Understanding where agents excel
Understanding where agents excel
5 min
Alban DumouillaWhat you've learned so far:
- Chapter 1: What AI agents are and how LLMs work
- Chapter 2: How agents work (instructions, knowledge, tools)
What you'll learn now: Where AI agents create real value in your daily work
The gap between "AI agents are interesting" and "AI agents are transforming how I work" comes down to one thing: recognizing the right opportunities. This chapter will help you spot them.
1. What makes a great agent opportunity?
AI agents deliver the most value when your task has these five characteristics:
1.1. Knowledge-intensive work
The task requires pulling together information from multiple sources and making sense of it all.
Example: Creating a quarterly business review that combines sales data, customer feedback, and market trends
1.2. Clear inputs and outputs
You can describe what goes in and what should come out, even if the specific content varies each time.
Example: "Take meeting notes → produce action items and decisions"
1.3. Measurable value
The agent either:
- Saves significant time, OR
- Improves quality noticeably, OR
- Makes something feasible that wasn't before
Example: Reducing RFP response time from 3 days to 3 hours
1.4. Structured process
There's a recognizable pattern or format, even if the details change.
Example: Contract reviews follow similar steps; content varies but the process doesn't
1.5. Low to medium stakes
Mistakes are catchable. A human reviews the output before it matters.
Example: Draft emails (reviewed before sending) vs. legal contracts (high-stakes, need extreme accuracy)
⚠️ What's NOT required: High frequency. That's a common misconception: You might think agents are only worth it for daily tasks. Not true.
The math of low-frequency, high-value tasks:
Task Type | Frequency | Time Saved Each Use | Annual Value |
Daily Slack summaries | 260x/year | 10 minutes | 43 hours |
Quarterly business reviews | 4x/year | 10 hours | 40 hours |
Annual planning | 1x/year | 24 hours | 24 hours |
That quarterly planning agent used just 4 times a year delivers as much value as a 10-minute daily task. Don't dismiss low-frequency opportunities just because they're infrequent.
2. High-value examples
Quarterly Business Reviews
- Agent compiles metrics, trends, and insights across systems
- Happens once per quarter
- Saves 8+ hours each time
Annual strategic planning
- Agent synthesizes previous year's performance + market research + competitive intelligence
- Happens once per year
- Saves days of preparation work
Board meeting preparation
- Agent creates comprehensive briefing books from scattered data
- Monthly occurrence
- Saves 10+ hours each time
Major client presentations
- Agent researches client, reviews past interactions, drafts customized pitch
- As-needed basis
- Saves 4-6 hours per presentation
Performance review preparation
- Agent summarizes employee contributions from Slack, emails, project updates
- Twice yearly per employee
- Saves hours per review
3. What agents DON'T replace
Knowing where agents fall short is equally important:
What agents struggle with | Why | Who should handle it |
High-stakes decisions | Errors are costly (legal, medical, financial) | Human experts with accountability |
Novel situations | No precedent or pattern to learn from | Human creativity and reasoning |
Creative strategy | Needs vision, taste, original thinking | Human strategists |
Ethical judgment | Nuanced trade-offs requiring human values | Humans with context |
💡 Remember: The best agent implementations amplify human expertise rather than replace it. A sales agent doesn't replace the salesperson—it frees them to spend more time building relationships instead of drafting emails.
🎯 Try it out: map your own work
Before exploring specific use cases, take 5 minutes to audit your work.
Step 1: List your tasks
Write down 5-10 tasks you do regularly (or infrequently but importantly).
Step 2: Score each task
Rate each on a scale of 1-5:
- Value impact: How much time/quality improvement? (5 = very high, 1 = minimal)
- Complexity: Does it require searching/synthesizing lots of info? (5 = very complex, 1 = simple)
- Predictability: Are inputs/outputs structured? (5 = very predictable, 1 = chaotic)
- Low stakes: What if it's wrong? (5 = no big deal, 1 = catastrophic)
Step 3: Calculate total score
Add the four numbers. Tasks scoring 15+ are prime agent candidates.
Example Scorecard:
Task | Value | Complexity | Predictability | Low Stakes | Total | Agent Fit? |
Daily meeting follow-ups | 3 | 3 | 4 | 4 | 14 | Maybe |
Quarterly business review prep | 5 | 5 | 4 | 3 | 17 | ✅ Yes |
Summarizing Slack updates | 4 | 4 | 4 | 5 | 17 | ✅ Yes |
Negotiating major contracts | 2 | 3 | 1 | 1 | 7 | ❌ No |
Annual strategic planning research | 5 | 5 | 3 | 3 | 16 | ✅ Yes |
Conclusion
Aaaand that's it! You now have a strong foundation of knowledge about AI. Let's put it into practice! Move on to the next course to take your first steps on Dust.
Test Your Knowledge
Answer 5 questions — score 3 or more to pass
Ready to test your understanding of this chapter?