https://vinhthang.dev/tags/ai-agents/Recent content in AI Agents on Vinh Thang's AI Tech NotesHugoen-usSat, 05 Sep 2026 14:00:00 +0700https://vinhthang.dev/posts/attention-guard-2/Sat, 05 Sep 2026 14:00:00 +0700https://vinhthang.dev/posts/attention-guard-2/<p>As autonomous AI agents evolve from isolated chatbots into orchestrators of complex, multi-pass software engineering tasks, the boundaries of execution become critical. In the <strong>Antigravity</strong> ecosystem, we rely heavily on specialized subagents—like <code>DeepCoder</code> and <code>DeepInvestigator</code>—to autonomously navigate codebases, test hypotheses, and execute shell commands.</p> <p>But what happens when an agent hallucinates a success? What happens when a long-running subagent hangs indefinitely? How do we prevent a deeply nested execution worker from going rogue and launching its own recursive subagents?</p>