From Cloud Native to Agentic Applications

for more info check: https://www.salaboy.com

Transcript

1. From Cloud Native to Agentic Applications 18-19 May 2026 Savoia Hotel Regency, Bologna FOLLOW US CLOUDNATIVEDAYSITALY.ORG
2. Ecosystem Engineer Dash0 Mauricio Salatino @salaboy.com @salaboy linkedin.com/in/salaboy ABOUT ME FOLLOW US CLOUDNATIVEDAYSITALY.ORG
3. - Agents - MCP (Tools) - A2A (Other Agents) - Skills (Context) New Landscape FOLLOW US CLOUDNATIVEDAYSITALY.ORG - Infra - GPUs to run LLMs - AI Gateways - KV Caches
4. Infra
5. Infra
6. Infra
7. What do we need to know about them as platform engineers? But Agents…? FOLLOW US CLOUDNATIVEDAYSITALY.ORG
8. Architecture FOLLOW US CLOUDNATIVEDAYSITALY.ORG
9. Business Agents vs Coding Agents Disclaimer FOLLOW US CLOUDNATIVEDAYSITALY.ORG
10. - Anthropic Nov 2024 -> Now AAIF - Client/Server architecture (LLM is the client) - Provide tool definitions that LLMs can call when needed Model Context Protocol FOLLOW US CLOUDNATIVEDAYSITALY.ORG
11. FOLLOW US CLOUDNATIVEDAYSITALY.ORG
12. - Google - April 2025 - Agent to Agent discovery and communication - Modality agnostic (Text, Audio, Video) - Enterprise focus - security, long running, built on standards - Complements MCP, it doesn’t compete Agent to Agent (A2A) FOLLOW US CLOUDNATIVEDAYSITALY.ORG
13. - Anthropic 15th Dec 2025 - https://agentskills.io/ Skills FOLLOW US
14. Key Takeaways <LINK>
15. #Demo
16. Pizza Order Agent LLM MCP Server APIs Architecture??
17. https://github.com/salaboy/pizza-vibe
18. Understanding Agentic Apps FOLLOW US CLOUDNATIVEDAYSITALY.ORG
19. FOLLOW US CLOUDNATIVEDAYSITALY.ORG
20. FOLLOW US CLOUDNATIVEDAYSITALY.ORG Workflow as Code
21. FOLLOW US CLOUDNATIVEDAYSITALY.ORG Durable Orchestrations
22. Multi-Agent Orchestration • Complex orchestrations involve multiple agents performing business operations • Enable your orchestrations to be resilient to agent and infrastructure failure • You have an out-of-the-box audit trail to trace back why an agent made or not a given decision or why it called a specific tool
23. #Demo
24. FOLLOW US CLOUDNATIVEDAYSITALY.ORG
25. FOLLOW US of respondents using OpenTelemetry in production. 49% of respondents evaluating OpenTelemetry. 26%
26. OpenTelemetry Collector Telemetry Backends Analysis Tools OTel API & SDK auto-instrumentation … Instrumentation Kubernetes … Infrastructure The OpenTelemetry Collector Receive Process Export Time-series database Log database Trace database … Generate and Emit transmit Collect, Convert, Process, Route, Export transmit Store & Analyze
27. OpenTelemetry Collector … … OTel API & SDK auto-instrumentation … Kubernetes … Infrastructure The OpenTelemetry Collector Receive Process Export Time-series database Log database Trace database … Generate and Emit transmit Collect, Convert, Process, Route, Export transmit Store & Analyze Collection of Telemetry is standardized Vendor space “The last observability agent you will ever install” … and many more.
28. OpenTelemetry Operator … … Instrumentation OpenTelemetry Operator Instructs how to inject auto-instrumentation Injects instrumentation in to the pod
29. Why are these apps so hard to observe? Traditional tracing assumes linear request/response flows 1 Context propagation often breaks between agents, tools, and model calls 2 Agent decisions create dynamic, non-deterministic workflows 3 Execution spans many systems (LLMs, APIs, MCP servers, skills) 4 Key context lives outside the runtime (prompts, reasoning) 5
30. Order pizza trace Numbers: ~275 spans ~48 seconds 10 services/agents Trace includes: • LLM reasoning loops • MCP tool calls • Workflow orchestration • Polling loops • Service calls
31. More Telemetry != Better Telemetry Observation: 275+ spans per order But, • many spans represent infrastructure • some represent retries or polling (it can create a lot of noise) • some represent framework calls (chatty MCP / A2A) Goal: meaningful spans, not just more spans
32. Agent Reasoning(GenAI SemConv) Example span from the trace: completion claude-haiku-4-5 Attributes: gen_ai.request.model = claude-haiku-4-5 gen_ai.prompt = "You are a pizza cooking agent..." gen_ai.completion = "I'll cook a Pepperoni pizza..." gen_ai.usage.prompt_tokens = 1693 gen_ai.usage.completion_tokens = 71
33. Agent Reasoning(GenAI SemConv)
34. What about skills? Skills are often: • shell or python scripts • external tools • subprocesses These components: • are not instrumented • do not propagate trace context • break traces Java Agent TRACEPARENT env Shell script curl with trace header Service span
35. Trace from Skills Execution
36. Key Takeaways • Observing simple agentic applications is quite hard • We have GenAI Semantic Conventions which are starting to align • We need more conventions, from workflows to skills and everything in between • Telemetry is becoming more important, as it can provide context for our agents to learn • So….
37. Maturity Model for OpenTelemetry Support • A shared framework for evaluating OpenTelemetry support • Inspired by the CNCF Platform Engineering maturity model • Descriptive, not prescriptive • Focused on evolution, not scoring Github Issue: https://github.com/open-telemetry/community/issues/3247 Project Proposal: https://github.com/open-telemetry/community/pull/3435
38. THANK YOU!