conda is the most-misunderstood package manager in this quest because it grew up in a non-web-developer world. These rules cross the cultural gap.
Use Miniconda or Miniforge, not Anaconda. The full Anaconda distribution is 3 GB of pre-installed scientific stuff most people never touch. Install conda; install what you need.
Set conda-forge as default channel. More packages, more recent versions, no commercial licensing strings.
One env per project. Same rule as venv — never install into the base env. conda activate per project; commit environment.yml to the repo.
Use --from-history when exporting envs to share. The default export pins exact build hashes which often don't exist on different platforms; --from-history exports your intent (just what you asked for), and conda re-resolves on the target machine.
For new Python projects, consider pixi. pixi (next track) is conda-forge compatible but 10x faster, with native lock files. It's 'conda done right' for projects where you'd otherwise reach for conda but want better DX.
Don't fight the slow base env. If conda update or conda install in the base env is sluggish, that's expected — base envs accumulate. Create per-project envs and let base be light.