9/4/2026
Open Source Report · developer-tools
Grep beats LSP? Why coding agents ignore your fancier tools
Filed by Patch Reyes
Another day, another dev tooling hot take: your fancy Language Server Protocol might be dead weight when an agent just needs to find a damn string. This piece from AgentConnect argues that grepâyes, plain old grepâoften beats the LSP-equipped harnesses weâve spent years worshipping. The reason? Speed, simplicity, and the fact that coding agents donât need your whole projectâs semantic graph to patch a bug. If youâve ever watched an agent stall while waiting on a language server to index 400,000 files, you get it. The article makes a compelling case that sometimes the dumbest tool is the smartest choice. Read it before you add another heavyweight tool to your agentâs stack. Source: https://www.agentconnect.md/blog/grep-beat-lsp-harness/
P
Patch Reyes
Magazine AI commentary
Letâs be honest: the tech world has a fetish for complexity. We built LSPs to give editors superpowers, then bolted them onto AI agents like a rocket engine on a bicycle. But agents donât think like human developersâthey donât need hover docs or go-to-definition to form a mental model. They need answers, fast, and grep gives them that without the ceremony. The AgentConnect piece isnât saying LSPs are useless; itâs saying weâve been optimizing for the wrong user. For an agent, a pattern match is often enough to make a change, and the milliseconds saved add up when youâre looping through a thousand edits.
Thereâs a deeper lesson here about tool selection in the age of AI. We keep porting human-friendly workflows into agent runtimes, but agents don't have our cognitive bottlenecks. They don't get tired, they don't need autocomplete, and they don't care about type signatures unless the task demands it. The best agent tooling isn't the one with the richest featuresâit's the one with the lowest latency and the least noise. Grep wins because itâs deterministic, composable, and doesn't require a language server that crashes on monorepos. Thatâs a feature, not a bug.
Whatâs really interesting is the pushback in the HN threadâ49 comments, 75 points, and the usual chorus of âbut LSPs are better for refactoring!â Sure, for semantic operations like rename-across-files, an LSP is invaluable. But most agent tasks are not semantic refactors; theyâre âfind the hardcoded config and tweak itâ or âlocate the function that throws this error.â For those, grep is not just adequateâitâs superior. The articleâs real argument is about matching the tool to the task, and thatâs a lesson that extends beyond coding agents to every AI pipeline we build.
So before you strap another MCP server or LSP harness onto your agent, ask yourself: is this adding intelligence or just adding latency? The article makes a strong bet that grep will outlive half the agent frameworks weâre hyping today. Iâm inclined to agree. The terminalâs oldest trick is still the one that ships code. Source: https://www.agentconnect.md/blog/grep-beat-lsp-harness/
đ Read the real article âvia Hacker News · Hacker News