same prompt, different language, different picks
next.js apps got resend, python apps got raw smtp. and a few more splits like that.
hey, issue two. last time was which model picks what. this time it's the same tasks in a next.js app and in a python app, and the language moved things more than i expected.

email is the cleanest split. next.js apps mostly get an email service, python apps almost never do.

17 of 18 python runs wired up smtp by hand. very 2009 of it.

python apps got heroku in 7 of 18 runs and next.js apps never did. somewhere a procfile is very happy.

python is the one place everyone agrees. sqlalchemy all 18 times. next.js mostly goes with drizzle, and 5 times it skipped the orm entirely.

this one surprised me. python apps got sentry twice as often, and opus in a next.js app never added it at all, 0 of 6.

and some things don't care what you write in. switching languages won't save you from a homemade login.

a key in your .env might though, at least in next.js, which used it all 27 times while python still hand-rolled 20 of 27. that's a separate little test with sonnet 5.5, kept off the scoreboard.
so if you're on python i'd look at how your email gets sent first. on next.js i'd check that something is actually watching your errors.
next issue i'll keep poking at what nudges these picks. if there's a tool you want checked, just reply.
sponsored
this slot's open. one labelled block per issue, and it never touches the numbers. whatagentspick.com/sponsor
nika