AI Resources
Two ways to get more out of AI: agents that walk you through specific tasks, and a playbook of named skills you can apply to anything.
Agents
Agents are AI assistants pre-configured for a specific task. Instead of writing a detailed prompt, you open the agent and start a conversation — it already knows what it’s for.
These agents are experimental. The Ignition team isn’t sure yet whether you’ll find them useful. Please try one and let us know in Slack what you think.
Compass
New to AI and not sure what to try? Compass looks at your actual work and finds something AI can genuinely help with — then does it with you.
When to use
Reach for Compass when you're curious about AI but don't have a specific task in mind, or when you're looking at your day-to-day work and wondering whether AI could help at all. Tell it what your organization does and what's eating up your time, and it'll find something concrete to try right then.
Example prompt to try
I spend hours each week scheduling, writing reminder emails, and pulling the same information together for reports. I'm not sure where AI fits in — can you help me figure that out?
Sage
Already working on something? Sage does the task with you and adds a quick tip at the end to sharpen your AI skills as you go.
When to use
Use Sage when you already have a task in hand — a document to draft, a report to tighten, a problem to think through — and you want it done well while picking up better AI habits along the way. It does the work first, then adds a short, skippable tip.
Example prompt to try
Here's a rough draft of a report section [paste it]. Can you tighten it and make the tone more confident for an external audience?
Skills
Foundation Skills
Your first week on the platform. Getting oriented, building intuition, finding what works.
The Taste Test — Find out which model fits which task
The situation
You don’t know which model to use for a task.
The skill
Take one real prompt from your work. Run it on three models: one fast and cheap, one mid-tier, one powerful. Read all three results side by side. Notice what’s different — not just quality, but character.
The example
“We serve 400 people a month through our program. Write a 200-word description of our impact for a stakeholder report.” Run it on three models. One will nail the tone. One will have better structure. One will include things you didn’t think of.
Watch for
You’re looking for the character of each model, not just quality. Fast models are confident but shallow. Powerful models are nuanced but sometimes overthink. This isn’t about finding the “best” model — it’s about learning when to reach for each one.
The upgrade
Run this same exercise every month. Models change. Your calibration should too.
Practiced in: Explore Module 1
The Briefing — Get AI to interview you before it starts working
The situation
You have a task for AI but you know the output will be generic if you just describe it cold.
The skill
Instead of giving AI the task, tell it what you need and ask: “Before you start, ask me 10 questions that would help you do this well.” Answer the questions honestly and in detail. Then let it work. The output will be dramatically better because AI now has the context it needed but you wouldn’t have thought to provide.
The example
“I need to write a board update for our program’s first quarter. Before you draft anything, ask me the questions you’d need answered to write a really good one.”
AI asks about audience, tone, key metrics, notable challenges, what the board cares about most, what happened last quarter, etc. You answer. Then the draft is 10x better than “write me a board update.”
Watch for
If the questions are too generic, tell it: “You can do better — ask me the questions that would make this specifically useful for MY situation.” AI will sharpen its questions.
The upgrade
Combine with The Sparring Partner. After AI drafts something using your answers, ask it: “Now challenge this. What’s weak? What am I not seeing?” You’ll get the draft AND the critique.
Practiced in: Explore Module 2
The 70% Draft — Get past the blank page
The situation
You need to write something and you’re staring at a blank page.
The skill
Tell AI who you are, what you’re writing, who it’s for, and what tone you want. Ask for a full draft. Don’t expect it to be done — expect it to be 70% of the way there. Your job is the last 30%.
The example
“I’m the director of a program. I need to write a 500-word update for our quarterly board report. The tone should be confident but honest — we had a great quarter on engagement but enrollment is down. Here are the key numbers: [paste data]. Draft it.”
Watch for
If the draft feels generic, you didn’t give enough context about your organization. The more specific your input, the less generic the output.
The upgrade
After you edit the draft, paste your edited version back and say: “Here’s what I changed and why. Next time I ask you to write something like this, apply these preferences.” You’re teaching the model your voice.
The Translator — Make complex things understandable for any audience
The situation
You need to take something complex and make it understandable for a different audience.
The skill
Paste the complex thing — a policy document, a technical report, a dense email from a stakeholder — and say: “Translate this for [audience]. Keep the meaning but make it accessible to someone who [describe their context].”
The example
Take your organization’s data privacy policy and say: “Rewrite this for the people we serve. They need to understand what we collect and why, but they shouldn’t need a law degree. Keep it under 300 words. Warm tone.”
Watch for
AI is genuinely excellent at this. If the result feels dumbed-down rather than translated, ask for a revision: “This feels condescending. Keep the clarity but treat the reader as intelligent.”
The upgrade
Chain it: ask AI to translate the same document for three different audiences. You’ll learn how much the framing shifts for each — and you’ll start doing this instinctively.
Practice Skills
You’ve got the basics. Now use AI with intention — for thinking, for iteration, for solving real problems.
The Sparring Partner — Think through a decision with AI pushing back
The situation
You need to think through a decision, a strategy, or a plan — not get a deliverable.
The skill
Don’t ask AI to write anything. Ask it to challenge you. “I’m considering [decision]. Here’s my reasoning: [explain]. Push back. What am I missing? What are the risks I’m not seeing? What would a skeptic say?”
The example
“We’re thinking about launching a new program focused on AI skills for adults in our community. Our budget is $50K and we’d partner with another local organization. Here’s our logic: [explain]. Play devil’s advocate. What could go wrong? What assumptions are we making?”
Watch for
This is one of AI’s most underrated capabilities. It’s a thought partner that doesn’t get tired, doesn’t get political, and will push back without ego. The value isn’t in the answers — it’s in the questions it surfaces that you weren’t asking.
The upgrade
After the sparring session, ask AI to summarize the strongest counterarguments and your responses. You now have a decision memo you can share with your team or board.
The Chain — Break complex tasks into steps AI can actually handle
The situation
You have a complex task that AI keeps getting wrong when you ask for it all at once.
The skill
Break it into steps. Each step is its own prompt. The output of step 1 becomes the input for step 2. Research → Analysis → Draft → Revision. Don’t ask for it all at once.
The example
Instead of “Write a proposal for our program,” try: Step 1: “Here’s our program description. Summarize our theory of change in 3 sentences.” Step 2: “Now, given that theory of change, here are the audience’s priorities [paste]. Which of our strengths align best?” Step 3: “Draft the needs statement section using those alignment points. 400 words, data-supported.” Step 4: “Now review what you wrote. Is anything unsubstantiated? Flag it.”
Watch for
Each step should produce something you can evaluate before moving to the next. That’s the whole point — you’re staying in the loop. If any step goes sideways, you catch it immediately instead of discovering a flawed final product.
The upgrade
Save your best chains. They become templates. They become agents. A chain you’ve refined three times is an agent waiting to be born.
The Mirror — Use AI to learn why AI gave you a bad result
The situation
AI gave you a bad result and you don’t know why.
The skill
Paste the bad result back into a new conversation and say: “I asked AI to do [task]. Here’s what it produced. It’s not what I wanted because [explain what’s wrong]. How should I have asked differently to get a better result?”
The example
“I asked AI to write a thank-you letter for a supporter and it produced something that sounds like a robot wrote it — formal, generic, no warmth. Here’s what it wrote: [paste]. How should I prompt differently to get something that sounds like it came from our director, who is warm and personal and always mentions specific things the supporter made possible?”
Watch for
This is the meta-move — using AI to learn how to use AI. The answer will usually reveal that you needed to give more context, a specific example of the tone you wanted, or explicit constraints.
The upgrade
Build a “style guide” prompt from the feedback you get. Save it. Use it every time you need that voice. You’re closing the Inputs Gap permanently for that task.
The Expert Panel — Get multiple perspectives on a single decision
The situation
You’re making a decision that benefits from multiple perspectives and you don’t have time to convene a focus group.
The skill
Ask AI to respond to your question from three different expert viewpoints. “Answer this question from the perspective of (1) a skeptical program evaluator, (2) a community member who would use this service, and (3) a decision-maker who has seen 100 pitches like this.”
The example
“We’re redesigning our intake process. Here’s the current process: [describe]. Evaluate it from three perspectives: (1) a staff member who uses this process daily, (2) someone going through the process for the first time, (3) a stakeholder evaluating our efficiency.”
Watch for
The perspectives won’t be perfect — AI doesn’t actually know what your community is experiencing. But it will surface angles you might miss, and the framing forces structured thinking. Use this to generate questions, not to replace actual community input.
The upgrade
Add a fourth perspective: “Now synthesize. Where do these three perspectives agree? Where do they conflict? What does that tell me about the tradeoffs I’m making?”
The Replay — Ask AI to coach you on your own prompting
The situation
You’ve just finished a long conversation with AI and you’re not sure if you could have gotten there faster or better.
The skill
At the end of the conversation, paste this: “Review this entire conversation. Evaluate my prompting — what did I do well, where was I vague or inefficient, and what could I have done differently to get better results faster? Be specific and direct.” Then read what it says and try one suggestion in your next conversation.
The example
You just spent 30 minutes going back and forth getting AI to draft a proposal. The final version is good, but it took 12 messages to get there. You ask the model to review the conversation. It tells you: “You gave good context about the audience in message 6 — but if you’d led with that, messages 2–5 wouldn’t have been needed. Your revision requests were vague (‘make it better’) until message 9 when you said ‘tighten the budget justification and add a theory of change paragraph.’ That specificity is what got you the final draft. Lead with that level of detail next time.”
Watch for
Use one of the more powerful models for the review — Claude Sonnet, Claude Opus, or GPT-4. These models give the most detailed, nuanced feedback on your prompting. And be ready — the feedback can be surprisingly blunt about habits you didn’t know you had.
The upgrade
Do this once a week for a month. You’ll watch your prompting get measurably better — fewer back-and-forth cycles, better first drafts, faster results. Then do it for someone else: review a colleague’s conversation (with their permission) and coach them. That’s how your whole team levels up.
Advanced Skills
Turn individual wins into team capability. Extract expertise. Build things that outlast the program.
The Handoff — Turn what works for you into something anyone can use
The situation
You’ve found something that works and you want AI to write the prompt for you so anyone can replicate it.
The skill
Describe what you do and the result you get: “I regularly need AI to do [task]. Here’s an example of a great result I got: [paste]. Here’s the context that matters: [explain]. Write me a reusable prompt template that I or anyone on my team can use to get this result consistently.”
The example
“I use AI to draft weekly parent newsletters for our school program. Here’s one I was happy with: [paste]. The tone is warm but informational, we always include 3 sections (what happened this week, what’s coming, one thing to try at home), and we always mention specific kids’ achievements without using names. Write me a prompt template I can hand to any staff member.”
Watch for
The output should be a reusable template with clear blanks to fill in — not a one-time result. If it’s too specific to one instance, ask AI to generalize it. If it’s too vague, give it another example of a good output to calibrate against.
The upgrade
This IS an agent’s system prompt. If you’re on the Build path, take this template and turn it into an actual agent on the platform. You’ve just gone from personal trick to team infrastructure.
The Audit — Have AI catch what you’ve become blind to
The situation
You have an important document and you need to check it for problems you might have missed.
The skill
Paste the document and ask AI to review it against specific criteria. Don’t say “review this.” Say: “Review this [document type] for [specific things]. Flag anything that is [unclear / unsupported / inconsistent / missing]. Format your review as a numbered list with quotes from the document.”
The example
“Review this proposal for our program. Check for: (1) claims without supporting data, (2) jargon a non-expert reader wouldn’t understand, (3) anything that promises outcomes we can’t actually measure, (4) gaps where the audience’s priorities [paste] aren’t addressed. Quote the specific lines so I can find them.”
Watch for
This is a classic Centaur move — clean division of labor. AI reviews, you decide what to fix. AI is often better at catching things you’ve become blind to from working too closely with a document. But it can also flag things that are actually fine — you’re the judge.
The upgrade
Build a standard review prompt for every major document type your org produces. Proposals, board reports, program evaluations, stakeholder communications. Each gets its own checklist. Save them as templates or agents.
The Interrogation — Extract your own expertise so AI can use it
The situation
You need to make your own expertise explicit — the stuff that’s in your head but has never been written down — so AI can work with it.
The skill
Have AI interview you. “I want to create a guide for how I [do specific task]. Interview me. Ask me one question at a time about my process, my decision-making, what I look for, what good looks like, and what common mistakes are. After 10 questions, synthesize what you’ve learned into a process document.”
The example
“I’ve been reviewing applications for 15 years. I want to document my review process so AI can help newer staff do preliminary reviews. Interview me. Start with: what’s the first thing I look at when I open an application?”
Watch for
This is possibly the highest-leverage Move in the entire collection. You’re extracting tacit knowledge — the stuff that makes your organization’s best people great at what they do. The output is usually surprising: people discover they have a more structured process than they realized, or that their instincts follow patterns they’ve never articulated.
The upgrade
Use the synthesized process document as a system prompt for an agent. You’ve just bottled your expertise. When that senior staff member retires or moves on, their judgment doesn’t disappear — it lives on the platform.
Invent your own
The best Moves come from practitioners, not program designers. When you discover a technique that reliably works for your organization — something you’d teach a colleague — name it and share it in Slack with the tag #new-skill. Include: the situation, what you did, and what happened. The best ones get added to this page. This collection belongs to the cohort.
When it’s not working
AI will frustrate you. That’s not failure — it’s information. When something doesn’t work, figure out which part broke:
Is it the brain (the model)?
Some models are better at certain tasks. If one model gave you a bad result, try another. Claude is often better at nuanced writing. GPT-4 handles structured data well. Fast models are good for simple tasks but struggle with complexity.
Try: Run the same prompt on 3 models and compare.
Is it the body (the platform)?
Sometimes AI could do the task, but the tool doesn’t support it well — file handling, formatting, long documents, memory across conversations. This is a real and common limitation right now. Most current AI limitations are platform limitations, not model limitations.
Try: Note it, share it in Slack, and know it’s improving constantly.
Is it the technique (your approach)?
This is the most common one — and the most fixable. Try giving more context. Break a big task into smaller steps. Ask AI to critique its own output. Or ask: “I tried to get you to do X and the result wasn’t great. How should I approach this differently?”
Try: Use AI to learn how to use AI. Seriously.
Post your “it didn’t work” moments in Slack. That’s where the best learning happens — for you and everyone else.