Compare use cases

The eight use cases side by side — each column a use case, each row a fixed seat. The seats never move, so the differences pop.

	Use case 1 Drafting & editing writing Communications	Use case 2 Research & background gathering Research & Development	Use case 3 Data & spreadsheets Planning & analysis	Use case 4 Admin & scheduling Scheduling & logistics	Use case 5 Summarizing documents & meetings Staying on top of it	Use case 6 Brainstorming & ideas Arts & Culture	Use case 7 Grants & funder reporting Fundraising	Use case 8 Translation & identification Translation & Language Access
Used at	The BentwayThe organization	The BentwayThe organization	The BentwayThe organization	The BentwayThe organization	The BentwayThe organization	The BentwayThe organization	The BentwayThe organization	The BentwayThe organization
The tools	ChatGPT · Claude · CopilotThe tools staff named	ChatGPT · Gemini · PerplexityThe tools staff named	ChatGPT · GeminiThe tools staff named	ChatGPT · GeminiThe tools staff named	ChatGPTThe tool staff named	ChatGPT · ClaudeThe tools staff named	ChatGPT · ClaudeThe tools staff named	Google Lens · GeminiThe tools staff named
The system	The system drafts, and a person decides what to do next.	The system understands and recommends, and a person decides what to do next.	The system understands and recommends, and a person decides what to do next.	The system drafts, and a person decides what to do next.	The system understands, and a person decides what to do next.	The system drafts, and a person decides what to do next.	The system drafts, and a person decides what to do next.	The system senses and understands, and a person decides what to do next.
Data flow	your rough draft — often naming partners or funders IdentifiableSensitive personal information pasted into a vendor chatbot · no rule yet on what is safe to share↓ rephrase, tighten, adjust toneLanguage Models ↓ a polished draft IdentifiableGenerated content edited by staff before it ships	your questions and rough notes Generated content about materials, artists, sites and precedents↓ a synthesis of sources it does not citeSearch & Retrieval ↓ a starting point — not an answer A recommendation or prediction cross-referenced before staff rely on it	spreadsheets — contact lists, budgets, survey results IdentifiableOperational data whole files uploaded to a vendor chatbot↓ compare, dedupe, reformatLanguage Models ↓ a cleaned or merged sheet IdentifiableOperational data errors have been caught in review	schedules, lists and itinerary details IdentifiableOperational data artist travel and contact details included↓ format, sort, lay outLanguage Models ↓ a tidy schedule or list IdentifiableGenerated content checked before it circulates	meeting notes and long documents — colleagues and partners named IdentifiableSensitive personal information shared with a vendor chatbot to condense↓ condense and synthesizeLanguage Models ↓ a digest staff act on Generated content what it drops, no one reads	a prompt and a half-formed idea Generated content early thinking, before it is shared with anyone↓ variations, angles, themesLanguage Models ↓ raw options to react to Generated content a seed staff and artists reshape	draft applications, plus the funder’s name and priorities IdentifiableSensitive personal information organizational strategy shared with a vendor↓ refine, reframe, tailor to the funderLanguage Models ↓ polished, funder-specific text Generated content staff would likely never tell the funder AI helped	a photo — a plant, a machine part, a phrase About a place taken on site, under the Gardiner↓ match against everything similar onlineComputer Vision ↓ a best guess at a name A recommendation or prediction verified on other sites before staff act
Risks staff flagged	Losing the human voiceThe concern staff named most, on this task above all: communications that stop sounding like the Bentway. Several staff described AI writing as recognisable and increasingly common across the sector. How it's managedStaff practice today: never the final product — staff edit afterwards to put the writing back in their own voice. Privacy & confidentialityDrafts carry partner and funder details into tools no one has vetted. Nothing yet says what is safe to paste. !No safeguard in place yet Environmental costRanked first by several staff — at odds with the Bentway’s sustainability work; some colleagues opt out of AI entirely over it. How it's managedProposed by staff: count AI use in the Bentway’s environmental impact reports, like commuting and guest travel.	Confident but unsourcedAnswers arrive with no source and no context — and are sometimes treated as if they were binding. Staff have seen chatbot pay figures presented as fact. How it's managedStaff practice: cross-reference everything — which can cost as much time as the task would have taken unaided. Outsourcing judgementThe worry that the starting point quietly becomes the answer — a person no longer making the final decision, verifying the information, or applying their own judgement. How it's managedA tool, not a substitute: AI offers assistance, and review of the work stays necessary. What the questions revealQueries about artists, partners and plans flow to vendors, with no guidance on what is safe to ask. !No safeguard in place yet	Errors that look finishedStaff report catching errors in the output — and cases where doing the task by hand would have been quicker. A wrong merge or figure can travel into decks, reports and decisions. How it's managedNothing is used until a person has checked it — staff re-check what the tool returns. Contact lists on vendor serversUploading a contact database sends hundreds of people’s details to a tool no one has vetted — the people listed never agreed to that. !No safeguard in place yet Environmental costHeavy file-crunching is exactly the kind of routine task whose convenience, several staff argue, does not outweigh the environmental cost — the reason some refuse AI outright. How it's managedProposed by staff: count AI use in the Bentway’s environmental impact reports.	Artists’ details in the promptAn itinerary is personal data — where someone sleeps and when they travel — pasted into a consumer tool with no rules about it. !No safeguard in place yet Small tasks, real footprintStaff who opt out point at exactly this: simple tasks whose convenience does not outweigh the environmental cost. The case where convenience and the Bentway’s values rub hardest. How it's managedAsk first whether the task needs AI at all — several staff simply don’t use it.	The summary becomes the recordIf the digest is wrong, the error propagates — the source was too long to check, which is why AI was used in the first place. How it's managedSpot-check against the original: the summary is a pointer, not the record. Whose work gets read?Summarizing colleagues’ documents changes whose words are actually read — part of the wider staff worry about effects on jobs and roles. !No safeguard in place yet	Ideas that all sound alikeA model trained on everyone’s ideas pulls toward the average — a real tension for an organization whose value is human connection and original public art. How it's managedIdeas are seeds, not selections: staff and artists make the creative calls. AI in creative work — where do we stand?Artists already send AI renderings, and AI imagery has already drawn public comment on a Bentway exhibition — the Bentway hasn’t yet said where it stands. !No safeguard in place yet	If funders found outStaff who use AI most on this work say they would not tell an external partner or funder. Undisclosed use is a standing bet on how funders would react. !No safeguard in place yet An error in an applicationA hallucinated metric or misstated program in a grant application puts real funding at risk. How it's managedDrafts start human: staff write the communication first and paste it in only to refine — and every claim is reviewed. A funder-shaped voiceTailoring language to each funder is the job — but the human voice is what staff most fear losing. How it's managedThe application always starts with the writer, not the model.	Wrong name, wrong fixThe site runs on old and one-of-a-kind machinery, so auto-generated help is often wrong — and a misidentified part or plant can mean the wrong intervention. How it's managedStaff practice: verify against other sources before acting on the answer.
Who it touches	Partners, funders & the public who read Bentway communications Wider community every email, caption and report AI touchedaffected together, beyond any one person it processes.	Staff who build on what it returns Wider community 8 of 18 staff use AI this wayaffected together, beyond any one person it processes.	People in the spreadsheets Caught incidentally contacts, artists and respondents — swept in with every uploadswept into the data; no decision is made about them.	Artists & guests being scheduled Caught incidentally their travel, timing and contact detailsswept into the data; no decision is made about them.	Colleagues & partners whose words are condensed Caught incidentally named in the notes and reports that go inswept into the data; no decision is made about them.	Staff & artists shaping programs Wider community ideas that end up in public spaceaffected together, beyond any one person it processes.	Funders reading applications & reports Wider community the grants that keep programs runningaffected together, beyond any one person it processes.	The grounds & site team Wider community novel plants and one-of-a-kind machinesaffected together, beyond any one person it processes.
You can	A person signs off Ask how it was made Questions or concerns — bring them to the Sept 2 workshop Email the AI policy group	Ask how it knows Verification is the job Questions or concerns — bring them to the Sept 2 workshop Email the AI policy group	Fix a wrong record Know your data was processed Questions or concerns — bring them to the Sept 2 workshop Email the AI policy group	Know AI touched your itinerary Correct a detail Questions or concerns — bring them to the Sept 2 workshop Email the AI policy group	Ask for the original A person owns the takeaways Questions or concerns — bring them to the Sept 2 workshop Email the AI policy group	Opting out is fine Ask where an idea came from Questions or concerns — bring them to the Sept 2 workshop Email the AI policy group	Disclosure is an open question A person owns every claim Questions or concerns — bring them to the Sept 2 workshop Email the AI policy group	The person on site decides Questions or concerns — bring them to the Sept 2 workshop Email the AI policy group