Agents in the Wild

This directory is no longer updated. Last update: November 16, 2026.

Tennr made an AI agent the maintainer of an internal tool; recruiters shipped 15+ tickets in a month

Tennr's recruiters and People team ask Claude in a Slack channel to change their offer portal. Claude ships the code and deploys it, with no PR review in that one channel. One bug reached candidates.

Organization
Tennr
Industry
Healthcare software
When
Jul 2026
Evidence
Vendor case study

Reported outcome

In roughly a month, non-engineers shipped 15+ tickets on the internal offer portal through Claude, and got production changes back within the hour.

  • 15+ in roughly a month
    Tickets shipped
  • within the hour
    Turnaround

Caveats

Published by the company that sells the agent. Treat numbers as marketing until confirmed elsewhere.

  • Published by Anthropic, which sells Claude Tag, with quotes from a Tennr executive.
  • One tool, one channel, about one month. The page gives no failure rate.
  • The page reports problems: a publicly exposed copy API was flagged, a bug kept resetting equity on live offers, and another bug briefly showed candidates variable pay incorrectly.
  • In this channel Claude skips PR reviews by design. Tennr says that posture would be wrong for product or engineering channels.

What the agent does

Tennr is a patient orchestration platform. In July 2026 it launched an internal offer presentation portal, built with Claude Code. It made Claude Tag the main maintainer of that tool through one Slack channel.

Recruiters, People team members, hiring managers and RevOps mention Claude in the channel and ask for a change in plain language, sometimes with only a screenshot. Claude ships the code change, deploys it and reports back. It also wrote the onboarding documentation and handles permissions for the tool.

What happened

The team shipped 15+ tickets this way in roughly a month. Examples are a benefits section built from a PDF, a signing banner and admin controls.

Things also went wrong. A publicly exposed copy API was flagged, and Claude locked it down the same day. An import bug kept resetting equity on live offers, and Claude fixed it and corrected the values. Another bug briefly showed candidates variable pay incorrectly. Claude posted a notice with the affected window and the fix without being asked.

The team taught Claude its conventions in the channel: an emoji status legend and a ticket format with requester, commit link, screenshots and a test link.

What to copy

  • Set autonomy per channel. Give the agent freedom only where a mistake is cheap.
  • Make the agent write a ticket record for every change, with a commit link and a test link.
  • Ask the agent to report its own mistakes, with the affected time window.

Starter prompt

Written by us, not by Tennr or Anthropic. Skipping review is a real risk; this tool showed wrong pay to candidates once.

You maintain the internal tool {tool}. Requests come from non-engineers in this channel.

For each request: restate it, list the files you will change, and say what could break.
After you deploy, post: ticket number, requester, commit link, screenshot, test link.
If a change touches money, permissions or anything a person outside the company sees, stop and ask for a named approver.
If you find a mistake you shipped, post the affected window and the fix at once.

Sources

  1. How healthcare organizations use Claude Tag — Anthropic, Sep 14, 2026

Added Nov 16, 2026 · Sources last checked Nov 16, 2026.

Get new use cases weekly

One short email. Sources included.

The newsletter is paused. Past issues stay in the archive.

Related use cases