---
name: Monica AI Assistant Workflow
description: Get real work out of Monica — the all-in-one AI assistant that sits across your browser and apps — with a repeatable workflow for research, drafting, translation, and summarization that keeps a human editing the output instead of pasting it. For anyone using a general AI assistant seriously.
audience: designer · writer · everyone
---

# Monica AI Assistant Workflow

## What this is

A working method for Monica, the all-in-one AI assistant (browser extension + apps) that offers chat, summarize, translate, and write across multiple underlying models. General assistants make it trivial to generate; the value is in *how* you use one so the output is trustworthy. This skill is the discipline: prompt with context, verify before you ship, and keep the human as the editor.

## What this is NOT

Not affiliated with Monica and not tied to a specific feature set — the exact tools and models change; the workflow doesn't. Not a claim that a general assistant replaces domain expertise or primary sources. It is a way to use one that amplifies judgment instead of outsourcing it.

## Method

1. **Bring the context, don't assume it.** Paste the actual text, page, or brief. A general assistant is only as good as what you give it — "summarize this" over a page it can see beats "tell me about X" from memory.
2. **Ask for structure, then edit it.** Request the output in the shape you need (bullets, a table, a rewrite in a set tone). Then edit it — the first draft is a starting point, not a finished artefact. Pasting raw output is how errors ship.
3. **Verify anything factual.** Claims, numbers, quotes, and citations get checked against a primary source before they leave your hands. A confident assistant is not a correct one; an invented citation is worse than none.
4. **Translate for meaning, review for nuance.** For translation, give the register and audience, then have a fluent reader check tone — a technically correct translation can still be wrong for the room.
5. **Summarize without losing the load-bearing detail.** Ask what was cut, not just for the short version — the sentence that matters (a deadline, a caveat, a number) is the one summaries drop.
6. **Keep sensitive data out.** Don't paste secrets, personal data, or anything under NDA into a third-party assistant. If the content is sensitive, it doesn't go in the box — the same rule the CDS privacy patterns enforce.
7. **Switch models deliberately.** If the assistant lets you pick the underlying model, use cheap-fast ones for drafting and translation, stronger ones where judgment matters — the same routing logic as any multi-model workflow.

## Quality bar

Real context is provided, not assumed · output is edited before it ships · every factual claim is verified against a source · translations are reviewed for nuance · summaries are checked for dropped load-bearing detail · no sensitive or personal data is pasted in · model choice matches the task's stakes.

## Guardrails & escalation

Regulated advice (medical, legal, financial) from a general assistant is a starting question for a professional, never an answer — route it to one. Content destined for publication passes a human editorial and fact-check pass. If the assistant's output would be attributed to you or your company, you own it: review it as if you wrote it, because as far as the reader is concerned, you did.

## References

- Catalogue: https://edwson.com/consumer-design-system.html · Contracts: https://edwson.com/cds/components.json · Agent brief: https://edwson.com/cds/AGENTS.md
- Related: the portfolio's "On using AI, honestly" method note. Confirm Monica's current feature set and data handling against its own documentation.