They are not the same. How to choose between a chatbot, RAG over your documents and an agent that executes, with COP ranges and Medellín examples.
Chatbot, RAG and agent are not synonyms. Hiring the trendy label is the most expensive way to not fix the process. In Medellín we see it every week: an “agent” that only answers FAQs, or RAG when what you need is a CRM write.
Short definitions
- Chatbot: it talks. Tree or LLM, no tools. Fits hours and closed questions.
- RAG: the model searches your documents and cites. Fits policies, contracts, catalogs, SOPs.
- Agent: it decides and executes (create lead, open a task, call an API, escalate to a human).
What to hire by pain
Support that copies the manual: RAG. Sales that does not log WhatsApp: agent + n8n + CRM. A landing that wants a widget: chatbot, and we will not bill it as autonomous.
Guide prices
Scoped chatbot: below an agent. RAG pilot: COP $12–25 million. Agent with tools: from USD 5,000. The 45-minute diagnosis avoids paying for an agent when RAG was enough.
Common mistake
Fine-tuning “so it sounds like the brand” when prices change every month. That is RAG or a table. Fine-tuning comes later if the output format is rigid and the corpus is stable.
How we start at Veflat
One process, one metric, one channel (often WhatsApp). If RAG and agent coexist, RAG is the agent’s tool, not a separate theatre product.
Next step
Agent, RAG and consulting pages. Or WhatsApp with the process that hurts most.