<?xml version="1.0" encoding="UTF-8"?><rss version="2.0"><channel><title>Agent Robbie — The AI briefing</title><link>https://agentrobbies.com/blog</link><description>Independent AI news and practical context.</description><item><title>More capable agents need clearer permission boundaries</title><link>https://agentrobbies.com/blog/claude-agent-permissions</link><guid isPermaLink="true">https://agentrobbies.com/blog/claude-agent-permissions</guid><pubDate>Fri, 11 Sep 2026 12:00:00 GMT</pubDate><description>Claude’s Managed Agents update adds automatic tool-call evaluation. It is a useful reminder to design the approval rules alongside the task.</description></item><item><title>AI image editing is becoming a workflow, not just a prompt</title><link>https://agentrobbies.com/blog/image-editing-gets-practical</link><guid isPermaLink="true">https://agentrobbies.com/blog/image-editing-gets-practical</guid><pubDate>Fri, 11 Sep 2026 12:00:00 GMT</pubDate><description>ChatGPT’s latest image update puts more emphasis on refining an idea. For creatives, the useful question is how much control survives the next edit.</description></item><item><title>The next useful AI model might forecast demand, not write copy</title><link>https://agentrobbies.com/blog/open-models-for-forecasting</link><guid isPermaLink="true">https://agentrobbies.com/blog/open-models-for-forecasting</guid><pubDate>Fri, 11 Sep 2026 12:00:00 GMT</pubDate><description>IBM’s new open time-series model brings another option for forecasting. The meaningful test is still your own historical data.</description></item></channel></rss>