--- id: prj-teach-your-assistant-your-jargon schema: project/1 layer: garden title: Teach your AI assistant your team's jargon, one confirmed term at a time outcome: A short glossary of your team's internal terms that the AI found in your own notes, guessed at, and saved only after you confirmed or corrected each one, so later answers stop misreading your shorthand. roles: [team lead, project manager, operations staff, programme coordinator, new joiner] time_box: 25 minutes tool_access: Any AI chat assistant. A projects feature or saved instructions makes step 4 easier (for example Claude Projects, ChatGPT Projects or Gemini Gems). Use only notes you are allowed to share with that tool, or the invented sample below. practises: [cmp-describe-the-task, cmp-check-the-result, cmp-make-it-repeatable, cmp-use-responsibly] concepts: [con-memory, con-context-engineering, con-projects-and-knowledge, con-hallucination, con-human-in-the-loop, con-clarifying-questions] skills: [build-brief, prove-it, make-it-repeatable] resources: [res-other-claire-vo-self-improving-pm-system, res-other-claire-vo-how-i-ai-workflows] source: url: https://www.chatprd.ai/how-i-ai/workflows/build-a-self-improving-ai-morning-brief-to-capture-action-items-and-learn-company-jargon claim: An assistant gets more useful when it notices internal terms it doesn't understand, offers its best guess, and saves a definition only after the person confirms it. evidence: >- In the How I AI episode "Claude Cowork for PMs: My Self-Improving Productivity System" (27 August 2026, hosted by Claire Vo), product manager Daniel Blum showed a scheduled Claude Cowork morning brief that scans his meeting notes and messages for unfamiliar terms and asks him before saving each definition. ChatPRD's workflow write-up lists those steps. limits: This is one person's working setup, shown on screen, not a measured result. His version ran on a schedule with Cowork, Granola and connected Slack and email; this card does the same loop by hand in an ordinary chat. Nobody has measured how often the AI's guesses are wrong, which is why every definition needs your yes. Prompts here are written for this card. origin: kind: ai-proposed by: "seed agent 2026-10-16" rationale: Takes the smallest reproducible part of a How I AI workflow (spot, guess, ask, save) and turns it into a manual exercise anyone can run before automating it. --- ## Why this matters Every team has shorthand: project nicknames, report acronyms, the name of the van. When you paste your notes into an AI, it either skips those words or guesses without saying so, and a wrong guess can end up in a summary someone else reads. Writing a glossary from scratch is tedious and you'll forget the terms you use most. In this project the AI finds the terms, you supply the meaning, and nothing gets saved without your say-so. ## Smallest useful build A glossary of five to ten terms, each with a one-line meaning, the note where it appeared, and a mark showing you confirmed it. Plus one test showing the AI now reads a new note correctly. One chat, about 25 minutes. ## Sample inputs Invented notes from a made-up food bank: ```text Mon standup: Falcon route short two drivers again. Priya to ask the Tuesday crew. FYI the PSR is due to the county on the 15th - use the new template. Tue: Pantry Plus pilot paused until the cold-room fix. Ask Marco about CR2. Wed: Reminder that green-tag items go out first. Don't mix with the Harvest bins. Thu: Board wants the Q3 PSR numbers before the retreat. Falcon also needs a backup van. Fri: New volunteers start Monday - send them the 3-2-1 sheet. ``` The intended meanings, for checking at the end only: Falcon is the Thursday delivery route to the east side; PSR is the Pantry Service Report sent to the county; Pantry Plus is a pilot that adds fresh produce; CR2 is the second walk-in cold room; green-tag items are near their use-by date; Harvest bins hold donations from a local farm; the 3-2-1 sheet is a one-page volunteer briefing. ## Steps 1. **List the terms yourself first.** Before using AI, read the notes and write down every word a new volunteer wouldn't understand. Don't write meanings yet. 2. **Ask the AI to find and guess, not decide.** Paste the notes and send: ```text These are my team's notes. Find every internal term, acronym, nickname or code a newcomer wouldn't understand. For each one, give your best guess at what it means, say how confident you are (high, medium or low) and quote the line where it appears. Don't treat any guess as fact. Then ask me to confirm or correct them, one at a time, starting with the one you're least sure of. ``` 3. **Confirm or correct each one.** Answer each question with "yes", a correction, or "skip". When you're done, send: ```text Write a glossary with only the terms I confirmed or corrected. For each: the term, a one-line meaning in my words, the line it came from, and "confirmed". Leave out anything I skipped. Put any term you're still unsure about in a separate list called "Still unclear". ``` 4. **Save it and test it.** Save the glossary in the project knowledge or saved instructions, or keep it as a file to paste at the start of future chats. Then start a new chat, add the glossary, and paste this new note: ```text Falcon can't run Thursday. Move green-tag stock to CR2 and flag it in the PSR. ``` Ask: `What does this note ask me to do, in plain words?` ## Check it - Every term in the glossary is one you confirmed or corrected; none is a guess the AI saved by itself. - Compare the AI's list from step 2 with yours from step 1. Note any term it missed and any you missed. - At least one guess in step 2 was wrong or marked low confidence. If every guess was "high" and correct, check whether it was guessing or just lucky. - In step 4 the plain-words answer uses your meanings, such as the Thursday east-side route rather than an invented one. ## Break it Add this line to the notes and repeat step 2 with the saved glossary included: `Sat: Falcon is now the name for the new volunteer app, not the route.` A term in your glossary now means something new. Does the AI flag the clash and ask which meaning to keep, or does it use the old definition without mentioning the change? Write down what happened. ## Reflect - Which terms did you assume everyone knew? - Did any of the AI's guesses sound so right that you nearly said yes without thinking? - Who else on your team should check this glossary before anyone else relies on it? ## Next iteration Run steps 2 and 3 once a week on that week's notes, adding only newly confirmed terms. If it keeps finding useful terms, turn it into a scheduled task or saved routine that asks you rather than saves for you. Before connecting it to email or chat tools, decide what it may read, keep personal details out of the glossary, and choose who owns updates when a term changes.