Large language models are only as useful as the context they can reach. Out of the box, an LLM knows nothing about your internal database, your company's ticketing system, or the CSV file sitting on y
Markdown supported · ⌘↵ to post
Kartik N V J K
Velocity gains from AI vanish when review debt piles up. I cap open AI pull requests so the team finishes one batch before the next wave lands.
The N x M integration problem framing is exactly why MCP took off, but the security side deserves equal space. I evaluate every MCP server my agents touch against a short checklist, tool descriptions that overstate scope, missing rate limits, and endpoints that accept arbitrary paths, because a malicious server can steer the agent through its own tool descriptions. I wrote my full checklist here: kartiknvjk.hashnode.dev/how-i-evaluate-mcp-server… . How do you sandbox servers you did not write yourself?