We just shipped a massive latency improvement in a critical Vercel API endpoint. One weird "trick" I've found again and again through my engineering career: move things *out* of the critical path. The issue is that it's been hard to do this reliably. 𝚠𝚊𝚒𝚝𝚄𝚗𝚝𝚒𝚕 is not enough, even though Fluid will try its darndest to run it, it's not a *guarantee*. In most cases, you absolutely need the asynchronous work to happen, no matter what. Network failures, machine failures, code rollouts notwithstanding. We're fixing this next week btw. You don't want to miss @pranaygp's talk at – what we'll present is one of the most exciting pieces of OSS.