Getting Started with Bits Agent Builder: Real-World Use Cases

Transcript

1. Getting Started with Bits Agent Builder : Real-World Use Cases 2026/08/10@JDDUG Sapporo Hisatomo Futahashi (@futahashi),Nulab Inc.
2. Hisatomo Futahashi • Principal Engineer / AWS Alliance Lead @ Nulab Inc. • Technical Advisor @ Horizon Technology Inc. • 2026 Datadog Ambassador • 2022 & 2023 APN AWS Top Engineers (Software) • 2023 Japan AWS All Certifications Engineers @futahashi 2
4. Agenda • Introduction • What Is Bits Agent Builder? • Bits Agent Builder Basics • Real-World Use Cases • Common Pitfalls • What We Gained • Wrap-Up 4
5. Introduction 5
6. Why Do We Need AI Agents? • From instruction-driven to autonomous execution ◦ Give it a goal, and it makes context-aware decisions and sees the task through. • Less manual work, less build effort ◦ Operates multiple tools and runs multi-step processes without programming. • Free humans from routine work ◦ Delegate complex, repetitive tasks to AI so people can spend time on higher-value work. Evolving AI from a tool into an autonomous worker 6
7. What Is Bits Agent Builder? 7
8. What Is Bits Agent Builder? • Custom AI agents that run inside Datadog Leverages the rich data already in Datadog ①Rich context Logs / Metrics / Traces / Signals / Events and more ~3,000 actions, ~50 services, and 4 MCP servers ②Broad tool support Datadog / AWS / Kubernetes / GitHub / Slack and more ~30 entry points across Datadog and external services ③Flexible triggers Workflow / Monitor / Incident / GitHub / Slack and more Agents that make the most of a unified observability platform 8
9. Bits Agent Builder Basics 9
10. How to Create an Agent • 3 ways to get started ①Create from a prompt ②Choose from Blueprints ③Build from scratch 10
11. Building Blocks ①Instructions Write the instructions for your agent e.g., role, procedures, constraints Feed lessons from each run back into the instructions ̶ the agent improves iteratively 11
12. Building Blocks ②Tools The tools your agent can use e.g., Datadog / AWS / Kubernetes and more ~3,000 actions, ~50 services, 4 MCP servers Knowing the Tools sharpens your sense of what you can delegate 12
13. Building Blocks ③Automation Set the conditions that trigger your agent e.g., Schedule, Monitor, Incident Get work done without manual intervention 13
14. Building Blocks ④Chat A conversational interface with your agent e.g., run, debug, ask questions Start here to test and refine your agent 14
15. Real-World Use Cases 15
16. ①AWS Commitment Monitoring and Recommendations Recommendations informed by org policy and actual usage Unified monitoring across organizations and services Never miss the few big purchases a year ̶ recommendations grounded in data and org policy 16
17. ②Datadog Cost Trend Analysis Analysis that goes beyond the invoice Cuts the burden of routine monitoring and trend investigations 17
18. ②AWS Cost Trend Analysis Pinpoints causes down to the ID / ARN level Analysis enriched with events and logs Deep insights powered by telemetry data 18
19. ③Incident Scribe Documentation that follows our in-house format Accurate records and summaries based on the actual conversation Structures incident communication and keeps it flowing 19
20. (Side Note) Working with Bits Investigation Reached the right conclusion in 12 minutes Can also execute Actions and Workflows AI drives the investigation ̶ humans focus on directing it 20
21. ④Security Signal Triage (Auto-Archive) Rationale and supporting context included Archives signals appropriately, with reasons attached Instant investigation and decisions on security alerts 21
22. ④Security Signal Triage (Escalation to Humans) Creates a Case and assigns it to a person What the agent can't decide goes to a human 22
23. ⑤WAF Log Analysis Takes over massive log analysis Can eventually connect to real-time response Log analysis and improvement proposals that make WAF operations self-sustaining 23
24. Common Pitfalls 24
25. Vague Instructions Lead to Wrong Procedures 🙋 🙅 Analyze our Datadog costs... you know, nicely. Got it ̶ I'll take service usage and unit prices and calculate everything through sheer willpower!!! There are cost metrics for that! 🤖 Check the evidence, and correct mistakes with follow-up instructions 25
26. Context Overflow Analyze ALL costs of ALL services in ALL AWS accounts and give me ALL the insights. 🙋 ALL of it!!! Through sheer willpower!!! ALL!!! *grinding noises* ... 😇 🤖 Design for it ̶ split the task and narrow the output scope 26
27. What We Gained 27
28. What We Gained • ①Work starts moving proactively • ②Operations get put into words • ③Start small, grow big 28
29. Outcome ①Work Starts Moving Proactively Task Task Human AI Instructions Result AI Result Human Agents get ahead of the work and carry it out 29
30. Outcome ②Operations Get Put into Words methods methods constraints triggers triggers exceptions exceptions 🧠 constraints Tacit knowledge and perspectives take shape as words 30
31. Outcome ③Start Small, Grow Big Research & proposals Autonomy & execution Decent quality High quality Single domain Multiple domains Single agent Agent collaboration Rich context Broad tools Flexible triggers Speed, depth, and breadth ̶ only possible on a unified platform 31
32. Wrap-Up 32
33. Wrap-Up • Agents can autonomously handle complex, recurring work • Bits Agent Builder makes the most of the unified observability platform • Ready to use right away, easy to grow • Building them right takes AI literacy and deep domain expertise The possibilities of Bits Agent Builder are endless! Put the unified platform to work for your organization! 33
34. The End Thank you for your attention. 34