For example you cant write a thread-safe map container in Go without sacrificing type safety (i.e. one that would not crash the program when there are concurrent writes to the map).
The standard library doesn't offer one either.
This means you end up copy pasting the lock related code for every map you want to make thread safe. You have to remember to declare and initialise its lock, to use that lock every time you access it, and to remember to unlock it.
Now for some things, this cost isn't too bad. For others, it makes the language a joke.
As a result, Go is a toy, or at least, a non-general-purpose language. If the problem you're solving fits its limited set of built-in abstractions, its great. Otherwise, its a disaster.
In my experience, well written programs evolve over time to gain new carefully chosen abstractions that cleanly implement shared underlying concepts of the problem domain. These abstractions often require the use of generics.
When this happens in a Go project, however, you're out of luck. Even worse, developing primarily in Go means distorting your thinking to fit its limits.
This is more or less true in any language, and its not a binary thing. Languages are all over the place on the ladder of abstraction (with lisps being close to the top, Go near the bottom).
p.s. Java also suffers because it added generics too late: by the time they were added there was already a clear picture painted of "idiomatic Java". To make matters worse lambdas were also added way too late, and the verbosity of type declarations is still staggeringly high. However, modern Java is a much better language than Go.
I've no particular love for Java's type system, but I can't criticize it as much because it was a lot more reasonable when it was released (1996). In 2016 there's no excuse for having a type system as broken as Go's.