No source we checked shows a company running itself end to end. Agents can run defined tasks and processes, with humans setting direction and checking in. This checklist lists the red lines to set before an agent runs any business process.
Source basis (fetched Oct 17-18, 2026): Microsoft 2025 Work Trend Index (https://www.microsoft.com/en-us/worklab/work-trend-index/2025-the-year-the-frontier-firm-is-born); Anthropic, Building Effective Agents (https://www.anthropic.com/engineering/building-effective-agents); Salesforce Agentforce platform page, vendor material (https://www.salesforce.com/platform/agentforce-platform/).
The 7 red lines
- Money. No payment, refund or price change without human approval.
- Contracts and promises. The agent drafts; a person commits to customers.
- Named owner per process. One accountable person, not "the system".
- Fixed steps where possible. Use coded workflows for well-defined tasks. Anthropic notes they offer predictability.
- Limited tools. Each task lists exactly which systems it can read or change.
- Logs. Every action recorded with inputs and outputs, reviewed on a schedule.
- Off switch. A tested way to pause a process, with a named person who can use it.
Why red lines matter
Anthropic warns that autonomy means higher costs and compounding errors, and recommends sandbox testing with guardrails. Early mistakes can travel through a chain of agents.
Limits
This is not legal advice. Ownership, liability and labor rules depend on country and sector. No claim is made that any company runs itself today.