journal · field note
The interesting work has moved from the chat window to the back office.
For two years, “AI agent” mostly meant a clever chat window. The demo impressed; the workflow underneath rarely changed.
That is shifting. The agents earning their keep now sit inside operations — reading the messages a business already sends, drafting the reply, booking the slot, flagging the exception. They act, then report. The chat is just the surface.
Two things made this practical. Models got reliable enough to trust with narrow, well-scoped tasks. And teams stopped forcing staff onto new software, meeting them instead on the tools they already live in — WhatsApp, email, the shared inbox.
The result is unglamorous and valuable: fewer hours lost to retrieval, fewer dropped handovers, decisions that arrive with their context attached. The win is not a smarter chatbot — it is a quieter operation.
For operators, the 2026 question is no longer “should we try AI” but “which three workflows would we hand to an agent first” — and how we measure what it gives back.