried from extraction: interview notes say the errant image was "of Iceland"; transcript says Falklands/British labelling. Transcript treated as primary; unresolved.EDITORIAL FLAG — attributed use of the incident in Part I judged acceptable in the extraction (newsroom's own account of its own lesson), but explicit sign-off from Anfibia recommended before publication.Cost
The cost structure comprised the external developer's fee for the two-to-three-month build, the n8n licence, two or three ChatGPT seats, and the newsroom's pre-existing design-software subscriptions. User training came free, as a favour from a developer who enjoys teaching.
Replicability potential
The pathway itself is the transferable asset: needs interviews, a written brief refined with a mentor, competitive selection of a developer, an iterative build, developer-led training, and a restricted low-risk pilot with mandatory human sign-off. So is the scoping principle — start with flyers, not journalism — and the lesson to involve the real end users from the beginning. Two things to adapt: Anfibia found its developer — and its free training — through a tech reporter's personal network, so others should budget explicitly for recruitment and training; and Anfibia's multi-branch structure generates unusually heavy flyer demand, so a single-title newsroom should check it has the request volume to justify the build.