Field notes from the work.
Ideas and deep-dives on building software and the AI inside it.
The Answer's in Your Docs. Your RAG Can't Find It.
Your RAG assistant gives a vague non-answer, and the maddening part is that the real answer was sitting in your documents the whole time. It didn't miss because the model is dumb. It missed because search is a funnel with three leaks: the answer got chopped up when you filed it, the search matched the wrong words, and the right passage got ranked too low to make the cut.
Read →Your Agent Remembers Too Much
An AI agent working a long task stuffs everything it sees into its short-term memory, until the memory is full of junk and it forgets what it was doing. The fix is not a bigger memory. It is teaching the agent what to throw away: keep the goal and the facts that matter, summarize the rest, and file the details somewhere it can look them up later.
Read →Your Agent Doesn't Need a Genius for Every Step
An AI agent fires model call after model call to handle one request, and almost all of them are clerical work: sort this, pull out that, pick a tool, fill in its form. Running every one on a top-tier model is hiring a genius to file paperwork. Let a small model do the routine steps and call the expensive one only where it earns its keep.
Read →Your Agent Returned 200 OK. It Was Still Wrong.
An AI agent can pass every check, throw no error, and quietly do the wrong thing. Why 'no errors' tells you nothing about an agent, and the two things you need instead: the trace and the evaluation.
Read →Does AI need its own filesystem?
Everyone blames storage when the GPUs sit idle. A few labs really did build new filesystems for AI. For everyone else the answer is no, and the reason is worth more than the storage.
Read →Fine-tune, RAG, or just prompt? Match the cure to the symptom
The three get pitched as rivals. They are not: each fixes a different failure. Pick by what sounds advanced instead of what is actually broken, and you can burn months fine-tuning a problem retrieval would have solved in an afternoon.
Read →A GPU is only as fast as the data you can feed it
Under every AI system runs a storage layer that decides whether your expensive chips are ever actually used, and its job flips completely between training a model and serving one.
Read →Before You Buy the GPUs
Your AI bill is climbing and someone asks: shouldn't we just self-host? The honest answer is a math problem most teams never run. Here's the real cost, the break-even, and when each option actually wins.
Read →Your Agent Is 95% Reliable. That's the Problem.
An AI agent can ace the demo and still fail most of the time in production, and it isn't the model's fault. Why agents break as the task gets longer, why benchmarks mislead, and the boring engineering that fixes it.
Read →No Model Can Outrun a Silo
Enterprise AI keeps stalling, not on the model, but on data trapped in silos. Why that happens, thirty years of fixes that didn't stick, and what actually makes data AI-ready.
Read →The Case of the Runaway AI Bill: the cost cliff between demo and production
Most AI pilots never reach production, and the usual culprit is the bill, not the technology. A detective's walk through one runaway AI bill: the six suspects, the red herring that hid them, and the easiest, highest-impact ways to close the case on yours.
Read →Your AI Edge Is for Rent: what a samosa stall teaches every company building with AI
Follow Priya, who runs a samosa stall and spends a year building her own fryer, her own taste-tester, and her own sauce factory, only to find the wholesaler now sells all three cheap. A plain two-question test for what's actually worth building yourself when you build with AI, and the few things genuinely worth owning.
Read →