Business chatbots should add RAG only when answers depend on company-specific documents
recro69 · reddit · 2026-07-21
A practical rule for business chatbots: only add retrieval when the answer truly depends on company-specific documents such as pricing, policies, or edge cases.
- If a well-prompted model can answer correctly from general knowledge, RAG adds latency, cost, and failure modes without much benefit.
- In production, failures are often caused less by model choice and more by poor message chunking, weak importance ranking, and missing retrieval tests.
- The author also argues for build vs. buy: keep the unique parts in-house, such as lead qualification, CRM integration, and domain-specific decision rules.
More from coding & agent
- OpenWiki adds Gemini AI Studio and Enterprise Vertex AI support — BraceSproul · 2026-07-22
- Video-based proof of work is emerging as a feedback layer for coding agents — Vjeux · 2026-07-22
- Salesforce open-sources MCP+ to filter tool output before it reaches agents — msrivastav13 · 2026-07-22
- Anaconda acquires Kilo Code to expand into agentic software development — anacondainc · 2026-07-22
- Poolside’s laguna s 2.1 claims 118B MoE, 1M context, and 78.5% on SWE-bench Multilingual — ben_burtenshaw · 2026-07-22
- OpenCode paired with Grok 4.5 can recreate a sharePDF site and backend — jasonkneen · 2026-07-22