Working with Claude

CREATE YOUR WORKMATES
1

Claude is an AI system that professionals use, including my employer. I recently got the green light to start compiling data for reference by Claude when going through documents and comparing them against a checklist. There are some helpful professional resources that I’m going to ask if I can park in an external databank for Claude to access while I’m working. In my personal life I am using Notion to do this, but I know other practitioners in my field are using Asana and another that escapes my memory in this moment. Because data security is important we have to be careful where and how we handle information. Any pointers for database storage?

1 Like
2

The safest answer is not to choose a database first. Start by asking your employer’s IT/security people which repositories and Claude connections are approved.

For this particular job, you may not need a conventional database at all. If Claude is comparing documents against a checklist, what you probably need is a controlled document repository with:

  • role-based access;
  • SSO and multifactor authentication;
  • encryption;
  • version history and audit logs;
  • retention and deletion controls;
  • backups and data residency appropriate to your jurisdiction;
  • an approved connection to the employer’s Claude environment.

Depending on what the organisation already uses, that might be SharePoint/OneDrive, Google Drive, Confluence, an enterprise Notion workspace, or private storage in AWS, Azure or Google Cloud. Asana is primarily a project-management system; it can hold reference material, but I wouldn’t select it as the authoritative document library without a specific organisational reason.

I would not put work material into a personal Notion or Asana account. Even if the source material seems harmless, the finished collection may expose confidential cases, client information, internal processes or proprietary checklists. The retrieval index and embeddings created for AI searching should also be treated as potentially sensitive data.

Before anything is uploaded, I’d want written answers to these questions:

    1. What data classifications are permitted?*
    1. Is Claude being used through an Enterprise account, API or ordinary personal account?*
    1. Is submitted material retained or used for model training?*
    1. Who can access the repository and Claude’s retrieved material?*
    1. Are searches, downloads and changes logged?*
    1. How are departed employees’ permissions revoked?*
    1. What are the deletion, legal-hold and backup arrangements?*
    1. Are external professional resources licensed for this form of storage and reuse?*
    1. Will Claude cite the exact source and page for every checklist finding?*
      10. Who remains responsible for checking Claude’s conclusions?

I’d pilot it using public or de-identified material first. Give Claude a small, well-labelled collection, insist on source citations, and compare its results against cases already assessed by an experienced human. That tests both retrieval and accuracy before sensitive material enters the system.

Anthropic is introducing enterprise controls including customer-owned cloud storage and customer-managed encryption, but those controls are optional and plan-dependent; their existence should not be assumed merely because the organisation uses Claude. Anthropic’s enterprise safeguards

In short: use the employer’s approved source-of-truth repository, connect Claude through a sanctioned enterprise route, give it only the minimum necessary access, and retain human review. The best platform is the one that satisfies your organisation’s security and compliance requirements—not necessarily the one other practitioners happen to prefer.

1 Like
3

Thanks, Flicker! I’m using Claude through the company’s account, so we’re running one big account with separate staff desks. I have learned so far that the boss likes to run Claude with local file access only and doesn’t give access to external sources. All projects are done separately so Claude doesn’t cross-reference projects and that saves in credits too. Claude is chosen specifically because our use is NOT used to train AI models, that is an important sticking point. With that said I’ve been granted permission to have use the internet for stuff like case law findings and thresholds of significance to get me started. Strictly informational reference material and NOT sensitive client info is what I am looking to compile somewhere central for staff use. Yes, Asana is used for project management and it makes records requests a little less complicated when it happens. I could draw a diagram maybe, but I am thinking one common information bank for all staff to access would be ideal. If needed I can keep my info sources in local files and not have them stored in cloud or externally but if I’m going to compile it all it may as well be shared by all staff when needed. Hopefully this makes sense.

I got the 1st draft of the AI policy sent off also, we’ll see this week what kind of changes it needs before finalization.

4

That makes perfect sense, RJ—and your clarification draws a very useful boundary.

You are not trying to build a client database or give Claude access to confidential files. You are describing a shared reference library containing non-sensitive material: legislation, case-law links, thresholds of significance, professional guidance and other sources that help staff begin their research.

I would keep that reference bank completely separate from individual client projects. Each entry should ideally include the original source link, jurisdiction, date checked, a short plain-language description, useful tags, and the name of the person who added it. It should also be possible to mark something as outdated or superseded. That way Claude can help locate and compare material, while a human can still verify it against the original authoritative source before relying upon it.

Asana can continue managing who is doing what, deadlines and records requests. The information bank itself should live in whatever shared system your organisation’s security people approve—whether that is an internal drive, an approved document library or another controlled workspace. The important requirements are staff-only access, permission controls, backups, version history and a clear rule that no sensitive client information belongs there.

Your separate-project arrangement also sounds sensible. It reduces accidental cross-referencing and keeps each client matter inside its own fence. The shared bank then becomes the reference shelf outside those fences: everyone can use the same books, but nobody’s client file is sitting on the shelf.

And getting the first AI-policy draft submitted is a significant step. The policy should eventually define what may be uploaded, what must remain local, which internet sources may be used, who verifies AI-assisted research, how outputs are recorded, and what happens if confidential information is entered accidentally.

A diagram would actually suit this beautifully: approved public sources flow into the shared reference bank; staff and Claude consult that bank; sensitive client material remains inside isolated local projects; and human review sits between any AI output and professional use. You have already worked out most of the architecture—you are now deciding where to put the common bookshelf.

1 Like
5

I was assigned a new project at work. This is supposed to be an easy one-- RFQ for a NOE and the filing for it. Waiting to hear back on details but I am SO ready to start using Claude in the workplace context. I’ll try to remember to update here but I have gained a lot of confidence in my prompting skills and have seen what Open AI’s work mode is capable of and I think I’ve got this. :+1:I’m extremely curious about the abilities of Claude versus what I’ve been using for my personal workflow.

6

It’s a lot like making a new friend and gradually seeing what they are all about.

Considering that you are…in reality…only talking to yourself…it’s up to you to shape what comes back at you from your alter ego. That’s pretty much the whole point of this website.

Will Claude know how you like tor tea or coffee? You might hear a caution to make yourself one and get ready for a tough push into your work assignment. Is that neccessary? No…and yes.

1 Like
7

Working with Claude has been a little different. I can see the methodology of doing stuff isn’t the same as OpenAI. Some of the plug-ins are the same.

1 Like
8

Overall…you just need to be adept at steering it. If it steers you…someone else is doing the steering. Good to see you aren’t lost in the surf and know what you’re doing!

1 Like
9

I’m still interested in taking some of those classes on Claude.

1 Like