OpenAI Python SDK 3.17.0: a Breaking Change and Six New Features
OpenAI's Python SDK 3.17.0 ships a breaking change to local shell tool output plus six new features. Here's what Python teams need to know before upgrading.
‹ Back to Blog
AI
Can we live dangerously? Sandboxing Claude, and the Claude foreman that runs the rest
How Scout runs a pen full of Claude agents behind one foreman, inside a sandbox with zero blast radius, credentials and all.
Prompts, skills, and the AGENTS.md nobody wants to write (and how Anthropic writes theirs)
The model isn't the only lever, and it's rarely the one you control. What you feed it is. A look at how Claude actually uses prompts, skills, and AGENTS.md, how to write each one well, and how we wire all three into Corral.
Monitoring AI Applications in 2026: What You Actually Need
AI applications need two layers of monitoring: LLM observability for output quality and token costs, and traditional APM for the application code underneath. A practical guide to what tools cover which layer.
Your AI Coding Agent Is Flying Blind in Production
Your AI coding agent can read your codebase but not your production data. MCP changes that. Here is how to connect monitoring to your development workflow.
You Are Building With AI. Who Is Watching What It Ships?
AI makes building faster. It also makes production problems faster. If you ship without monitoring, you are flying blind at 10x speed.
Stop Starting Your Day in a Stack Trace
A workflow that connects Scout's MCP server to an AI assistant to automatically triage production errors overnight and open draft PRs with proposed fixes — so your team starts the day reviewing solutions, not stack traces.
Getting Scout Data Into Your AI Workflow
We've shipped API updates, two MCP server options, and a new CLI — all designed to get Scout's performance data to the AI tools you already use.
Adding FastMCP support to the Scout Python Agent
Scout Senior Developer Quinn walks us through adding FastMCP instrumentation to our Python agent. The Scout MCP server enables Scout users to query their Scout agent about application performance so that the agent can make decisions and fix issues based on that data.