Introducing the HIA Agent Scale: A Better Way to Describe AI Agents
Not all AI agents work the same way, and existing classification frameworks don’t tell the whole story. We developed the HIA Agent Scale to describe both how much responsibility an AI system carries and the role it plays with the person using it. Here’s why we created it and how the six Levels and five relationship categories work together.
OpenAI is doing away with custom built GPTs
OpenAI’s move away from custom GPTs is more than a product migration. For organizations that turned them into internal tools, the shift to Projects, Skills, Plugins and MCP-based integrations can mean rebuilding how those systems work, testing them again and asking nontechnical builders to manage a more complex architecture.
Your AI agent may be working for someone else.
AI agents can read emails, documents, webpages, and other material created by people you don’t control. If they mistake someone else’s instructions for yours, the problem can move from a bad answer to an unauthorized action. This post looks at prompt injection, why agent permissions matter, and why a few seconds of human review can still be worth keeping.
Dear Claude Cowork, I Love You …Sometimes
AI agents can do things a normal chatbot can’t, which is both the appeal and the problem. After reluctantly giving Claude Cowork access to a dedicated laptop, I discovered how much time an agent can save when the job actually warrants it. The trick is figuring out how much access and autonomy the task really needs, what happens if something goes wrong, and whether you’ve brought in a backhoe to plant flowers.