https://tommyeriksen.dk/tags/ai-agents/Recent content in AI agents on tommyeriksen.dkHugoenWed, 14 Oct 2026 00:00:00 +0000https://tommyeriksen.dk/posts/what-my-agents-arent-allowed-to-do/Wed, 14 Oct 2026 00:00:00 +0000https://tommyeriksen.dk/posts/what-my-agents-arent-allowed-to-do/<p><img src="https://tommyeriksen.dk/posts/what-my-agents-arent-allowed-to-do/least-privilege-flow.png" alt="Diagram: an agent commits to a repository, a human review gate stands between the repository and production, and the agent only has read-only access to observability" /> </p> <p>Most of what I write about agents is what they do. Commit code, run tests, review each other&rsquo;s work. This one is about what they can&rsquo;t do, because that list is the one I spend the most time on.</p> <p>The rule is old and boring: least privilege. Give a thing the access it needs to do its job, and not a bit more. We&rsquo;ve applied it to people and service accounts for decades. It feels slightly silly to apply it to an agent that&rsquo;s very keen and very polite. That&rsquo;s exactly why you should.</p>https://tommyeriksen.dk/posts/making-myself-redundant/Wed, 30 Sep 2026 14:00:00 +0200https://tommyeriksen.dk/posts/making-myself-redundant/<p><img src="https://tommyeriksen.dk/posts/making-myself-redundant/agent-guardrails-hero.jpg" alt="Small robot agents racing along a track, kept on course by guardrails, towards a human approval gate" /> </p> <p>So. When I did infrastructure engineering, I had a mantra: make yourself redundant. Whatever you&rsquo;re doing, automate it so it can run over and over. That freed me up for the important stuff, like coffee. Or YouTube (which hadn&rsquo;t really been invented yet).</p> <p>Come to think of it, when I made the roadmap for my first PO gig (an operations team), there was a dot on it where it made sense for me to exit and a new skill set to take over.</p>