Will the lift coefficient (Cl) and drag coefficient (Cd) I obtain by using software simulations or subsonic wind-tunnel tests be the same as on an actual aircraft while cruising? Or will there be some error?
In computational fluid dynamics (CFD) there are limits for the calculation of turbulent flow. This means that large flow separation cannot be modeled precisely; especially its time-dependent characteristics are computationally much too complex to be modeled. Instead, statistical methods must be used.
For attached and far-field characteristics the methods are refined enough to yield very accurate results, so much that what had historically been studied in wind tunnel tests or only on the finished aircraft can now be calculated with confidence up to hypersonic speeds. Errors still exist - after all, if the geometric model does not include the manufacturing irregularities of the real aircraft, the results will not be in total agreement, but close enough for engineering purposes.
In wind tunnel testing you cannot expect the same results as long as the Reynolds number and Mach number are different from the real thing. This is adjusted with correction factors, and again the results become good enough for engineering purposes. Cryogenic wind tunnels even remove this restriction, but are very expensive to build and operate, because they run with pure nitrogen at very low temperature and several bars of pressure. Small errors remain, because all the details of the real aircraft cannot be modeled in a wind tunnel model and their ratio between dynamic pressure and stiffness cannot be adjusted for all flow conditions, so aeroelastic effects will remain different.
Windtunnel tests are still valuable for predicting forces in post-stall conditions and for calibrating new CFD codes. Most of the operational envelope can then be covered with CFD. The combination of both is enough to predict the force coefficients with confidence, and only aeroelastic effects and details of the stall behavior might turn out slightly differently than anticipated from tests.
As @TrebiaProject points out in the comments, CFD can help with the corrections. The influence of the sting the model is mounted on would be a primary example. CFD can help to understand its influence and correct for it. This makes CFD complementary, so both techniques will lead to better results if combined correctly.