6d • General discussion
Automation using Claude and Python
Looking for people who've run LLM automations inside a large enterprise
I'm starting a contract building small internal automations for a few non-engineering teams (technical writing, training, process). They need to be in real use quickly, then handed off to someone else to maintain.
My plan:
- Python does the deterministic work: pull closed tickets from Jira, filter, check fields, format.
- Claude is called only at a few judgment steps, with structured output. Python validates the result, e.g. every ticket it cites must exist in the input.
- A person reviews every output before it goes anywhere.
- No agent loops. Fixed steps, so it's testable and someone else can maintain it.
Example: a monthly summary of closed tickets that someone currently writes by hand.
Questions for anyone who's done this:
1. Did you go with scheduled scripts calling an API, or skills a person runs inside Claude? What decided it?
2. How do you catch invented facts before a human sees the output?
3. Anything break when reading Jira during a Data Center to Cloud migration?
4. What made your handoff docs actually usable by the next maintainer?
Less interested in tool recommendations, more in what went wrong and how you fixed it.
2
1 comment
Automation using Claude and Python
powered by
skool.com/zazencodes-agentic-coding-club-7823
Learn how to use coding agents like the top 1%. Become the AI expert on your team. Welcome to the official ZazenCodes community.
Suggested communities
Powered by