Let's frame the problem before we pick a stack. What are we optimizing for — latency, cost, or developer speed?

Developer speed, mostly. We're pre-PMF, so shipping fast beats shaving milliseconds. I'd reach for a managed queue.

Agreed on managed, but let's not paint ourselves into a vendor corner. A thin adapter keeps the swap cheap later.

With a 3-month runway the adapter is a luxury. Managed queue, direct integration, revisit at scale.

Then we agree on the queue. I'd still time-box a spike on the worker model — retries are where these designs go sideways.