My opinion, after working for 4+ years with pathom, in at least 3 teams, ~5 distinct projects: Pathom is way ahead of the time. It is difficult to explain all the advantages of using pathom. And pathom requires many consensus from the team: • consensus about use clojure • consensus about use graph API • consensus about "data oriented" • consensus about qualified keywords • consensus about "do not type things" Only once all your team (for team here, include your CTO) agree on those things, you are free to use pathom and use all its advantages in its maximum power. Otherwise, it will only create conflicts. TLDR: kind of lisp curse

In a simple case, it feels quite powerful to be able to add calcualted and further derived values to any data model you have and not have to think about where or how the values you need get to you. You previously said X comes from Y and so X you know is in the graph of resolvable attrs. If you want to calculate Z as a function of X then you can write that behavior without worrying about where X is in your data model, as long as it's present in the graph (and resolvable at run time). Then you can really easily just request Z. Some other points I've noticed Pathom can do well: • Dependency injection • Runtime validations against schemas for attrs, or just using the https://pathom3.wsscode.com/docs/plugins in general. • More intentional, prolific, and shared naming of attributes (qualified keywords). I wouldn't say I was a skeptic about Pathom, but it certainly is a different technology that takes some getting used to. It's a very powerful tool and I'm glad to have it rather than not. But just like any tool, it's not right for every situation, but when you find a good fit, it's amazing. But I'd say the most convincing thing is seeing it in action.

Thank you for the replies, I appreciate the time and thoughtfulness. I was looking more for advice on choosing between pathom 2 vs 3 and the tradeoffs for building something new.

On the good side I’d add that pathom3 has a great batching mechanism to solve the n+1 problem On the bad side authorization is tricky, and performance is low when dealing with hundreds of entity in a response

pathom3 is a huge improvement in basically every way

This is really helpful, thank you! I didn’t realize that pathom2 didn’t have the batching mechanism. I know that will be needed.