blog.md · page 2 of 5
Articles on AI, engineering, and building products by Subodh Jena.
Multi-agent architectures amplify both capability and error. Google DeepMind measured the amplification across 180 configurations. Anthropic reports 90 percent cost reductions from a planner-plus-workers split. Both results inform how to scale production agents.
8 min read
Semantic RAG is a fixed pipeline: chunk, embed, search, generate. Tuned well, it handles most production lookup workloads at a cost the alternatives cannot match. A short walkthrough of the pieces and the upgrades that matter.
6 min read
A harness is the set of components around a language model that turn a research prototype into a production system. Anthropic's three-agent architecture (planner, generator, evaluator) mirrors the GAN discriminator-generator dynamic and survives long-running applications.
8 min read
A provider's SDK earns its keep on six concerns: auth, request shape, transport, retries, streaming reassembly, and response parsing. Everything above that, the agent loop, planning, routing, memory, parallelism, guardrails, and workflow state, is application architecture and is better written in pure Python.
16 min read
An agent that cannot be measured cannot be improved. Evaluation starts with twenty queries and an LLM-as-judge, scales up through pass-at-k metrics and standard benchmarks, and never trusts any single layer alone.
8 min read
Subquadratic launched yesterday with a 12-million-token model and a $29M seed. The architecture pitch rests on a selector that has to be subquadratic in sequence length. The launch materials describe the property without describing the mechanism.
10 min read
The lunch-table conversation about AI agents and the production-engineering conversation about AI agents are both right. They are talking about three different things and calling all of them "agent."
10 min read
Safety instructions in a system prompt are not sufficient for production. Benchmark agents still behave unsafely in a significant fraction of cases. The fix is layered defense: prompt-level, tool-level, model-level, and system-level guardrails stacked so any one failure is caught by the next.
8 min read
Full autonomy is not the right target for every agent. Put a human in the loop at the points that matter: before dangerous actions, for ambiguous decisions, as an escalation path when the agent is stuck. Checkpointing makes the pauses cheap.
8 min read
Anthropic's Skills pattern packages reusable procedural knowledge as folders of instructions, scripts, and resources. Progressive disclosure loads only the metadata at startup and brings in the details on demand. Context capacity becomes effectively unbounded.
7 min read