Our AI philosophy
Here's the short version: AI is a tool that makes good Optimists better. It doesn't replace your expertise, your creativity, or your relationships with clients. What it does is handle the repetitive 10 to 20% so you can spend more energy on the valuable 80 to 90%.
We are not an "AI-first" agency. We are a people-first agency that uses AI well. The work we do for clients depends on human judgment, creative intuition, and genuine care about results. AI helps us get there faster. It doesn't get us there on its own.
The golden rule
Never hand off to AI something you don't understand well enough to check.
AI is an assistant. You are the expert. If Claude gives you a report on a client's analytics performance and you can't tell whether the numbers make sense, that's a problem. Run the numbers yourself first, or ask a colleague, before handing any data work to AI.
The four-tier usage policy
This policy comes from leadership and applies to everyone at OO. When you're not sure whether to use AI for something, run it through this framework.
Must use AI
For these tasks, you should be using AI. If you're not, you're working harder than you need to.
- Meeting recordings and transcription
- Problem-solving before escalating to another Optimist
- Coding, formulas, and regex
- Menial, repetitive tasks
- Summarizing long documents
Should consider AI
AI is a good fit here. Use your judgment on how much to rely on it versus doing it yourself.
- Brainstorming and ideation
- Bulk content creation for lower-priority items
- Low-priority copywriting
- Validating content against client brand standards via Claude Projects
Use cautiously
AI can help here, but the risk of errors is real. Always verify the output before using it.
- Data analysis and reporting (hallucination risk)
- Creative elements, unless you're on the Design team
Do not use AI
These are off-limits. Full stop.
- High-priority or sensitive client communications
- Any work for a client who has previously flagged AI-generated content
Not sure which tier something falls into? Ask yourself: "If this output were wrong, how bad would that be?" The worse the answer, the more cautious you should be. And always ask your Director or a VP if you're genuinely uncertain.
Data privacy: what can go into an LLM?
This is one of the most common questions we get, and it matters. When you paste something into Claude or ChatGPT, where does it go? Here's the practical breakdown.
What NOT to put into free-tier AI tools
- Personally identifiable information (PII): names, addresses, phone numbers, emails tied to real people
- Passwords or login credentials of any kind
- Financial data (bank account numbers, payment info, revenue figures with client names attached)
- Confidential client information (unreleased campaigns, proprietary processes, internal strategies)
- Anything marked confidential in a client contract
The agency's Claude Pro accounts are okay for client work, because Anthropic's professional agreements limit how that data is used for training. Free-tier tools have different terms. When in doubt, ask your Director or a VP before pasting anything sensitive.
⚠️ Quick test: Before you paste something into a free-tier tool, ask yourself: "Would I be comfortable if this appeared in a training dataset?" If the answer is no, don't paste it.
AI note-taker etiquette
AI note-takers are genuinely useful. They also require some basic courtesy rules, because a lot of people don't love being recorded without knowing.
- Always ask before using an AI note-taker, whether it's an internal meeting or a client call. "I'm planning to use AI to take notes — is everyone okay with that?" takes 10 seconds.
- If a client shows up with their own AI note-taker and didn't ask first, it's okay to politely say: "We're happy to have that running, and going forward we'd appreciate a heads-up before the call. Just so everyone knows what's being recorded."
- This applies to internal meetings too. New Optimists and guests especially may not know they're being recorded.
- If someone asks you to turn it off, turn it off. No exceptions.
- Online Optimism's Google Calendar automatically adds an AI note-taker to calendar events with 3+ attendees. Event organizers may toggle this off ahead of a meeting if they deem it unnecessary.
- To manually turn a note-taker on in a Google Meet call, ask everyone inside the meeting first, then toggle on in the upper right corner.
- Google will alert anyone entering a meeting about a note-taker, so they can give consent by entering the meeting.
The ethical and environmental stuff
Optimists have brought up real concerns about AI: the environmental impact of running large models, and the broader ethical questions around data, labor, and what AI does to creative fields. We're not going to paper over those.
OO's value "Keep It Human" is directly relevant here. We believe in the people who work here, the people we work for, and the communities around us. Part of using AI responsibly means acknowledging its costs, being thoughtful about when it's the right tool, and not treating it as a replacement for expertise or human connection.
The environmental cost of AI is real. Training large models and running billions of inference calls uses significant energy. We don't have a perfect answer to this, but we do believe that using AI intentionally (rather than reflexively) is better than using it for everything. Don't run a giant model to do something you could do in two minutes yourself.
We also believe that by learning how to use AI effectively (by completing this course), we can also use it more efficiently, and reduce irrelevant usage as much as possible.
If you have concerns about specific AI uses at OO, and any ideas on how to solve these, please bring them up to your Director, a member of the Top Trio, or Flynn. These conversations are worth having.
What to do when Claude is down 😅
It happens. Claude goes glitchy, the API times out, or the whole service is down. Here's your troubleshooting flow:
- Refresh the page. A lot of Claude weirdness resolves itself.
- Check status.anthropic.com. If there's an active incident, you'll see it there.
- Try ChatGPT as a backup for non-brand-voice work (formulas, summaries, general research). Don't use it for client-facing copy because it doesn't have the Communication Codexes.
- If the task requires brand voice or a client Claude Project, wait for Claude to come back. It usually doesn't take long.
- If everything's truly broken and you have a deadline: do the task manually. You know how. You were doing it before AI existed.
🛠️ Post in #ai-and-automations on Slack if Claude is down and you're blocked. Someone on the team usually knows what's going on and can help you find a workaround.
Try it yourself
Open the Using (and not Using) AI Notion page and read through the full policy. Then, using the four-tier framework above, write down:
- One specific task from your last week of work that probably belongs in the "Must use" tier — name the task, and name the tool you'll use for it next time.
- One specific thing you've used AI for recently that might actually belong in the "Use cautiously" or "Do not use" tier for one of your clients — name the client and the task.
No wrong answers. This is just for you to think through.
We suggest you bookmark this Notion page in Chrome for future reference, if you find it helpful!
Common mistakes to avoid
- Treating AI output as final. Claude's first draft is a starting point. Read it. Edit it. Own it.
- Putting client PII into free-tier tools. Always check which Claude account you're logged into before pasting anything sensitive.
- Using AI as a knowledge source for facts you can't verify. Claude will confidently make things up. Use it for drafts and structure, not as a reference for claims.
- Skipping the note-taker permission ask. It's a small thing that makes a big difference to the people in the room.