Make a clear, safer operating decision.
You will be able to write a one-page agent contract before configuring an agent, so everyone can see what it is meant to do and what it must not do.
Good agent work is useful before it is impressive.
‘Answer leads’ is not a job. An accountable job identifies who arrives, what starts the work, what good looks like, which actions are allowed, what is out of bounds, and who receives the exception.
Make the relationship visible.
The language that keeps the work clear.
Work through the decision in order.
Write the trigger
Use a specific event rather than ‘whenever needed.’
Name the user and result
State whose request the agent supports and the useful outcome they should reach.
Set allowed actions
List the exact questions, records, messages, or recommendations the agent may create.
Define failure and ownership
Describe when it stops, who receives the case, and how the owner knows what happened.
A realistic, bounded implementation.
A B2B software company uses an agent for inbound demo requests. Its contract says: when a new demo form arrives with a work email, summarize the stated problem, ask only for the missing qualification field approved by sales, create a CRM task for the regional owner, and send the approved acknowledgement.
It may not quote custom prices, promise product capability, or change account data. If a request concerns security, procurement, or an existing customer incident, it routes the conversation to the named team without improvising.
The acceptance condition is a complete record, clear owner, appropriate acknowledgement, and no unresolved high-risk issue.
Use this copyable working template.
Adapt it to the client’s evidence, policy, people, and tools. Do not treat placeholders as approved instructions.
Put the operating system around the agent.
Use a contact record, workflow trigger, task owner, approved message template, and automation conditions as the visible implementation of the contract.
Before you move on
- Write one agent contract on a single page.
- Ask the operating owner to identify one unsafe ambiguity.
- Remove any action you cannot explain or measure.
- The job is narrow and outcome-led.
- Permitted context and tools are named.
- Prohibited actions are explicit.
- A handoff owner and completion condition exist.
Build the operating layer around your agent.
Use the free Spacebrain workspace to keep contact context, handoffs, tasks, automation, and reporting together.