If a job is failing to write to a std*file then we would advise engaging the system admin to identify what existing process has a lock on the file preventing the new update. If it is in fact the prior run of the job then looking into the agent's log and spool files may shed some light on the situation.
NOTE by default the spool files are removed if a job runs to SUCCESS. So you would need to disable that feature and restart the agent in order to dig further.
Set that to false in the agentparm.txt file on the machine which has the issue and restart the agent. This is so when you have a failure you could look at the job's prior run's logs to see if there was anything unusual as compared to when things run fine.
Anything special about the std*files?
Are they on a share?
Also please confirm the version of autosys and the version of the agent.
Also please provide any evidence you have gathered showing that the prior run is actually holding some lock on the std*files and open a support case if needed.