Review the assignment, inputs, actions, and result.
Keep the context, tools, output, and operational result in the same review.
CAMPUS · AI AGENT MANAGEMENT FOR OPERATIONS TEAMS
Campus helps operations teams review and improveAI agents.
AI agent review and coaching
Campus stores the assignment, agent run, failure, human feedback, changed instruction, approval, and retest in one review record. It is designed for teams that need to understand why an agent failed and whether a correction worked.
THE REVIEW LOOP
Campus uses a four-step review process.
Review the run, identify the failure, approve a correction, and compare the next result.
Classify the failure before changing the agent.
Separate instruction, context, tool, judgment, and execution errors.
Write and approve the correction.
Record the changed instruction, examples, boundaries, permissions, and required approval.
Run the same test cases again.
Compare the next result with the prior failure and record who approved the change.
MARKDOWN · SAFETY · TRUST
Campus stores instructions and feedback in Markdown.
We chose a Markdown-first foundation because Markdown is pure text: readable by humans, native to AI systems, easy to diff, and durable enough to hold shared instructions, feedback, coaching records, and trust boundaries.
People and agents use the same instructions.
The instructions remain readable without converting them into a proprietary format.
Review what changed between runs.
Versioned instructions make the approved change visible during the next review.
Record who approved each change.
Campus keeps the reviewer, instruction change, approval requirement, and later result together.
THE CAMPUS MAP
Campus uses named review modes.
Each mode supports a specific part of reviewing, testing, or improving an agent.
Review a miss without overreacting.
Separate evidence, cause, and corrective action before changing the agent.
Review one assignment together.
Put the assignment, explanation, manager feedback, and next expectation in one conversation.
Test a new instruction before live use.
Challenge edge cases and verify the approval boundary before the agent returns to work.
Keep corrections that improve performance.
Convert coaching into durable instructions and compare later runs.
Each review remains tied to the assignment, evidence, feedback, and next run.
THE AGENT PIP
A performance improvement plan built for agents.
When an agent misses the standard, Campus gives leaders a practical response: define the gap, diagnose the cause, teach the expected behavior, and review the next runs together.
Define the expected behavior and approval conditions.
Record the assignment, boundaries, required judgment, and human gate.
Record where the run failed.
Keep the relevant context, tools, decisions, handoffs, and failure point.
Apply the approved change and compare later runs.
Monitor the same failure category after the new instruction is in use.
CONVERSATIONAL REVIEW · VERSIONED RECORD
Campus stores the review conversation and the resulting changes.
The record includes the assignment, context, feedback, instruction change, approval, and later result.
Store the review with the agent run.
Keep the assignment, context, feedback, and follow-up together.
Turn feedback into a versioned instruction.
The next run uses an explicit, reviewable change instead of a note lost in chat history.
Record the reviewer and approval.
Leaders can see who changed the instruction, who approved it, and what the agent did afterward.
Review one failed run and its retest.
See the evidence, correction, approval, and next result in one record.