Replace an agent currently in use

Stand up a Replacement Plan agent that turns an agent your team relies on into a clean handoff to a successor — who to use instead, until when, and what to detach — so bookmarks and muscle memory do not send people into a wall.

Plus: three prompts that catch replacement candidates before they quietly rot in Settings.

Audience Admins · Everyone
Time ~15 min
Prerequisites An Auxot account on any tier. [Build your agent directory](/tutorials/build-your-agent-directory) — you have or can update a directory. [Update your agents without breaking the team](/tutorials/update-your-agents-without-breaking-the-team) — you announce changes. Helpful: [Audit and clean up your agents](/tutorials/audit-and-clean-up-your-agents) for roster cleanup mindset (when the agent is unused, just delete; this tutorial is for agents still in use); [View your audit logs](/tutorials/view-your-audit-logs) to confirm traffic dropped; [Manage your Credentials](/tutorials/manage-your-credentials) when replacement means revoking dedicated credentials; [Split an overloaded agent into two](/tutorials/split-an-overloaded-agent-into-two) if merge beats delete.
You'll end up with A Replacement Plan agent, one written replacement memo for a real or candidate agent, and executed steps — redirect prose, directory row, optional credential cleanup — so nobody discovers the gap from a customer thread.

When a tutorial shows italic text in quotation marks, it usually mirrors a label or helper string inside Auxot. Product copy changes between releases — if something reads differently in your workspace, trust what you see on screen.

Callouts with a Worth knowing gold accent are meant as must-read context before you move on. Blockquotes that open with Tip are lighter, optional depth.

Why this matters

Agents outlive their usefulness: the workflow moved, the integration died, or a split (Split an overloaded agent into two) replaced them. What lingers is worse than clutter: stale instructions, old API keys still attached, and teammates who open the wrong chat because the name feels familiar. Replacement is not only delete row in Settings → Agents. It is redirect, communicate, detach, and verify traffic actually moved (View your audit logs).

The replacement-plan agent does not press delete for you: it produces the checklist and the one-paragraph “use B instead of A until [date]” language that belongs in Update your agents without breaking the team and your directory (Build your agent directory). Same respect you give migrations (Migrate agents when models or providers change) and credential work (Rotate credentials without surprising your agents), applied to people who type prompts.

Today you rehearse one replacement you have been avoiding.

Concept diagram: 59-concept Replacement Plan: usage check → successor → redirect instructions → comms → detach tools → archive.


Quick start

  1. Sign in: open Auxot in your browser and log in.

  2. Open chat with the Admin Agent: click Chat in the left menu, make sure the agent picker reads “Admin Agent.”

  3. Ask for a Replacement Plan agent: paste something like:

    Create an agent called "Replacement Plan." Its job: help me replace an Auxot agent that the team currently relies on — users, tools, and credentials included. It should prompt for: agent name, reason for replacement, and named successor agent or human process if any. Outputs: (1) replacement criteria checklist — traffic low enough, owner agrees, successor tested; (2) redirect block for job-description instructions (Give your agent its job description) — short, impossible to miss, with end date if we will delete the agent later; (3) announcement language for Update your agents without breaking the team — Slack and directory; (4) detachment list — tool policies, credentials used only here (Manage your Credentials), context files exclusive to this agent (Keep your context files honest and fresh); (5) post-replacement monitoring — what to skim in Jobs for two weeks; and (6) merge alternative — when Split an overloaded agent into two–style consolidation beats hard delete. Tone: respectful to users who relied on the agent; no blame.
    
  4. Answer the Admin Agent’s questions: it may ask whether the agent was customer-facing.

  5. Plan one replacement: pick an obvious weakling or a duplicate. Finish the checklist on paper before you change Settings.

Done? Replacement Plan is in Settings → Agents. You have a replacement memo you could execute in one maintenance window.


The agent can do that?

Replacements fail when successors are untested. These three prompts reduce thrash.

1. Have the Admin Agent compare recent usage to your guess

Open chat with the Admin Agent:

We are considering replacing agent [name]. Based on recent threads or jobs if visible, is it still in active use compared to [successor]? Three bullets — evidence, uncertainty, what to watch after redirect.

Why it’s non-obvious: Product judgment says dead; Audit Logs may disagree. Checking first prevents surprise outrage (Collect useful feedback on your agents).

2. Ask the replacement-plan agent for a hard delete vs soft redirect

After facts exist:

Recommend hard delete on date X vs soft redirect forever — pros and cons for our risk level. If soft, what minimal instruction block keeps the agent from doing real work?

Why it’s non-obvious: Hard delete cleans Settings; soft redirect preserves bookmarks during long migrations. Naming the tradeoff picks the right drama level.

3. Inverted usage: bring-it-back criteria

Ask the replacement-plan agent:

Under what conditions would we bring this agent back or fork it again — one paragraph for the runbook so we do not debate from scratch during an incident.

Why it’s non-obvious: Sometimes replacement is temporary. Writing the return clause prevents “we deleted the only thing that knew how to do X.”


Go deeper

Order of operations

Usually: announce (Update your agents without breaking the team) → add redirect to instructions → monitor → detach tools (Define a tool policy) → revoke dedicated credentials (Rotate credentials without surprising your agents) → remove directory row → delete agent when traffic is flat.

Merges first

Split an overloaded agent into two: if two agents overlap, merge capabilities into the survivor, then replace the loser: fewer user-facing names to track.

Troubleshooting

Variations & edge cases

  • Compliance retention: you may keep read-only exports outside Auxot even after delete; say so in internal policy.
  • Seasonal agents: soft redirect with return date in directory instead of delete.

Walkthrough

Step 1: Create Replacement Plan

Use the Quick start prompt in Chat with the Admin Agent.

Screenshot: 59-step-1 Admin Agent confirms: checklist, redirect, comms, and detach list.

Step 2: Confirm usage and owner

Skim Audit Logs (View your audit logs). Get explicit OK from the named owner: replacement is a people process too (Hand off agent ownership cleanly).

Screenshot: 59-step-2 Jobs or threads — evidence of remaining use.

Step 3: Patch instructions and announce

Paste the replacement-plan agent’s redirect block into the agent being replaced (Give your agent its job description). Post announcement notes per Update your agents without breaking the team; update Build your agent directory.

Screenshot: 59-step-3 First lines of instructions — bright redirect.

Step 4: Detach and delete on schedule

Remove tool policies from the agent; move or revoke credentials per Rotate credentials without surprising your agents; delete the agent row when the monitoring window says traffic is gone.

Step 5: Log the lesson

Add one line to Run a quarterly review of your agents prep or internal wiki (why we replaced it) so the agent-roster narrative stays honest.


What’s next

Reference