How do you write your applications to be database independent?

You could use one of the many Object/Relational Mapper tools, like Hibernate/NHibernate, LLBLGen, and so forth. That can get you a long way to database portability. No matter what you do, you need to have some abstraction layer between your model and the rest of your code. This doesn't mean you need some sort of dependency injection infrastructure, but good OO design can get you pretty far. Also, sticking with plain SQL and thinking that will get you portability is rather naive. That would be true if your application was trivial and only used very trivial queries.

As for always writing an application to be "any database ready," I usually use some sort of abstraction layer so it is not hard to move from one database system to another. However, in many circumstances, this is not required, you are developing for the Oracle platform or SQL Server or MySQL whatever so you shouldn't sacrifice the benefits of your chosen RDBMS just for the possibility of an entirely seamless transition. Nevertheless, if you build a good abstraction layer, even targeting a specific RDBMS won't necessarily be too difficult to migrate to a different RDBMS.

To decouple the database engine from your application, use a database abstraction layer (also data access layer, or DAL). You didn't mention what language you use, but there are good database abstraction libraries for all the major languages.

However, by avoiding database-specific optimizations you will be missing out on the advantages of your particular brand. I usually abstract what's possible and use what's available. Changing database engines is a major decision and doesn't happen often, and it's best to use the tools you have available to the max.

Tell your boss to mind his own business. No, of course one can't say such things to one's boss, but stay tuned.

What's interesting is what business value is supposed to be supported by this requirement. One obvious candidate seems to be that the database code should be ready for working on other database engines than the current. If that's the case then that's what should be stated in the requirement.

From there it's up to you as an engineer to figure out the different ways to achieve that. One might be writing ANSI SQL. One might be using a database abstraction layer.

Further it's your responsibility to inform your boss what the costs of the different alternatives are (in terms of performance, speed of development, etcetera).

"Write ANSI SQL"... gah!