Coder

our team itself providing the developer environments, whether it's through coder.com CDEs, or whether it's the local workstation environments that are always hardened and locked down

Natalie and Jessie taught me a lot about optimizing a website in the AI era. Then Kyle scanned coder.com and found yet different optimizations. What did this tell me? There is a difference between AEO (answer engine optimization) and agent-ready (can an agent use the site). This matters when product lines blur. Let's take Claude as an example. Are you using Claude.ai (the chat), Claude Code, or Claude Cowork? Modality matters. In the case of Claude Code and Claude Cowork it's pretty straight forward. It's acting as an agent. Claude.ai ? Not so clear. It defaults to acting as an answer engine, unless it does a tool call in which it acts as an agent. We increasingly rely on AI for search. Sites need to be AI-ready. But we're clearly in the messy middle where the technology, the best practices, and the tools are all rapidly advancing. Getting it right is the difference between a user finding your product or service versus your competitor's. If you want a more technical breakdown, I wrote about it here: https://lnkd.in/gn4WUviG

"Every project I shipped with vibe coding ended up with API keys in four places, no rotation strategy." That's from a developer who loves AI coding tools describing what happened after the demo phase ended, not a CISO. The vibe coding movement has produced real results. People are shipping faster than ever. But speed without structure creates a specific kind of debt that compounds fast: credentials scattered across workspaces, agents making network calls nobody authorized, code that works in staging but wasn't written with production constraints in mind. Just recently, a Cursor agent deleted a company's production database. It found an over-scoped API token in the filesystem, used it to call a destructive API, and wiped both the database and its backups. The thing is, the agent used every tool available to complete the task. The problem? No one had defined what tools should be available. Development costs are dropping. But operational costs — maintenance, monitoring, governance — are going up when AI-generated code hits production. Most of us are measuring the first number and ignoring the second. Stop asking whether AI coding tools work. They clearly do. The question is whether your infrastructure assumes agents will always behave as intended. Because they won't. And the answer has to exist at the infrastructure layer, before the agent runs — not in a policy doc it will never read. We wrote about this issue on the Coder blog recently if you want to read more: https://lnkd.in/g8rwv9aZ

Out at Red Hat Summit this year? I will be working our booth Tuesday and Wednesday afternoon, drop by and checkout what we're doing with Coder Agents deployed to Red Hat OpenShift!