SYSmith Yaqoobinaeoranker.hashnode.dev·Aug 24 · 4 min readHow to Evaluate an AI Tool Before You Pay for ItEvery AI product demo is optimised for the same thirty seconds. Impressive output, clean input, ideal conditions. What it does not show is behaviour at the edges, and the edges are where the tool eith10
SYSmith Yaqoobinaeoranker.hashnode.dev·Aug 23 · 7 min readDesigning a Corpus: Ownership, Staleness, and the Quarterly ReviewMost RAG write-ups are about retrieval. Chunk sizes, embedding models, rerankers, hybrid search. All of it matters and I've written about it. None of it helps if the corpus underneath is wrong. I buil10
SYSmith Yaqoobinaeoranker.hashnode.dev·Aug 20 · 7 min readInstrumenting AI Workflows: What to Log So You Can Tell If It's WorkingMost AI workflows are unmeasured. Not lightly measured unmeasured. They get built, they run, and the only signal anyone has about whether they're worth keeping is a vague sense that things feel faster00
SYSmith Yaqoobinaeoranker.hashnode.dev·Aug 19 · 8 min readDesigning the Boundary Layer: Least Privilege for AI AgentsIn the last piece I audited my stack and found 5 tools that were still authorised to read my email months after I'd cancelled them. The interesting question isn't how that happened it happened because00
SYSmith Yaqoobinaeoranker.hashnode.dev·Aug 18 · 7 min readAuditing an AI Agent Stack: Cost, Scope, and the Credentials You Forgot You IssuedMost "AI stack" posts are shopping lists. This one is an audit, the thing you run on a stack you already have. I did it on my own setup last month across [NUMBER] tools and services. The cost finding 00
