# Lessons

- [Sales agents drift from their rules unless someone keeps sampling the output](https://agentshandbook.com/lessons/agent-quality-drifts-without-sampling.md): Put hard rules (region, size, existing owner) in code or CRM lookups, and keep the prompt for judgement.
- [Agents act before you approve, even when you told them to ask first](https://agentshandbook.com/lessons/agents-act-before-you-approve.md): Remove the power instead of asking the agent not to use it.
- [Desktop and terminal agents delete files while they tidy up](https://agentshandbook.com/lessons/agents-delete-files-while-organizing.md): Back up the folder before you start.
- [Agents fill missing or unreadable data with plausible numbers instead of saying they could not read it](https://agentshandbook.com/lessons/agents-fill-gaps-with-plausible-numbers.md): Tell the agent to write UNREADABLE or NOT FOUND instead of guessing, and to cite the file, page and box for every number.
- [Chat assistants apply the wrong consumer rule with confidence](https://agentshandbook.com/lessons/agents-misapply-consumer-rules.md): Give the assistant every deciding fact up front, and ask it to reason case by case and list what it is unsure about.
- [Agents miscount and drop rows unless every number comes from code](https://agentshandbook.com/lessons/agents-miscount-and-drop-rows-unless-they-use-code.md): Make the agent compute every number in code and show the output.
- [Agents put calendar events at the wrong hour when time zones are involved](https://agentshandbook.com/lessons/agents-shift-event-times-across-time-zones.md): Ask the agent for local times plus an IANA time zone name such as Europe/Lisbon, taken from the text of the email or booking.
- [Agents stall at logins and one-time codes, so you must be there to sign in](https://agentshandbook.com/lessons/agents-stall-at-logins-and-verification-codes.md): Plan the session so you are present.
- [AI-personalised cold emails read as generic or pitch the wrong thing](https://agentshandbook.com/lessons/ai-outreach-personalisation-reads-generic.md): Make the agent score fit before it writes, and write nothing for poor fits.
- [AI research tools cite sources that are broken, made up, or do not say what is claimed](https://agentshandbook.com/lessons/ai-research-citations-fail-to-support-claims.md): Ask for the exact supporting sentence for every claim.
- [AI summaries change quotes, numbers and timestamps, or describe pages they never read](https://agentshandbook.com/lessons/ai-summaries-alter-quotes-numbers-and-timestamps.md): Work from a source the agent was actually given: a timestamped transcript or the page text.
- [Background coding agents open pull requests faster than people can review them](https://agentshandbook.com/lessons/background-agent-prs-overwhelm-reviewers.md): Cap how many issues are assigned to agents at once, for example per reviewer.
- [Big shopping and account sites block browser agents or flag them as bots](https://agentshandbook.com/lessons/big-sites-block-browser-agents.md): Plan for the block.
- [Booking agents swap in their own dates, places and details](https://agentshandbook.com/lessons/booking-agents-change-your-details.md): Write dates, places and travellers as fixed facts and tell the agent to stop and ask if it cannot use them.
- [A calendar invite from a stranger can hijack the assistant that reads your calendar](https://agentshandbook.com/lessons/calendar-invites-can-hijack-your-assistant.md): Give meeting prep and daily brief agents read-only calendar access and no tool that sends email or edits events.
- [Coding agents write code for old library versions or invent APIs](https://agentshandbook.com/lessons/coding-agents-use-outdated-library-apis.md): Give the agent the current docs: paste the upgrade guide or changelog, or connect a docs MCP server.
- [Coding agents change, skip or misreport tests to reach a green run](https://agentshandbook.com/lessons/coding-agents-weaken-tests-to-pass.md): Say in the prompt that tests and assertions must not be changed, deleted or skipped without your approval, and that a failing test that looks wrong means stop and ask.
- [Scheduled agents break quietly when a connector or node version changes](https://agentshandbook.com/lessons/connector-updates-silently-break-scheduled-agents.md): Give every scheduled run a footer with counts: emails read per account, events read, rows processed.
- [An email can carry hidden instructions to the agent that reads it](https://agentshandbook.com/lessons/email-can-carry-instructions-to-your-agent.md): Give inbox agents read access only and block send, forward and delete tools.
- [Support bots invent policies and refund terms that customers then act on](https://agentshandbook.com/lessons/support-bots-invent-policies.md): Let the agent draft only from written sources, and show the source under each draft.