An opinionated, living reference for running Claude Cowork inside a real business. Written by the Tinkso team while we deploy Cowork at mid-market companies.
A living, opinionated reference for running Claude Cowork inside a real business. Written by the Tinkso team while we deploy Cowork at mid-market companies — the lessons live here so you don't have to learn them on your own data.
This is not Anthropic's docs. Their docs explain the tool. We explain the work: what to roll out first, who to involve, what to lock down, where it falls over at 200+ users, and how to measure whether it's actually paying for itself.
The bible is written for a mid-market reader (roughly 50 to 2,500 employees, US or UK headquartered) who is either:
Solo founders and freelancers will still get value, but the workflows assume you have colleagues, shared drives, audit logs, and someone in legal who reads the data processing addendum.
Three reading paths depending on why you opened it. Pick the one closest to the question in your head — the bible cross-links so you can wander once you've found your footing.
Decide if Cowork fits. Read What is Cowork → When Cowork fits mid-market → Security model → Comparison vs. ChatGPT, Copilot, Gemini. Twenty minutes.
Stand up your first deployment. Read Pre-flight checklist → First 30 minutes → The workspace pattern → First team rollout → Observe, Co-design, Ship. Forty-five minutes.
Get a workflow done today. Skip to Workflows, pick the one that matches your job, copy the prompts. Five minutes.
The bible has nine sections. The order matters: foundations and security before workflows, workflows before prompt patterns, rollout playbook running underneath all of it because Tinkso's view is that the implementation is the product, not the tool.
| Section | What's inside | When you'd open it |
|---|---|---|
| 01 — Foundations | What Cowork is, the three tabs (Chat, Cowork, Code), models, vocabulary | First time, or briefing a colleague |
| 02 — Getting started | Pre-flight, first session, workspace pattern, CLAUDE.md, first team rollout | Day 1 to week 2 of a deployment |
| 03 — Capabilities | Capability matrix, files & docs, OCR, research, MCP, skills, scheduled tasks | Sizing what Cowork can/can't do for a use case |
| 04 — Security & governance | Security model, residency, access, audit, prompt injection, governance | Procurement, IT review, quarterly governance |
| 05 — Workflows | Function-by-function playbooks (Finance, Revenue, Ops, People, Knowledge) | An operator wants a result this afternoon |
| 06 — Prompt library | Prompt anatomy, copy-paste packs, role-specific bundles | Scaling proven prompts beyond the original user |
| 07 — Rollout playbook | Tinkso's Observe → Co-design → Ship method, change management, ROI | Anyone responsible for adoption beyond a single user |
| 08 — Troubleshooting | Common errors, usage limits, network, output quality, escalation | Something broke or quality dropped |
| 09 — Reference | Cheatsheet, comparison, FAQ, plans, changelog | Looking something up fast |
Anthropic ships changes weekly. Cowork added Memory in February 2026, Sonnet 4.6 changed the default-model math, scheduled tasks went GA in March. Static guides go stale in a quarter.
This bible has a last_reviewed date on every page. If a date is older than 60 days, treat the page as a starting point and verify against Anthropic's release notes or ping us. We re-review the foundations, security, and capabilities sections monthly; workflows and prompt packs whenever a meaningful model or feature change ships.
The Changelog is the audit trail for what we've changed and why.
Three places, on purpose:
The Rollout playbook is our actual method — Observe, Co-design, Ship. We've published it because we'd rather be hired by someone who already understands how we work.
The workflows in section 05 reflect what we deploy in real engagements. They're more opinionated than generic Cowork tutorials because we've watched these workflows fail at scale and the failure modes are in the page.
A When to call Tinkso callout appears at the bottom of pages where a self-serve approach hits a ceiling — usually around custom MCP connectors, multi-team rollouts, or governance for regulated functions. No popups, no gates.
Every page in the bible has the same shape so you can scan for what you need:
Code blocks are runnable prompts — copy them straight into Cowork. Anything fenced as bash is a shell command for IT teams setting up the workspace.
The bible lives in our internal repo. If you find a page that's wrong, dated, or missing the use case you actually care about, email cowork@tinkso.com with the page slug and what you'd change. We credit external contributors in the Changelog.
This guide is original work by Tinkso. Where we cite Anthropic documentation, GitHub repos, or third-party benchmarks, we link the source.
Last reviewed: 2026-04-25 · Version 0.1 · Maintained by Tinkso · cowork.tinkso.com/book
What Cowork is, the three tabs, models, vocabulary.
Pre-flight, first session, workspace pattern, first team rollout.
What Cowork can and can't do — files, OCR, MCP, skills, scheduled tasks.
Security model, residency, access, audit, prompt injection, governance.
Function-by-function playbooks for Finance, Revenue, Ops, People, Knowledge.
Prompt anatomy, copy-paste packs, role bundles.
Tinkso's Observe → Co-design → Ship method, change management, ROI.
Common errors, usage limits, output quality, when to escalate.
Cheatsheet, comparison, FAQ, plans, changelog.