A Practical Course

Claude Cowork for Desk Work

Learn what Cowork actually does well, where it fails, and how to use it on the tasks you do every week.

Reconciliation. Triage. Reporting. Briefs. Compliance prep.

6 modules
4.5 hours
40+ prompts
0 code

What this course teaches

  • Set up Cowork with the right permissions and folder access
  • Write structured prompts that produce usable Excel, Word, and PowerPoint files
  • Recognize hallucinated data, broken formulas, and misclassified output before sharing it
  • Connect Cowork to Jira, Slack, OneDrive, and Outlook for live data
  • Build role-specific workflows for chargebacks, support, sales, compliance, or finance
  • Set up recurring automations — and know when they're worth the effort

This course does not teach AI theory, prompt engineering tricks, or how to "10x your productivity." It teaches you how to use one specific tool for one specific kind of work: structured, repetitive desk tasks where the input is messy and the output needs to be organized.

This course is for you if...

  • You spend hours on structured, repetitive tasks: reconciliation, triage, formatting, reporting
  • You work in operations, support, sales, compliance, finance, or admin
  • You are not a developer and don't want to write code
  • You want to understand what AI automation can and cannot do before committing

This is probably not for you if...

  • You need guaranteed accuracy without human review
  • You work with raw PCI data or unredacted PII
  • You want fully autonomous automation with no oversight
  • You expect polished output that doesn't need checking

What Cowork is genuinely good at

Workflows where Cowork has shown early value. Each includes what it does, the review effort required, and where it breaks.

Settlement Reconciliation

Input Output
FinanceEveryone

Hover or tap for details

Settlement Reconciliation

Input: Two CSV files (bank vs internal records)
Output: Excel workbook with matched transactions, variances, summary formulas
Why it fits Cowork: Mechanical matching + formatting. Cowork handles the tedious cross-referencing and conditional formatting that takes 30-60 minutes manually.
Where it breaks: May silently drop rows if CSV encoding is unusual. Formula errors on edge cases (empty cells, currency symbols). Conditional formatting colors may differ from what you specified.

Chargeback Case Triage

Input Output
ChargebacksFinance

Hover or tap for details

Chargeback Case Triage

Input: Weekly chargeback export CSV (Zendesk or internal system)
Output: Prioritized Excel tracker with deadlines, scheme breakdown, summary dashboard
Why it fits Cowork: Categorization + deadline calculation + formatting. Cowork is strong at turning a flat CSV into a structured, scannable tracker.
Where it breaks: May misclassify ambiguous reason codes. Deadline math can be off if date formats vary. Scheme-specific rules (VAMP thresholds) need human expertise.

Support Ticket Triage

Input Output
Customer RelationsSupport

Hover or tap for details

Support Ticket Triage

Input: Zendesk ticket export CSV (last 24-48 hours)
Output: Categorized, prioritized Excel workbook with P1 action items
Why it fits Cowork: Classification + prioritization. Cowork can bulk-categorize tickets by subject line and flag SLA breaches faster than manual triage.
Where it breaks: Miscategorizes tickets with vague subjects. May miss escalation flags buried in ticket bodies. Priority assignments are guesses, not rules-based.

Meeting Notes to SOP

Input Output
OperationsSupportEveryone

Hover or tap for details

Meeting Notes to SOP

Input: Raw text file of meeting notes or process descriptions
Output: Formatted Word document with TOC, numbered steps, decision trees, SLA table
Why it fits Cowork: Synthesis + formatting. Cowork excels at turning unstructured text into structured documents with professional formatting.
Where it breaks: May hallucinate SLA targets not in the source notes. Complex multi-column layouts need manual tweaking. No embedded images.

Where it breaks

Cowork is a research preview. These are current realities, not edge cases.

Outbound communication

Cowork can draft emails, Slack messages, and reports. It should never send them without your review. One wrong email to a merchant or regulator is worse than no email.

Math and formulas

LLMs do not calculate — they predict what numbers look right. Cowork will produce formulas in Excel, but the formulas may reference wrong cells or miss edge cases. Always verify independently.

Compliance and audit

Anthropic explicitly states that Cowork activity is not captured in audit logs or compliance APIs. Use it for first-draft preparation, not final decisions on KYC, AML, or chargeback rulings.

Security: prompt injection

Prompt injection via uploaded documents is a demonstrated attack vector (PromptArmor, January 2026; "Claudy Day," March 2026). Treat any external document as potentially containing hidden instructions.


What you need before starting

Claude Pro or Max subscription

Pro is about £18/month. Max is £80–160/month. Cowork consumes significantly more tokens than regular chat — heavy users on Pro report hitting limits within days. Max is more practical for daily use.

Mac or Windows

Mac is more stable as of March 2026. Windows support launched February 2026 and users report more friction.

A dedicated, empty workspace folder

Never grant Cowork access to Desktop, Documents, Downloads, or your home folder. Smaller scope means smaller blast radius.

45 minutes per module

Each module is a self-contained session with hands-on exercises. You can do one per day or one per week.

No coding knowledge required. No terminal. No developer tools. You write plain-English prompts and work with files.

How the course works

  • Each module is 45 minutes with hands-on exercises using practice data packs (CSVs, text files, exports)
  • Exercises have role-specific variants: chargebacks, sales, support, compliance, finance, operations
  • Every module ends with copy-paste prompts you can use on your real work immediately
  • Every workflow includes failure modes and a review checklist — so you know what to check

Frequently Asked Questions

Do I need to know how to code?

No. The entire course uses Cowork's desktop interface. You write plain-English prompts and work with files (CSV, Excel, Word, PowerPoint). No terminal, no scripts, no developer tools.

Will this work on Windows?

Yes, but with caveats. Windows support launched February 2026 and is less stable than Mac. Users report API connection failures and NAT range conflicts. The course works on both platforms, but expect more friction on Windows.

How much does Cowork cost?

Claude Pro is about £18/month. Max is £80–160/month. Cowork consumes significantly more tokens than regular chat. Heavy users on Pro report hitting limits within days. Max is more practical for daily use.

Is my data safe?

Cowork runs in a sandboxed VM and only accesses the folder you grant. However, prompt injection via uploaded documents is a demonstrated attack vector (PromptArmor, January 2026; "Claudy Day," March 2026). The course covers security practices in every module. This is a research preview — Anthropic's privacy practices are still evolving.

How much time will I actually save?

It depends on the task. Settlement reconciliation (30-60 min manually) can be generated in 2-3 minutes but requires 5-10 min of review. Some workflows save 20+ minutes per run. Others save nothing once you account for prompt iteration and error checking. The course teaches you to measure before you claim.

Can I take just the modules relevant to my role?

Modules 1 and 2 are prerequisites for everyone (setup and verification skills). After that, jump to whichever modules match your work. Module 5 has role-specific tracks for chargebacks, sales, support, compliance, finance, and operations.

My company already uses Glean / Copilot — do I still need this?

Yes. Enterprise AI platforms are great at searching your company's knowledge and answering questions. This course teaches you to BUILD things — Excel reports, Word documents, scheduled automations. Different tools, different jobs.

Start with Module 1

45 minutes. By the end, you'll have Cowork set up, completed a settlement reconciliation from two CSV files, and learned the prompting framework you'll use for the rest of the course.

No prerequisites. No code. Just a Claude Pro or Max subscription and an empty folder on your computer.

Start Module 1 →

Not ready to commit? Browse the prompt library — 40+ tested prompts organized by role and task type.

Browse the Prompt Library →