Operations | Monitoring | ITSM | DevOps | Cloud

Every week there's a new agent sandbox, and they all promise the same thing: a safe place for your agent to run code. In this video we split them into two types, dumb sandboxes and smart sandboxes, so you can sort any provider by what it can actually do instead of how it's marketed. You'll see where today's providers sit, why dumb sandboxes still have a real place, and how to pick the kind your agent's job needs.

XERJ hit number one on GitHub promising to stop coding agents from burning tokens, so I plugged it into a large TypeScript repo I work in every day and the results weren't what I expected. That sent me down a rabbit hole testing nine agent search tools head to head on my own codebase, with exact run counts, token numbers, and setup costs. If you're thinking about adding a search tool to your agent, watch this first and save yourself the install.

I built a small software factory: I file a Linear ticket, and a coding agent spins up in its own cloud sandbox, fixes the bug, pushes a branch and reports back. I can watch it work or take over from my laptop using my normal dev setup. In this video I walk through every piece of it in plain terms, from ticket to pushed branch, and explain why I made each decision so you can build your own.

I scanned 10,705 public GitHub repos for leaked credentials. After removing obvious placeholders, 3,543 still had something that looked like a key, and 1,680 held almost 7,000 credentials in recognizable formats like Stripe keys. If you're new to coding with agents (or you've been a dev for decades), this is the setup I'd want you to have before you ship anything public.

AI-generated code security is the part nobody plans for. Ross Hendrickson, CTO of Inspectiv, on what happens after the AI writes it. He can think of a feature and ship it the same day, generating more code than a week of writing it by hand would have produced. The vulnerabilities don't disappear along with the typing. Something still has to review what was generated, secure it, and get it out the door without becoming the new bottleneck.

I pulled every core-service outage AWS, Azure, and GCP published from January 2025 through September 2026 and measured what each deployment pattern would have saved.

If you're trying to run more than a few AI coding agents at once, the real problem shifts from prompting to managing where they live and what they can reach. This video walks through the five levels of agent management, from the IDE all the way to a fleet running in the cloud with access to your own services.

Kintsugi puts sales tax compliance on autopilot for 7,300 companies selling into 110 countries, and its tax engine sits inside other companies' checkouts. The answer has to arrive before the shopper finishes paying. "Typically, a customer's expectation is that we are returning the sales tax estimate on the invoice within 100 milliseconds." The cost of missing is not abstract: "For every minute that our sales tax API, if it ever goes down, our customers are not able to collect roughly $4 million in sales tax that they should be collecting.".

Matt Zytaruk, VP of Technology at Minga, on what changed once his developers could point their AI tools at their own infrastructure. Minga checks in 1.5 million students across the eastern seaboard every school morning. They run that platform on Control Plane, which gives their team an MCP server, a CLI, and Terraform-native resources — so a developer can investigate an issue and generate the code for the change without waiting on anyone.

"Hey, I built this over the weekend. I want to get it in front of customers." And it always hits architecture, security, and infrastructure. Ross Hendrickson, CTO at Inspectiv, calls that gap the chasm. His team crosses it on Control Plane: AI-written code secured, reviewed, and released in a day. Control Plane combines AWS, GCP, Azure and your own hardware into one virtual cloud shaped to your workloads.