Vibe Check : There's no agent without the harness

In this episode, Melanie was joined by Elizabeth Fuentes Leone, AI Engineer and Developer Advocate at AWS, and co-host Shubham Londhe from Temporal who is a leader in helping people learn about and build platforms and DevOps. The premise: there is no agent without the harness, the code and infrastructure around the model that turns it into something you can actually run in production. Elizabeth built a harness from scratch with AWS Bedrock AgentCore using a single CLI command, then walked through the pieces that do the work and where durability fits in.

We went beyond the demo and got into:
The harness:
- What a harness actually is: the agent loop (your code) plus all the infrastructure around it that feeds the model and handles its output
- Bedrock vs AgentCore: Bedrock as the umbrella for AWS GenAI services, model invocation through a single API, and how AgentCore sits on top

Building it
- Building a harness from scratch with one AgentCore CLI command, generating about 30 lines
- Swapping model providers (Bedrock, Anthropic, OpenAI, Gemini) by changing a single model ID, and using global model IDs to avoid per-region invocation limits
- CDK for infrastructure as code in Python or TypeScript instead of hand writing CloudFormation YAML
- Deploying to a container on AWS, and what gets built first: roles, policies, the security layer
- Invoking the harness locally with session and actor IDs to test before spending on cloud deploys

What AgentCore gives you
- AgentCore Gateway with semantic search across your MCP tools
- Two kinds of agent memory: session memory built into the runtime, and user memory that persists across sessions in a vector store, with four built-in memory strategies
- Built-in observability: traces, spans, session IDs, token usage, and response latency with nothing extra to configure

Production and durability
- Where Temporal fits: durable execution, state management, retries, and observability around the agent loop, added as a tool inside the harness
- Why Temporal does not replace LangGraph or LlamaIndex: frameworks build the agent, Temporal is the durability layer you put inside the loop
- Elizabeth's dev workflow: sharing skills across Claude and Kiro, and using AWS's MCP server to keep the assistant from hallucinating APIs
- Why knowing agent architecture still matters when an assistant can one shot the code, and the black box that quietly burns tokens when you don't

If you have built an agent that runs fine on your laptop and then fell apart the moment you tried to put it in production, this one is for you.

Resources:
- Bedrock AgentCore: https://aws.amazon.com/bedrock/agentcore/
- AgentCore credits (shared on the stream): https://events.elifuentes.tech/events/agentcore-harness/
- Elizabeth Fuentes Leone: https://elifuentes.tech/
- Shubham Londhe: https://www.trainwithshubham.com/
- Temporal on GitHub: https://github.com/temporalio 
- How Temporal uses Bedrock AgentCore to Create Robust AI Systems:  https://aws.amazon.com/blogs/apn/how-temporal-uses-amazon-bedrock-agentcore-to-create-robust-ai-systems/
---

Temporal is the simple, scalable, open source way to write and run reliable cloud applications. 

Learn more
Blog: https://temporal.io/blog
How Temporal Works: https://temporal.io/how-temporal-works
Community Slack: https://temporal.io/slack

Developer resources
Docs: https://docs.temporal.io
Courses: https://learn.temporal.io/courses
Support forum: https://community.temporal.io