Assuming id_pracownika is The PRIMARY KEY of the table. Or at least defined UNIQUE. (If it's not NOT NULL, NULL is a corner case.)
SELECT or INSERT
Your function is another implementation of "SELECT or INSERT" - a variant of the UPSERT problem, which is more complex in the face of concurrent write load than it might seem. See:
With UPSERT in Postgres 9.5 or later
In Postgres 9.5 or later use UPSERT (INSERT ... ON CONFLICT ...) Details in the Postgres Wiki. This new syntax does a clean job:
CREATE OR REPLACE FUNCTION hire(
        _id_pracownika integer
      , _imie varchar
      , _nazwisko varchar
      , _miasto varchar
      , _pensja real)
  RETURNS text
  LANGUAGE plpgsql AS
$func$
BEGIN
   INSERT INTO pracownicy
          ( id_pracownika, imie, nazwisko, miasto, pensja)
   VALUES (_id_pracownika,_imie,_nazwisko,_miasto,_pensja)
   ON     CONFLICT DO NOTHING;
   IF FOUND THEN
      RETURN 'OK';
   ELSE
      RETURN 'JUZ ISTNIEJE';  -- already exists
   END IF;
END
$func$;
About the special variable FOUND:
Table-qualify column names to disambiguate where necessary. (You can also prefix function parameters with the function name, but that gets awkward quickly.)
But column names in the target list of an INSERT may not be table-qualified. Those are never ambiguous anyway.
Best avoid ambiguities a priori. Some (including me) like to prefix  all function parameters and variables with an underscore.
If you positively need a column name as function parameter name, one way to avoid naming collisions is to use an ALIAS inside the function. One of the rare cases where ALIAS is actually useful.
Or reference function parameters by ordinal position: $1 for id_pracownika in this case.
If all else fails, you can decide what takes precedence by setting #variable_conflict. See:
There is more:
- There are intricacies to the - RETURNINGclause in an UPSERT. See:
 
- String literals (text constants) must be enclosed in single quotes: 'OK', not - "OK"
 
- Assigning variables is comparatively more expensive than in other programming languages. Keep assignments to a minimum for best performance in plpgsql. Do as much as possible in SQL statements directly. 
- VOLATILE COST 100are default decorators for functions. No need to spell those out.
 
Without UPSERT in Postgres 9.4 or older
...
   IF EXISTS (SELECT FROM pracownicy p
             WHERE  p.id_pracownika = hire.id_pracownika) THEN
      RETURN 'JUZ ISTNIEJE';
   ELSE
      INSERT INTO pracownicy(id_pracownika,imie,nazwisko,miasto,pensja)
      VALUES (hire.id_pracownika,hire.imie,hire.nazwisko,hire.miasto,hire.pensja);
    
      RETURN 'OK';
   END IF;
...
But there is a tiny race condition between the SELECT and the INSERT, so not bullet-proof under heavy concurrent write-load.
In an EXISTS expression, the SELECT list does not matter. SELECT id_pracownika, SELECT 1, or even SELECT 1/0 - all the same. Just use an empty SELECT list. Only the existence of any qualifying row matters. See: