Skip to main content
Agent Chat plans and executes tasks on its own. You don’t need to break your request into explicit steps. Instead, focus on describing your goal clearly and let the agent figure out how to get there. That said, how you describe your goal matters. The patterns below help you get better, more consistent results.

​
Describe the goal, not the steps

Instead of walking the agent through each action, tell it what you want to end up with.
Instead of thisTry this
”First read the document, then identify the risky clauses, then redline them""Review this contract and redline the high-risk provisions"
"Search the web for data breach notification standards, then compare them to section 8""Are our data breach notification terms consistent with market standards?"
"Read our playbook, then check each clause against it""Review this contract against our SaaS playbook”
The agent already reads, plans, and executes in sequence. Describing the end state lets it choose the most effective path.

​
Include the context that matters

The more relevant context you provide up front, the less the agent needs to ask for along the way. Key details to include:
  • Your role: Vendor, customer, licensor, licensee
  • Document type: MSA, NDA, DPA, employment agreement
  • Jurisdiction: If specific laws or regulations apply
  • Risk tolerance: Light-touch review vs. aggressive negotiation
  • Audience: Who the output is for (internal team, counterparty, executive summary)
Example: 
Review this SaaS vendor agreement as the customer. We're in healthcare,
so flag any HIPAA compliance gaps. Focus on indemnification, liability,
and data security. We have moderate leverage.

​
Common workflows

​
Comprehensive contract review

Give the agent a contract and your standards. It handles the rest. 
Review this vendor agreement against our playbook and company profile.
Redline the provisions that deviate from our standards and add comments
explaining each change.

​
Compliance review

Point the agent at specific regulations or policies. 
Review this DPA for GDPR compliance gaps. Flag any provisions that
don't meet the requirements and suggest compliant alternatives.

​
Template comparison

Upload both documents and ask for the differences. 
Compare this counterparty draft to our standard template. Identify
the material deviations and redline the ones that favor the counterparty.

​
Learning from prior redlines

Attach previous redlined contracts so the agent can apply consistent positions. 
I've attached three prior redlines of similar vendor agreements.
Review how we handled indemnification, liability, and termination
in those contracts, then apply those positions to this one.

​
Negotiation prep

Ask the agent to think from multiple perspectives. 
Review this MSA from the counterparty's perspective. What are their
strongest positions? Then draft responses to the objections they're
most likely to raise.

​
Post-review cleanup

After applying redlines, use a follow-up message to tidy up. 
Proofread the document for any typos, undefined terms, or formatting
inconsistencies introduced during the redlining process.

​
Iterating in conversation

Agent Chat is multi-turn. After the agent delivers its initial results, you can refine in follow-up messages without starting over. Narrow the focus: 
Go deeper on the indemnification section. What are the specific risks
if we accept the current language?
Adjust the approach: 
That liability cap is too aggressive. Revise it to be more balanced
while still protecting us on direct damages.
Change the output: 
Now summarize all the changes we've made in a table I can send to
the business team.
Each follow-up benefits from the full conversation history, so the agent’s responses get more tailored as you go.

​
Agent Chat vs Playbook Checks

​
Use Agent Chat when

  • Dealing with novel or one-off contracts
  • The review requires judgment calls and back-and-forth
  • You need drafting, research, or document generation alongside the review
  • You want to iterate on specific provisions

​
Use Playbook Checks when

  • You have recurring contract types with consistent criteria
  • You want standardized, repeatable pass/fail results
  • Speed is critical for high-volume work

​
Use both together

Run Playbook Checks first to catch standard issues, then switch to Agent Chat to address anything the checks missed or to draft alternative language for flagged provisions.

​
Effective prompting

DoDon’t
Describe the end result you wantMicromanage individual steps
Include your role, jurisdiction, and risk toleranceAssume the agent knows your business context
Specify the output format (redlines, summary, table)Leave the format ambiguous
Iterate in follow-up messages to refine resultsCram everything into one long prompt
Attach relevant documents as referenceReference documents the agent can’t access

​
Troubleshooting

Agent makes changes in the wrong location: Highlight the specific section in Word before prompting. Incomplete redlines: Request explicitly: “Review the entire contract from Section 1 through the end.” Too many style changes: Specify: “Only suggest substantive changes, not stylistic improvements.” Missing company context: Add information to your Company Profile so it’s automatically included.