From Platform Engineering to AI Automation: Building Infrastructure for Agent Systems

Originally presented at WWCode Summit.

What does a platform engineer need to know to support developers building AI agents? From the initial step of data consolidation and processing to deploying and maintaining infrastructure to run AI agents, this session explores the important infrastructure components to support AI automation.

You'll learn how to generate production-ready, secure infrastructure-as-code using open-source AI automation tools, including Ollama, Granite, Docling, Langflow, and OpenSearch. With this setup, you can generate new code to deploy cloud infrastructure. We'll explore the basics of prompt and context engineering, MCP servers, retrieval augmented generation, and more to help make code generation more specific and relevant to your purposes.

Transcript

1. From Platform Engineering to AI Automation: Building Infrastructure for Agent Systems WWCode Summit | © 2026 IBM Corporation
2. # whoami.md Name: Rosemary Wang Role: Developer Advocate Company: IBM Website: joatmon08.com WWCode Summit | © 2026 IBM Corporation
3. Workshop Resources github.com/joatmon08/ infrastructure-agent WWCode Summit | © 2026 IBM Corporation
4. WWCode Summit | © 2026 IBM Corporation
5. WWCode Summit | © 2026 IBM Corporation MODELS
6. WWCode Summit | © 2026 IBM Corporation MODELS Neural network trained on text for natural language processing tasks
7. WWCode Summit | © 2026 IBM Corporation
8. Models WWCode Summit | © 2026 IBM Corporation Large Language Models (100B+ parameters) Small Language Models (1M-10B parameters) GPT Gemini Claude Llama Copilot Granite Llama Granite SmolLM
9. Models for Platform Engineers WWCode Summit | © 2026 IBM Corporation Large Language Models (100B+ parameters) Small Language Models (1M-10B parameters) GPT Gemini Claude Llama Copilot Granite Llama Granite SmolLM APIs Workload Scheduling
10. Models for Platform Engineers WWCode Summit | © 2026 IBM Corporation Large Language Models (100B+ parameters) GPT Gemini Claude Llama Copilot Granite APIs AuthN/AuthZ Response Caching
11. Models for Platform Engineers WWCode Summit | © 2026 IBM Corporation Large Language Models (100B+ parameters) GPT Gemini Claude Llama Copilot Granite APIs AuthN/AuthZ Response Caching
12. WWCode Summit | © 2026 IBM Corporation Small Language Models (1M-10B parameters) Llama Granite SmolLM Workload Scheduling Compute Storage Models for Platform Engineers
13. Models WWCode Summit | © 2026 IBM Corporation Small Language Models (1M-10B parameters) Llama Granite SmolLM Workload Scheduling Compute Storage Models for Platform Engineers
14. ollama.tf env { name = "OLLAMA_CONTEXT_LENGTH" value = "131072" } volume_mount { name = "ollama-data" mount_path = "/root/.ollama" } resources { requests = { memory = "10Gi" cpu = "3500m" } limits = { "nvidia.com/gpu" = "1" } } WWCode Summit | © 2026 IBM Corporation
15. WWCode Summit | © 2026 IBM Corporation CONTEXT
16. WWCode Summit | © 2026 IBM Corporation CONTEXT The input to a GenAI model
17. Context Tokens WWCode Summit | © 2026 IBM Corporation What ‘s my name ?
18. Context Tokens WWCode Summit | © 2026 IBM Corporation What ‘s my name ? name my ‘s What ?
19. Context Tokens WWCode Summit | © 2026 IBM Corporation What ‘s my name ? name my ‘s What ?
20. Context Tokens WWCode Summit | © 2026 IBM Corporation What ‘s my name ? name my ‘s What ? I’m Rosem ary. Rosem ary. I’m
21. Context Cache WWCode Summit | © 2026 IBM Corporation What ‘s my name ? Rose mary name my ‘s What ? I’m Rosem ary. Rosem ary. I’m Rose mary
22. Context Cache WWCode Summit | © 2026 IBM Corporation What ‘s my name I’m Gabe. ? name my ‘s What ? Gabe. I’m
23. Chat Context WWCode Summit | © 2026 IBM Corporation User: What’s my name User: What’s my name AI: I don’t know User: Hi there AI: I don’t know AI: Hello! User: What’s my name AI: I don’t know User: Hi there AI: Hello! User: Tell me a story
24. Context Patterns WWCode Summit | © 2026 IBM Corporation Prompting Knowledge Action Memory Length
25. WWCode Summit | © 2026 IBM Corporation
26. Context Patterns WWCode Summit | © 2026 IBM Corporation Prompting What’s my name?
27. Context Patterns WWCode Summit | © 2026 IBM Corporation Prompting Knowledge What’s my name? I’m Rosemary.
28. Context Patterns WWCode Summit | © 2026 IBM Corporation Prompting Memory What’s my name? My name is Rosemary. <recall> Rosemary <remember>
29. Context Patterns WWCode Summit | © 2026 IBM Corporation Prompting What’s my name? My name is Rosemary. Length Tell me a story Hi Rosemary! Once upon a time... ... Max Context Length Key Information
30. Context Patterns WWCode Summit | © 2026 IBM Corporation Prompting What’s my name? My name is Rosemary. Length Tell me a story Hi Rosemary! Once upon a time... ... Max Context Length Key Information
31. Context Patterns WWCode Summit | © 2026 IBM Corporation Prompting Knowledge Action Memory Length Prompt Engineering MCP Prompts Attachments RAG MCP Resources Tool Calling MCP Tools Skills Scratch on Disk R/W RAG Observational Truncation Compression Sub-Agents
32. Context Patterns WWCode Summit | © 2026 IBM Corporation Prompting Action What’s my name? <look up name> Rosemary Tool: name_lookup
33. Context Patterns for Platform Engineers WWCode Summit | © 2026 IBM Corporation Prompting Knowledge Action Memory Length Prompt Engineering MCP Prompts Attachments RAG MCP Resources Tool Calling MCP Tools Skills Scratch on Disk R/W RAG Observational Truncation Compression Sub-Agents
34. Prompting WWCode Summit | © 2026 IBM Corporation
35. Prompting WWCode Summit | © 2026 IBM Corporation
36. WWCode Summit | © 2026 IBM Corporation
37. Knowledge – Attachments 36 NY Tech Week | © 2026 IBM Corporation User: What is my company’s infrastructure as code tool called? Document terraform.md: # Terraform Terraform, maintained by IBM, is an infrastructure as code tool that lets you define both cloud and on-prem resources in human-readable configuration files that you can version, reuse, and share. Document me.txt: My name is Rosemary and I work for IBM. I bridge the technical and cultural barriers between infrastructure, security, and application development as platform engineer, writer, and speaker. ...
38. Knowledge – Attachments WWCode Summit | © 2026 IBM Corporation
39. Knowledge – Attachments WWCode Summit | © 2026 IBM Corporation
40. Knowledge – Retrieval Augmented Generation (RAG) WWCode Summit | © 2026 IBM Corporation Document me.txt (chunk 1): My name is Rosemary and I work for IBM. Document terraform.md (chunk 3): Terraform, maintained by IBM, is an infrastructure as code tool Answer the question using the following documents: User: What are my company’s SLMs called?
41. Knowledge – Retrieval Augmented Generation (RAG) #!/bin/bash # Wait for OpenSearch to be ready echo "Waiting for OpenSearch to start..." until curl -s http://localhost:9200/_cluster/health > /dev/null; do sleep 2 done echo "OpenSearch is ready. Creating 'langflow' index..." # Create the langflow index with vector search configuration curl -X PUT "http://localhost:9200/langflow" -H 'Content-Type: application/json' -d' { "settings": { "index": { "knn": true, "knn.algo_param.ef_search": 512 } }, "mappings": { "properties": { "chunk_embedding": { "type": "knn_vector", "dimension": 384 } } } } ' echo "" echo "Index 'langflow' created successfully!" WWCode Summit | © 2026 IBM Corporation
42. Knowledge – Retrieval Augmented Generation (RAG) WWCode Summit | © 2026 IBM Corporation
43. Knowledge – Retrieval Augmented Generation (RAG) WWCode Summit | © 2026 IBM Corporation
44. Knowledge – Retrieval Augmented Generation (RAG) WWCode Summit | © 2026 IBM Corporation
45. WWCode Summit | © 2026 IBM Corporation
46. Action – MCP Tools WWCode Summit | © 2026 IBM Corporation What’s my name? @tool def get_name(include_last: bool) -> str: """ ... { "description": "Get the user’s…", "properties": { "include_last": { "title": "Include Last", "type": "boolean" } }, "required": [ "include_last" ], "title": "get_name", "type": "object" } @tool def get_name(include_last: bool) -> str: """ ... @tool def get_name(include_last: bool) -> str: """ ...
47. Action – MCP Tools WWCode Summit | © 2026 IBM Corporation What’s my name? @tool def get_name(include_last: bool) -> str: """ ... { "description": "Get the user’s…", "properties": { "include_last": { "title": "Include Last", "type": "boolean" } }, "required": [ "include_last" ], "title": "get_name", "type": "object" } @tool def get_name(include_last: bool) -> str: """ ... @tool def get_name(include_last: bool) -> str: """ ... <tool_call>{ "name": "get_name", "arguments": {"include_last": true} }
48. Action – MCP Tools WWCode Summit | © 2026 IBM Corporation What’s my name? @tool def get_name(include_last: bool) -> str: """ ... { "description": "Get the user’s…", "properties": { "include_last": { "title": "Include Last", "type": "boolean" } }, "required": [ "include_last" ], "title": "get_name", "type": "object" } @tool def get_name(include_last: bool) -> str: """ ... @tool def get_name(include_last: bool) -> str: """ ... <tool_call>{ "name": "get_name", "arguments": {"include_last": true} } <tool>Rosemary Wang
49. Action – MCP Tools WWCode Summit | © 2026 IBM Corporation
50. Action – Skills WWCode Summit | © 2026 IBM Corporation How should I build my vector database? Tool: bash System: You have a set of skills available in the skills directory. skills/ ├── architect │ └── SKILL.md └── engineer ├── scripts │ └── iac_test.py └── SKILL.md 4 directories, 3 files ?>
51. Action – Skills WWCode Summit | © 2026 IBM Corporation How should I build my vector database? Tool: bash System: You have a set of skills available in the skills directory. skills/ ├── architect │ └── SKILL.md └── engineer ├── scripts │ └── iac_test.py └── SKILL.md 4 directories, 3 files ?> for skill_dir in $(find . -name SKILL.md); do echo $skill_dir | xargs dirname | xargs basename; done architect engineer ?> cat skills/engineer/SKILL.md --- name: engineer <tool_call>{ "name": "bash", "arguments": {"command": "for skill_dir in $(..."} } <tool>architect engineer
52. Action – Skills WWCode Summit | © 2026 IBM Corporation How should I build my vector database? Tool: bash System: You have a set of skills available in the skills directory. skills/ ├── architect │ └── SKILL.md └── engineer ├── scripts │ └── iac_test.py └── SKILL.md 4 directories, 3 files ?> for skill_dir in $(find . -name SKILL.md); do echo $skill_dir | xargs dirname | xargs basename; done architect engineer ?> cat skills/engineer/SKILL.md --- name: engineer <tool_call>{ "name": "bash", "arguments": {"command": "for skill_dir in $(..."} } <tool>architect engineer <tool_call>{ "name": "bash", "arguments": {"command": "cat skills/engineer/..."} } <tool>--- name: engineer
53. Action – Skills # Terraform Example Generator Agent ## Overview You are an agent that generates HashiCorp Configuration Language (HCL) code for Terraform configurations, following best practices and security standards. ## Capability Statement When invoked, the agent will: 1. Analyze infrastructure requirements and constraints 2. Query Terraform Registry for latest provider versions and module patterns 3. Generate idiomatic HCL with proper resource dependencies 4. Apply security best practices and compliance standards ## Prerequisites - Understanding of target infrastructure provider (AWS, Azure, GCP, etc.) - Access to Terraform Registry APIs - Knowledge of organizational naming conventions - Awareness of security compliance requirements (CIS, SOC2, etc.) ## Execution Steps ### 1. Provider Discovery ```markdown - Use `get_latest_provider_version` to fetch current provider release - Call `get_provider_capabilities` to understand available resources - Retrieve `get_provider_details` for specific resource documentation ``` WWCode Summit | © 2026 IBM Corporation
54. Action – Agent AuthN/AuthZ WWCode Summit | © 2026 IBM Corporation Identity Provider (Assignment & Attestation) Policy Enforcement Engine (Service Mesh? Gateway?) Deterministic Policy or Predictive Index? Delegation Chain (On-Behalf-Of) Coding Agent Architecture Agent Security Review Agent CMDB Check Agent
55. WWCode Summit | © 2026 IBM Corporation
56. Memory – Observational WWCode Summit | © 2026 IBM Corporation My name is Rosemary. Don’t forget it! AI: Hi Rosemary! I work for IBM Who do I work for? Memory: name is Rosemary AI: Hi Rosemary, thanks for letting me know Memory: name is Rosemary Memory: works for IBM AI: Hi Rosemary, you work for IBM
57. Length – Sub-Agents WWCode Summit | © 2026 IBM Corporation My name is Rosemary. Don’t forget it! Tool: remember Tool: recall <tool_call>{ "name": "remember", "arguments": {"content": "name: Rosemary"} } <tool>success AI: Hi Rosemary! I work for IBM <tool_call>{ "name": "remember", "arguments": {"content": "employer: IBM"} } <tool>success Tool: remember Tool: recall You are an assistant that remembers and recalls facts. Tool: remember Tool: recall You are an assistant that remembers and recalls facts.
58. WWCode Summit | © 2026 IBM Corporation
59. Learn more 1 2 3 github.com/joatmon08/ infrastructure-agent hashi.co/vault-a2a-obo langflow.org/ WWCode Summit | © 2026 IBM Corporation
60. Thank you! Rosemary Wang | joatmon08.com