Improve your AI agent

You can tell Chatlane when your AI agent did well or poorly—and review a log of what the agent did—so your team can improve behaviour and quality over time. This article explains how to give feedback (e.g. thumbs up/down and categories) and how to use agent logs.

Giving feedback on agent replies

When the agent suggests a draft reply or adds a note, you can give quick feedback so Chatlane (and your team) knows how that reply performed:

• Thumbs up — The reply was helpful, accurate, or on-brand. Use this when the agent got it right.
• Thumbs down — The reply was off, unhelpful, or not what you wanted. Use this when the agent missed the mark.

When you give feedback, you can often choose a category (e.g. "accurate," "tone," "incomplete," "wrong topic") and add an optional comment. That helps managers and founders see patterns—for example "the agent often gets the tone wrong" or "incomplete answers on billing questions"—so you can adjust the agent's instructions or knowledge base.

Feedback is stored with the conversation and the agent run, so you can look back at what the agent did and what feedback it received. You don't have to give feedback on every reply; even a little feedback over time helps improve the agent.

When feedback is required

Your team can turn on optional per-inbox settings so that:

• Require feedback on deletion — When someone deletes an agent's draft without sending it, they're prompted to give feedback first (e.g. why it wasn't used). That way you capture "why we didn't send this" instead of losing that signal.
• Restrict draft deletion — While an agent draft exists, the Reply or Agent Reply buttons can be disabled until the draft is sent, edited, or deleted (with feedback if required). That helps avoid accidentally sending a different reply without reviewing the agent's suggestion.

These settings are useful when managers want to ensure the team is consistently feeding back on agent quality.

Reviewing agent logs

Agent logs show what the agent did: when it ran (automatically or manually), whether it succeeded or failed, how long it took, and what actions or tools it used (e.g. updated status, added a tag, called an external system). Logs are grouped by conversation so you can open a conversation and see all agent runs for that thread.

Managers and founders can use logs to:

• Tune behaviour — See which runs failed or got negative feedback, and adjust the agent's instructions or knowledge base.
• Check quality — Spot patterns (e.g. the agent often fails on a certain type of question) and decide whether to add more content or change the agent's role.
• Audit — See what actions the agent took (e.g. tags added, status changed, webhooks called) for compliance or process review.

You can filter logs by date, status (success/failed), and other criteria so you can focus on the runs that matter.

Why this helps your team

• Support teams can quickly mark good and bad replies so the product improves without extra meetings.
• Managers can use feedback and logs to improve agent quality, assign follow-up work (e.g. "review failed runs this week"), and report on agent effectiveness.
• Founders get a clear picture of agent ROI and where to invest in better instructions, knowledge, or actions.

For giving your agent a knowledge base (files and message attachments), see Give your AI agent a knowledge base. For letting the agent take action (e.g. update status or tags), see Let your AI agent take action. For high-level agent and team metrics, see See how your support team is performing: reports and analytics.