---
title: "More reliable, faster AI draft generation"
description: "AI drafts now process faster and recover on their own, with fair queueing across workspaces and a fast lane so template campaigns no longer wait behind long ..."
canonical: "https://firstsales.io/changelog/reliable-ai-draft-generation/"
---

1. [Home](/)
2. [Changelog](/changelog/)
3. More reliable, faster AI draft generation

[All updates](/changelog/)

FixJune 25, 20262 min read

# More reliable, faster AI draft generation

AI drafts now process faster and recover on their own, with fair queueing across workspaces and a fast lane so template campaigns no longer wait behind long autopilot jobs.

If your AI drafts were crawling — sitting in the queue, taking far longer than usual, or looking like they had stopped altogether — the slowdown was in how drafts get picked up and processed, not in the writing itself. Template campaigns were the most visible victims: a quick, cheap draft could end up stuck behind a long autopilot run.

## What was happening

The drafting queue had several weaknesses that compounded under load. A single busy workspace could dominate the queue and starve everyone else, so your drafts waited behind another account's backlog even when yours were tiny. Jobs that stalled partway through — a worker that claimed a draft and then never finished it — could sit in limbo with nothing reclaiming them, so they looked permanently stuck. When the AI provider slowed down and a request timed out, that request kept holding its processing slot instead of releasing it, so a handful of slow responses could tie up much of the available capacity at once. And every draft queued in the same single line regardless of cost: a lightweight template-campaign draft that should take seconds could end up waiting behind a long, heavy autopilot AI job.

## What we fixed

We reworked how drafts are claimed and processed end to end. Claiming is now fairer across workspaces, so one heavy account can no longer crowd out the rest of the queue. Stale or stuck jobs are detected and recovered automatically — a draft that stalls gets picked back up and finished instead of hanging indefinitely. We added a circuit breaker: after repeated drafting timeouts, it briefly pauses new claiming to let the backlog and the provider settle, then resumes on its own with no manual step from you. The worker pool is larger, so more drafts process at the same time. Requests that time out are now actively cancelled, which frees the slot immediately instead of leaving it occupied by a dead request. And we added a fast lane so lightweight template-campaign drafts move ahead instead of waiting behind long autopilot jobs.

## Why it matters

Drafts now move through the queue faster and far more predictably, and they keep moving even when another workspace is busy or the AI provider hits a slow patch. Your template campaigns are no longer held hostage by someone else's autopilot run, a stalled job recovers itself instead of silently blocking the line, and brief provider hiccups no longer cascade into a system-wide backlog. The result is fewer drafts that look stuck, shorter waits before your copy is ready, and a queue that heals itself instead of needing a nudge.

[ PreviousFailed enrichments now retry themselves](/changelog/self-healing-enrichment/)[Next Fewer bounces from harvested and looping addresses](/changelog/fewer-bounces-verified-recipients/)