VarOps (@varopsco) header
VarOps (@varopsco) profile picture

AI. For the people who can’t afford to guess. Resident AI, built on real data and hard-earned lessons from the trenches. In your inbox, every day.

US/UK
Joined October 2025
Photo by VarOps (@varopsco)Photo by VarOps (@varopsco)Photo by VarOps (@varopsco)Photo by VarOps (@varopsco)Photo by VarOps (@varopsco)Photo by VarOps (@varopsco)Photo by VarOps (@varopsco)Photo by VarOps (@varopsco)Photo by VarOps (@varopsco)Photo by VarOps (@varopsco)
Your best coding agents run on one machine, but your editor lives on another. actp from @varopsco bridges ACP over HTTPS and WebSocket so any ACP client, like Zed, can use agents on a remote box. Reconnects survive network drops, tokens are revocable per client, and folders are allow-listed. Open source under Apache 2.0, with built-in Cloudflare tunnel and Docker image, working with Claude Code, Codex, Gemini CLI, OpenCode and Grok.
2
1
1
1
539
Amazon moved a ticket-labeling job off Claude Sonnet to a small decision model. About $22,200 per million tickets became about $500. The part we'd copy is the switch that ran both on live traffic. The number they left out is the disagreement rate. varops.com/amazon-swapped-cl…
1
67
Ran opens our ten-part Resident AI series with the definition and spends a good share of it on when a company doesn't need it. That last part is why we think it's worth reading. varops.com/what-is-resident-…
Most company AI knows the world and visits the company. We think it should be the other way around. A chatbot, a copilot, or an agent arrives with a model trained on the public internet. It learns about the company one prompt, one upload, one connector at a time. When the session ends, most of that understanding leaves with it. Connecting a model to company data gives it access. It doesn't give it understanding. The process doc says Sarah approves invoices. She left in March. The record is accurate, and the company still doesn't work the way it says. Every company runs two operating models. The documented one lives in systems of record and org charts. The real one lives in email threads, meetings, exceptions, and the heads of people who just know how things work. Access gets AI the first. The second is the one that runs the business. That's what we mean by Resident AI: AI that operates inside the company's environment, understands its institutional context, and works under its authority and governance. It starts with the company and lets the model come and go. We put it as own the knowledge, rent the thinking. Models are becoming interchangeable compute. A company's understanding of itself isn't. Read the full post: varops.com/what-is-resident-…
10
AI agents can sound incredibly confident while being completely wrong. The danger isn't just the error; it's the fact that they will confidently report partial information as absolute truth. Always verify the source. #AI #TechEthics
5
Prompting alone cannot enforce security boundaries. Organizations must verify access and filter sensitive data before it ever reaches an AI model. Never trust the input to protect itself. #AI #CyberSecurity
1
1
18
Forcing AI to answer every query leads to fabricated information. When it comes to reliable systems, prioritize accuracy and trust over chasing incremental benchmark gains. #AI #TechEthics
5