https://builderdecade.com/tags/ai-agents/Recent content in AI Agents on Builder DecadeHugoenFri, 16 Oct 2026 07:30:00 +0800https://builderdecade.com/posts/trust-scope-reversibility/Thu, 08 Oct 2026 08:00:00 +0800https://builderdecade.com/posts/trust-scope-reversibility/The framework for what to hand an agent isn't about how smart the model is. It's reversibility × blast radius × stakes — and here's where the line sits for a one-person build.https://builderdecade.com/posts/decisions-are-correlating/Wed, 07 Oct 2026 08:00:00 +0800https://builderdecade.com/posts/decisions-are-correlating/When a platform hands the same AI to everyone, the sameness is the risk — one judgment multiplied into a market. Here's why differentiation is now a system-safety property, not just a moat.https://builderdecade.com/posts/what-your-agent-can-reach/Mon, 21 Sep 2026 18:00:00 +0800https://builderdecade.com/posts/what-your-agent-can-reach/Seven 2026 sandbox escapes, no exploit in any of them. The vulnerability is the trust you extend after the agent writes something — here's the check.https://builderdecade.com/posts/mcp-security-solo-builder/Sun, 20 Sep 2026 08:00:00 +0800https://builderdecade.com/posts/mcp-security-solo-builder/You pasted an MCP server in from a blog post. It holds your files and your keys. Here's the 10-minute check before you connect the next one.https://builderdecade.com/posts/agents-fail-boringly/Thu, 10 Sep 2026 08:30:00 +0800https://builderdecade.com/posts/agents-fail-boringly/AI agent failures are rarely dramatic, they're mechanical: one missing bracket, one bad permission. The skill is reading the logs. Here's the discipline.https://builderdecade.com/posts/memory-is-the-product/Mon, 07 Sep 2026 08:20:00 +0800https://builderdecade.com/posts/memory-is-the-product/The agent isn't the model. Memory, rules and trust rituals are the product. Here's why the smarter-bot upgrade won't fix your workflow.https://builderdecade.com/posts/done-is-a-claim/Sun, 06 Sep 2026 08:30:00 +0800https://builderdecade.com/posts/done-is-a-claim/With AI agents, "done" starts meaning "the agent says it's done." Here's the verification ritual that separates product from hallucination.https://builderdecade.com/posts/supervisor-class/Sat, 05 Sep 2026 08:00:00 +0800https://builderdecade.com/posts/supervisor-class/The supervisor class: your job shifts from doing the work to specifying, checking and owning it. Here's the discipline that requires.https://builderdecade.com/posts/agent-patterns-worth-knowing/Tue, 01 Sep 2026 07:30:00 +0800https://builderdecade.com/posts/agent-patterns-worth-knowing/Three agent patterns worth knowing before you blame the model: routing, tool use, reflection. Diagnose the failure, not the pattern.https://builderdecade.com/posts/api-vs-mcp/Sat, 29 Aug 2026 00:00:00 +0000https://builderdecade.com/posts/api-vs-mcp/API or MCP? They solve different problems and the choice shapes your whole integration. Here's how to tell which one you actually need.https://builderdecade.com/posts/what-agents-cant-do/Sun, 23 Aug 2026 00:00:00 +0800https://builderdecade.com/posts/what-agents-cant-do/AI agents are extraordinary at the last mile and terrible at the first. Here's the line worth knowing before you hand one the job.