https://blog.aiaugmentedsoftwaredevelopment.com/tags/agent-safety/Recent content in Agent-Safety on AI Augmented Software Developmenthttps://blog.aiaugmentedsoftwaredevelopment.com/images/og-default.pnghttps://blog.aiaugmentedsoftwaredevelopment.com/8831Hugo -- 0.150.0en-us2026 AI Augmented Software DevelopmentFri, 18 Sep 2026 05:10:32 -0400https://blog.aiaugmentedsoftwaredevelopment.com/posts/ai-safety-engineering-problem/?utm_source=rss&utm_medium=feed&utm_campaign=rss-subscribersFri, 18 Sep 2026 00:00:00 +0000https://blog.aiaugmentedsoftwaredevelopment.com/posts/ai-safety-engineering-problem/Agreeing that AI safety is an engineering problem is a much bigger claim than most people making it intend. Model-side safety is necessary. It cannot be load-bearing, and harm assembled from locally benign pieces is why.https://blog.aiaugmentedsoftwaredevelopment.com/posts/who-gets-the-keys-to-ai/?utm_source=rss&utm_medium=feed&utm_campaign=rss-subscribersTue, 15 Sep 2026 00:00:00 +0000https://blog.aiaugmentedsoftwaredevelopment.com/posts/who-gets-the-keys-to-ai/Who should be allowed to use AI? Comparing access rules for cars, guns, and alcohol to who gets the keys — and who's accountable when it goes wrong.https://blog.aiaugmentedsoftwaredevelopment.com/posts/designing-the-control/?utm_source=rss&utm_medium=feed&utm_campaign=rss-subscribersTue, 08 Sep 2026 00:00:00 +0000https://blog.aiaugmentedsoftwaredevelopment.com/posts/designing-the-control/How to design human review for AI agents: label actions by danger, match the check to the stakes, verify by trying to refute, and add automatic brakes.https://blog.aiaugmentedsoftwaredevelopment.com/posts/the-missing-review-tier/?utm_source=rss&utm_medium=feed&utm_campaign=rss-subscribersTue, 01 Sep 2026 00:00:00 +0000https://blog.aiaugmentedsoftwaredevelopment.com/posts/the-missing-review-tier/Why AI agents with real tools need human review: guardrails leak, danger hides in the sequence of steps, and the fix starts at the connections, not the model.