An AI Agent that write queries to the data source based on a natural language question

### Discussed in https://github.com/grafana/grafana/discussions/71618 <div type='discussions-op-text'> ^{Originally posted by **samber** July 13, 2023} What about a little companion for writing PromQL, SQL, Elasticsearch... queries faster, based on the actual data source schema? **Example:** The user type the following question: "http request rate per API version for api.acme.org endpoint". Grafana would send it to a LLM (GPT4?) with all Prometheus metric names+descriptions. As soon as the LLM replies with a few metrics matching the request, a new LLM query could be sent, containing the labels and some value samples. Then, the returned PromQL query is inserted into the panel. Selecting a data source dynamically or even the panel type would be awesome. With an advanced prompt, I'm pretty sure it would be possible to prefill the query with board variables. Metabase built a similar feature recently: Metabot. </div>