Have a system that allows you to deploy only the ones needed to run the test.

If you are creating non-text output, try finding ways to turn results into
plaintext. This enables version-control, search&replace of test acceptance
criteria.

Add tracing IDs to requests which should end up in logs as well so you can find
where something failed.

Once more than one team is working on a microservice architecture it gets
important to coordinate tests between these teams.

Version control of microservices is very important to enable this.