Day 17: Agent User Messages - The Art of Agent Communication
Learn best practices for crafting user messages that elicit optimal agent responses, enable complex task completion, and establish effective human-agent collaboration patterns.
Day 17 challenge
Goal: master user message engineering for optimal agent collaborationTheme: context engineering week - communication optimizationTime investment: ~20 minutes
Welcome to Day 17! Yesterday you mastered system prompts that guide agent
behavior. Today you’ll learn the other half of the conversation: user message
engineering. You’ll discover how to craft requests that elicit optimal agent
responses and enable complex task completion.Great user messages are the difference between agents that frustrate you and
agents that feel like the perfect colleague.
Learn communication patterns that maximize agent capability
Master techniques for complex task delegation
Develop strategies for iterative problem-solving with agents
Build repeatable communication frameworks for consistent results
This builds on Day 16’s prompt engineering foundation. You’ll be practicing
with the agents you’ve optimized to see how message quality affects outcomes.
"Prepare for tomorrow's client meeting at 2 PM: research their recent announcements,update the project proposal based on our last conversation, and create a follow-up task list based on likely outcomes"
Communication principle Agents perform better when they understand not
just what to do, but why they’re doing it and how it fits into larger goals.
Good: "Research our top 3 competitors' pricing strategies, focusing on their enterprise tiers. I need this for our pricing review meeting next week - include key differentiators and market positioning."Poor: "Look up competitor pricing"
Action-taking requests:
Copy
Ask AI
Good: "Create a Slack message for the #engineering channel announcing our API v2.0 release.Include the key improvements (performance, new endpoints, breaking changes) and link to the migration guide. Schedule it for 9 AM tomorrow."Poor: "Post about the API release"
Analysis requests:
Copy
Ask AI
Good: "Analyze last quarter's sales pipeline data from HubSpot.I want to understand our conversion rates by source, identify which stages have the biggest drop-offs, and recommend 3 specific improvements for next quarter."Poor: "Look at sales data"
1. Initial request: "Help me prepare for tomorrow's board meeting"2. Add context: "Help me prepare for tomorrow's board meeting where we're discussing Q4 performance and 2025 strategy"3. Add specifics: "Help me prepare for tomorrow's board meeting on Q4 performance and 2025 strategy.I need: current metrics summary, comparison to goals, key wins/challenges, and our top 3 strategic priorities for next year"4. Add constraints: "Help me prepare for tomorrow's board meeting on Q4 performance and 2025 strategy.I need: current metrics summary, comparison to goals, key wins/challenges, and our top 3 strategic priorities for next year.Keep everything concise - the deck should be 10 slides max, and focus on actionable insights rather than raw data."
Instead of: “Plan our product launch”Try this sequence:
Copy
Ask AI
1. "Research successful product launches in our space over the last year - what tactics, timelines, and channels did they use?"2. "Based on that research and our product features, create a high-level launch timeline with key milestones"3. "For each milestone, identify what assets we'll need, who's responsible, and what success looks like"4. "Create a detailed project plan for the first 4 weeks with specific tasks and deadlines"
"I need to create a customer onboarding email sequence.First, analyze our current customer feedback to identify the top 3 onboarding pain points.Then, draft email templates that address each pain point.Finally, suggest an automation workflow in HubSpot to deliver these emails based on user behavior."
"Create a competitive analysis report.After you gather the initial research, summarize your findings and ask if I want you to dive deeper into any specific areas before proceeding to the full report."
Complex task strategy Think of agents as highly capable interns. Give them
meaningful work, clear context, and check in at key decision points.
1. "What are the main approaches to improving our customer retention?"2. "Focus on the top 3 approaches - give me specific tactics for each"3. "For the email engagement approach, create a detailed implementation plan"4. "Draft the first email in that sequence targeting customers who haven't used our key features"
"That analysis is helpful, but I need more focus on the financial impact.Can you redo the recommendations section with specific revenue/cost projections for each suggestion?"
"I appreciate the comprehensive research, but let's refocus on the originalquestion about pricing strategy. Can you distill those findings into 3 actionable pricing adjustments we could implement this quarter?"
Research [topic] for [purpose/context].Focus on [specific aspects].I need this by [deadline] for [intended use].Format as [output type] and include [specific elements].
Analysis requests:
Copy
Ask AI
Analyze [data source] to answer [specific question].Consider [key factors] and provide [output format].Include recommendations prioritized by [criteria].
Creation requests:
Copy
Ask AI
Create [deliverable] for [audience/purpose].Use [style/tone] and include [required elements].Format should be [specifications] and optimized for [intended use].
"Run my weekly [department] review: check progress on active projects, identify any blockers, summarize key metrics vs. goals,and flag anything needing my attention this week."
Meeting preparation:
Copy
Ask AI
"Prepare me for [meeting type] with [attendees] about [topic]. Include: recent context about [relevant area], talking points for [specific agenda items], and potential questions I might face."
Project updates:
Copy
Ask AI
"Create a project status update for [project] covering progress since [last update],current status vs. timeline, upcoming milestones, and any risks or blockers."
"Building on the competitive analysis we discussed yesterday,now create a positioning strategy that addresses the gaps we identified in our enterprise messaging."
"Draft this announcement from the perspective of our customer success team - they'll need to field questions about the changes,so include likely concerns and talking points for addressing them."
"Given that we only have 2 weeks and a $5K budget,what's the highest-impact marketing campaign we could run for this product launch?Focus on tactics that leverage our existing assets and team expertise."
In 20 minutes, you’ve mastered user message engineering:Communication frameworks: learned structured approaches to agent requestsTask decomposition: developed strategies for breaking complex work into
manageable piecesIterative refinement: mastered techniques for progressive problem-solving
with agentsQuality optimization: built methods for getting better outputs through
better inputsRepeatable patterns: created templates for common interaction types
Well-crafted user messages unlock agent potential:Before optimization Vague requests leading to generic responses and multiple
rounds of clarificationAfter optimization Precise requests that elicit targeted, actionable
responses on the first tryCombined with yesterday’s prompt engineering, you now control both sides of the
human-agent conversation.
Practice the spiral approach with one of your agents:
Copy
Ask AI
"Let's practice iterative problem-solving.Start by giving me a high-level overview of [complex topic relevant to your work].Then we'll drill down into specifics based on what I find most interesting or useful."
This builds your intuition for guiding agents through complex reasoning
processes.Time to complete: ~20 minutesSkills learned: message structure optimization, task decomposition
strategies, iterative problem-solving, communication frameworks, advanced
delegation techniquesNext: day 18 - Retrieval systems with PostgreSQL and Supabase
Remember: agents are mirrors of the communication they receive. Invest in
crafting better requests, and you’ll get exponentially better results.