How easy is it to use LLBLGen with an existing project. e.g. say I have an application in production but periodically do some enhancements. In such a scenario, would introducing and using LLBLGen be useful or not much useful or just too risky? Just like to hear some opinions.

A bit more specifics related to how you're doing data-access at the moment would be helpful

The project (web application) I am refering to has I guess two sections. One section is the admin section. i.e. only a few people have access to this section. So here I have used the SQL Datasource for CRUD operations.

The second section where many people may access at the same time, I have implemented a DAL, BLL and Entities (i.e. class representing a table). For binding to a datagrid I have used datasets. The dataset is returned from a class in the BLL, which in turn calls into the DAL. I have also sometimes created a relation in a dataset and then bind it to an Infragistics grid to display hierachical data.

I have not used much cacheing or transactions but have used session variables.

I can give more details, if required.

From my previous experience it should be safe to introduce LLBLGen to the solution. Use the Adapter model, and pass entities through the different tiers.