Terminal coding agents grep 500k-line repos like it's a 30-year tech regression
Evening_Hawk_7470 · reddit · 2026-10-28
A Reddit post argues that watching terminal coding agents (Claude Code, Aider) work on 500k-line repos feels like technological regression: they spend ten minutes in ripgrep loops, miss overloaded methods in test fixtures, confuse shadowed variables, and burn 50,000 input tokens just locating an interface implementation.
- The contrast: software engineering spent 30 years building compilers, language servers and IDE indexers that resolve type hierarchies and cross-package call graphs in milliseconds — agents are throwing that away for raw text search.
- Author's test: standalone CLI agents guess at symbols, while Junie connected to IntelliJ via /ide queries the IDE's running AST index (Find Usages, type hierarchies, compiler-verified renames) before generating code.
- Boundary: text search + embeddings in a big context window work fine for self-contained scripts or greenfield microservices; multi-module Java/Kotlin monoliths are a disaster.
- Open question: can pure text-search CLI agents ever be reliable on massive codebases without burning a fortune in input tokens?
More from coding & agent
- Enterprise workflow automation is still in its 'autocomplete era' — but not for long — saranormous · 2026-10-28
- Garry Tan shows his Capy AI software factory: 10-20 subagents running constantly, 19,101 contributions/year — garrytan · 2026-10-28
- He Read vLLM Line by Line for Hours—Now With Codex He'd Never Do It Again — gordic_aleksa · 2026-10-28
- Developers report Apple is automating App Review, with trusted accounts approved almost instantly — rudrank · 2026-10-28
- N-queens solved entirely in Python type annotations with ty, each solution a tuple — charliermarsh · 2026-10-28
- MCP gap: services can't push to agents, dev argues spec needs server-initiated events — braelyn_ai · 2026-10-28