EIEshan Inamdarineshaninamdar.hashnode.dev·Jun 6 · 12 min readAgentic AI Deep Dive — Evaluation & ProductionWhere We Left Off Part 8 was about failure modes — the six ways agents break in production without anyone trying to break them. Infinite loops. Hallucinated tool calls. Silent failures. Context poison00
EIEshan Inamdarineshaninamdar.hashnode.dev·May 31 · 11 min readAgentic AI Deep Dive — Failure Modes: What Goes Wrong and WhyWhere We Left Off Part 7 was about deliberate attacks — prompt injection, trust boundary violations, agents with too much permission. The threat model was an adversary trying to manipulate your agent.00
EIEshan Inamdarineshaninamdar.hashnode.dev·May 23 · 12 min readAgentic AI Deep Dive — Security & TrustThe Story So Far — Full Series Recap Six parts in. Here is what we've built up. Part 1 — LLMs have four fundamental limitations: knowledge cutoff, no private data, statelessness, no actions. RAG fixed00
EIEshan Inamdarineshaninamdar.hashnode.dev·May 20 · 10 min readAgentic AI Deep Dive — Multi-Agent SystemsWhere We Left Off In Part 5 we solved the memory problem. A well-designed agent uses four memory types — in-context for the active task, external for large persistent stores, episodic to learn from pa10
EIEshan Inamdarineshaninamdar.hashnode.dev·May 19 · 11 min readAgentic AI Deep Dive — Memory: How Agents RememberWhere We Left Off In Part 4 we gave agents hands. Tools are how an agent reaches outside itself — into APIs, databases, browsers, file systems. The LLM describes an action, the harness executes it, th00
