Well, even most optical telescopes aren't steered by hand and guided by human eyes - aside from a small fraction of objects, most sources are too dim to be seen by the naked eye or even with assistance, and observations require precision far beyond the ability of humans. A hobbyist telescope where less precision is needed and most sources are relatively bright could be adjusted by hand; a telescope being used for research couldn't - or certainly shouldn't.
Radio telescopes use a procedure also utilized by telescopes at other wavelengths: observing scripts and scheduling blocks. These contain information about the observing parameters - such as the telescope configuration, receivers and backends used, and the sources themselves - and commands that tell the telescope to slew to a point in the sky, take data, track the source, etc. Here's an extremely simple demo scheduling block for use on the Green Bank Telescope (see Section 6 of Observing with the Green Bank Telescope for additional details on it):
# load the configurations file
# load catalogs file
# configure the GBT
# slew to the source
# balance the IF system
# now observe the source for ten minutes
This block can use Python commands within it, allowing it to, for instance, loop through a list of sources. Notice that it loads in a source from a catalog (the Catalog() command) and automatically slews to the source (Slew()) without any human intervening; you just submit the block and let it run. The catalog contains all the information needed for the telescope to find a source, including its right ascension and declination.
Pointing and focus errors can creep in and do require additional calibrations, particularly for high-frequency continuum observations. This can be somewhat achieved by applying a pointing model (e.g, White et al. 2021) taking into account the telescope's structure, current positioning, thermal deformations, etc. However, for even better pointing correction, calibration sources should be used, which may be chosen from a specific catalog of pointing sources. Again using the GBT as an example, this can be achieved by running
AutoPeakFocus() in a scheduling block to perform pointing and focus scans prior to gathering the actual data, or simpler variants to just do one of the two. This will then automatically update the corresponding corrections. Again, this is significantly more important in certain regimes and certain targets.
It is possible for the telescope operator - the observatory staff member on site in the control room coordinating and aiding observations - to interrupt scripts and make unscheduled changes. This is typically in the event of high winds (the GBT, for example, must be stowed if winds exceed 35-40 mph) or if snow or ice has built up on the dish, in which case observations must be paused for the snow to be dumped. That said, this isn't really done by the operator looking at the telescope (two kilometers away from the control room), but by computers that can calculate and move the telescope to the precise position necessary to keep it safe.