Most automation backlogs aren't blocked by a lack of developers. They're blocked by the gap between the person who understands the problem and the person who can write the script.
Claude Code closes a lot of that gap. It's Anthropic's agentic coding tool: it reads files, writes and runs code, and works through multi-step tasks while a person reviews each change. That means an analyst who knows exactly why the weekly pack takes four hours can build the fix, as long as they know how to scope the task, guide the tool and check its output.
This post is a practical plan for teaching that to analysts, MIS, finance and operations teams. The focus is not programming syntax. It's scoping, guiding, reviewing and maintaining.
Who should be in the room
- Analysts and MIS teams with recurring reports built from exports
- Operations staff doing the same clean-up every week
- Finance teams running reconciliations and recurring packs
- Technical and product teams who want a faster way to prototype internal tools
One prerequisite from managers: name the repeated tasks before the session. The best builds come from real work, not toy examples.
Coding skills aren't required, but people should be comfortable with files, spreadsheets and simple logic.
A six-block workshop agenda
| Block | What happens | Output |
|---|---|---|
| 1. Setup and safety | Create a project folder and agree rules on data and permissions | A safe working space |
| 2. Scoping a task | Write a brief: inputs, steps, expected output | A task brief |
| 3. First build | A script that cleans or merges a real export | Working script |
| 4. Review and fix | Add tests and checks so the output can be trusted | Verified output |
| 5. Second build | A weekly summary or report generated from source files | Second automation |
| 6. Handover | A readme so a colleague can run and maintain it | Documentation |
Treat this as a template. The actual builds should come from tasks the team shares beforehand.
Claude Code, chat tools, n8n or Copilot?
A common mistake is forcing one tool to do every job. A simple rule of thumb:
| Task type | Best fit |
|---|---|
| One-off drafting and analysis | Chat tools such as ChatGPT or Claude |
| Document work inside Microsoft 365 | Microsoft Copilot |
| Custom logic over files and data | Claude Code |
| Recurring, multi-app processes | n8n |
Most teams end up using all of them. Claude Code earns its place when the logic is specific to your data and too fiddly for a no-code workflow.
Use cases that work well as first builds
- MIS: a script that assembles the weekly pack from source exports
- Finance: a reconciliation check that flags only the mismatches
- Sales operations: a clean-up script that standardises names, fields and duplicates
- Procurement: a comparison table generated from each new quote file
Each of these has clear inputs, a clear output and a person who can tell immediately if the result is wrong. That last part matters more than anything.
Five guardrails for agentic coding at work
Agentic coding stays safe when every build follows a few rules agreed before the session:
- Work on copies of data or approved sample files during training.
- Review every change before it runs on anything important.
- Keep secrets out of prompts, files and scripts.
- Name an owner for each automation once it's in weekly use.
- Write a readme so the build survives a team change.
Rules 4 and 5 are the ones teams skip, and they're the reason a useful script stops working the month its author changes roles.
What a team should leave with
- At least one working automation per participant
- A repeatable way to scope tasks before building
- Agreed review and safety rules
- A shortlist of next tasks, each with an owner
A hands-on workshop like this usually runs one to two days. Participants need Claude Code access under the company's plan and a laptop they can install it on.
The takeaway
The people closest to the problem can now build the fix. The job of training isn't to turn them into developers. It's to give them a safe, repeatable way to scope, build, check and hand over.
Originally published on Special Forces of AI by Rushabh Mehta, Agentic AI Enabler, who has trained teams at the Times of India, HDFC Bank and L'Oréal and built 100+ AI workflows.
Top comments (0)