Cowork Mode: The Complete Guide to Context Vault, Automations, and Production Work

Most people open an AI assistant, type a prompt, copy the answer, and paste it somewhere else. That workflow is fine for quick questions. It is not how production work gets done.

Axionomy Personal Assistant offers two modes inside the same workspace. Chat answers in the thread. Cowork reads your context vault on Google Drive, plans multi-step work, asks clarifying questions when the brief is unclear, and saves deliverables to folders you control.

This guide walks through Cowork end to end: how it differs from chat, how to set up your vault, global instructions, clarifying forms, automations, skills, token discipline, optional visualization presets, and the limitations you should plan for.

Chat vs Cowork: the mindset shift

Chat mode is conversational. You ask, the model responds, you decide what to do with the text. There is no persistent folder contract, no standing orders, and no obligation to produce a file in your Drive.

Cowork mode assumes you have work that should land somewhere specific—a briefing doc, a comparison table, a content draft, a research brief. The agent is expected to read your ABOUT ME context, execute across connected tools when you approve, and write outputs under your vault's OUTPUTS path.

If your goal is "explain this concept to me," use Chat. If your goal is "produce this artifact so my team can use it Monday," use Cowork.

Cowork also surfaces dedicated UI: vault setup, global instructions, token coaching, optional chart presets, and an Automations tab for recurring schedules. Those controls exist because Cowork sessions are longer, cost more tokens, and touch real files.

What you need before you start

Cowork runs inside Autonomous Agent → Personal Assistant on axionomy.xyz. You need an account on a paid plan that includes agent automation and sufficient monthly AI tokens.

Connect Google in Settings → Integrations so Cowork can use your context vault on Google Drive. Optional but valuable: Gmail, Google Calendar, Notion, LinkedIn, Slack, and messaging channels (Telegram, WhatsApp, email) for automations that pull live context.

Select a Drive folder as your Cowork vault root. Cowork reads and writes inside that folder only—you choose the boundary.

Step 1: Build your context vault

The vault is a folder structure on your Drive, not a hidden database. A practical layout has three subfolders: ABOUT ME, OUTPUTS, and TEMPLATES.

ABOUT ME holds standing context the agent should load at the start of substantive tasks. OUTPUTS is where deliverables go—often under project-named subfolders. TEMPLATES stores reusable structures (report skeletons, post formats) stripped of one-off content.

Most generic AI output happens because the model does not know who you are. The vault fixes that once, not on every prompt.

File 1 — about-me.md: who you are, how you work, what "done" looks like, and what you refuse to accept in output. Keep it under roughly 2,000 tokens. Larger files load every session and burn budget before real work begins.

File 2 — anti-ai-writing-style.md (or equivalent): banned phrases, sentence patterns, and formatting rules. Without this, Cowork writes like a model. With it, Cowork writes closer to your voice.

File 3 — my-company.md: north-star goals, active priorities, and what you are saying no to this quarter. Update quarterly, not daily.

Axionomy includes a Cowork interview wizard that can help you generate these files through guided questions instead of staring at a blank page.

The vault panel in Personal Assistant shows which ABOUT ME files exist and estimates token load, with warnings when individual files grow too large.

Step 2: Global instructions

Your folder alone is not enough. Global instructions tell Cowork how to use the vault on every session.

In Personal Assistant, open Global Instructions and paste standing orders—for example: read all ABOUT ME files before substantive tasks; never read OUTPUTS or TEMPLATES unless pointed to a specific file; save deliverables under OUTPUTS/[project]; use clarifying questions when the brief is unclear; do not pad with filler.

You set this once. It applies before each Cowork run, which is why a ten-word task prompt can still produce on-brand output—the context is already loaded.

Step 3: Clarifying forms (AskUserQuestion)

When a brief is ambiguous, Cowork can stop and ask structured questions instead of guessing. You get clickable options, multi-select choices, and ranked lists—not another wall of text to edit.

This is the difference between polishing the wrong answer and approving the right plan before execution.

For non-trivial work, add a line like: "Use AskUserQuestion before you start." Cowork interviews you in under a minute, then builds a plan you review.

Human-in-the-loop still applies: sensitive actions (sending email, publishing posts, creating calendar events) require your approval unless you explicitly pre-approve tools on an automation.

The one prompt pattern that scales

Stop writing novel-length prompts. A durable pattern:

"I want to [TASK] so that [SUCCESS CRITERIA]. Read my ABOUT ME folder first. Then use AskUserQuestion to clarify before you start."

Cowork loads vault context, asks what it needs, proposes steps, and—after you approve—creates files in OUTPUTS. The interaction feels like briefing a colleague, not wrestling with a text box.

Step 4: Templates folder

When Cowork produces something you want to repeat, ask it to save the structure—not the content—to TEMPLATES. Next time: "Use the template in TEMPLATES/[filename]."

Over months, TEMPLATES becomes a library of formats that worked: executive one-pagers, research briefs, client update emails. You do not need to maintain it manually; you point to files when needed.

Step 5: Skills library

Skills are reusable workflows—step-by-step instructions for a specialist job: weekly content draft, competitive comparison, morning briefing, and more.

In Axionomy, skills live in the Personal Assistant skills library and agents-workspace pipelines. A skill encodes process, quality checks, and tool usage so Cowork behaves like a specialist rather than a generalist.

Good skills are specific: not "analyze the data," but "compare this week to last week, flag >20% variance, write one-sentence insight per metric."

Skills pair with slash commands where configured, so recurring workflows start with a short trigger instead of re-pasting instructions.

Step 6: Automations

Automations are scheduled Cowork runs—weekday morning briefings, end-of-day wrap-ups, or recurring production blocks.

In Personal Assistant, switch to the Automations tab, pick a template or create your own, set cadence, and choose which tools may run without per-step approval. Plan review still applies; pre-approved tools skip repeated confirmation for routine actions.

Example morning automation: read Gmail and Calendar (and Slack if connected), tier inbox items, draft responses for low-urgency mail, flag what needs your brain, write meeting prep notes, save Morning-Briefing-[DATE].md to OUTPUTS/daily-briefings/.

Automations are assistive, not autonomous law. You remain responsible for reviewing outputs and revoking tool access in Settings → Integrations.

Step 7: Token discipline

Cowork sessions cost more than quick chat because they reload vault context and run multi-step plans. Paid plans include monthly AI token budgets; heavy daily Cowork without discipline will hit limits faster.

Practical rules from production use:

Keep ABOUT ME files small. Every session starts by reading them.

Batch related tasks in one run instead of three separate sessions that each reload context.

When the thread goes off track, restart with a cleaner brief rather than ten corrective follow-ups—long threads re-send growing context.

Use lighter models for formatting and simple edits; reserve frontier models for research, synthesis, and multi-tool missions.

After roughly twenty exchanges, ask Cowork to summarize, start a fresh session, and paste the summary as message one.

Personal Assistant includes a token coach that surfaces usage guidance in the Cowork UI.

Optional: visualization presets

Cowork can attach lightweight chart and table presets to certain reports—SWOT briefs, comparison tables, executive one-pagers, trend summaries, positioning matrices, research briefs, morning briefing tables, and content drafts.

Presets are opt-in: toggle "Include charts" when starting a run. They add structured visuals without running a full heavy pipeline, and they respect token tiers (low/medium) per preset.

If you do not need charts, leave the toggle off to save tokens.

Integrations Cowork can use

With your OAuth grants, Cowork can read and act across Google Workspace (Gmail, Calendar, Drive, Contacts), Notion, LinkedIn, Slack, and messaging channels. Browser sessions may be used for approved web workflows.

Each integration is optional and revocable. Cowork only accesses what you connect and what you approve per mission or automation.

Limitations (read this before you rely on Cowork)

Cowork is not fully hands-off. You must review plans and approve sensitive actions unless explicitly pre-approved on an automation.

Output quality depends on vault quality. Empty or generic ABOUT ME files produce empty or generic work.

Token budgets are finite on every plan. Viz presets and long sessions consume more.

Some marketing copy references voice dictation and advanced slash-command browsers; those capabilities are not fully shipped—typed briefs and in-app clarifying forms are the supported path today.

Cowork writes to your Drive vault; it does not replace legal, financial, or compliance review of AI-generated content.

Automations run on schedules you define; they do not guarantee perfect categorization of email or calendar edge cases—treat briefings as drafts.

EU AI Act and enterprise policy: Axionomy positions Cowork as limited-risk assistive automation with transparency (you see plans, approvals, and run history) and human oversight. Your organization may impose additional rules on automated email or external posts.

Getting started checklist

Connect Google and select a vault folder.

Create ABOUT ME files (use the interview wizard if helpful).

Set global instructions.

Run one Cowork task with AskUserQuestion enabled.

Save a good output as a template.

Optionally enable viz presets for reporting tasks.

Schedule one automation only after a manual run proves the workflow.

Where to go next

Open Personal Assistant on axionomy.xyz, complete vault setup, and run your first Cowork mission with a real OUTPUTS path.

For plan limits and token tiers, see /pricing. For integration details and security, see /privacy and /docs.

Cowork does not remove you from decisions. It removes you from the production steps that should not require your brain—so you can spend time on the work only you can do.