A technical publication by Shubham Jindal

Not the LLM

The interesting engineering happens around it.

Notes on agents, context, tools, runtimes, evals, governance, and the systems that make AI actually work.

Production AI / system cutawayA model call, wrapped for reality.
  1. Identity & Policyauth · permissions · audit
  2. Governancerisk · review · accountability
  3. Observabilitytraces · latency · cost
  4. Evalsquality · regressions · task success
  5. MCP + toolscapabilities · data · action
  6. Harnessplans · loops · state · retries
  7. Contextinstructions · retrieval · memory
raw capabilityLLMreasoning + generation
Runtimestate · queues · latency · cost · scale
Capability starts at the center.Reliability accumulates outward.

Latest

Writing

MCP

It's Not the LLM. It's the MCP.

Why capable agents fail when their tool interfaces overwhelm context, and how small verbs plus registry dispatch change the architecture.