Vercel 
-
The problem
Vercel gets a large volume of feedback from the field: sales calls, support cases, Slack threads, posts on X and notes submitted by customer-facing staff. There was no reliable way to connect that feedback to the feature requests product teams were already tracking.
-
What they built
A system that reads through all of that feedback, links each piece to an existing feature request, and sends a weekly report to 22 product teams. Each report says how many calls and support cases were checked and how many were relevant, summarizes them, and links to the original evidence. A dashboard shows the top requested features, the accounts linked to each one, and how the evidence builds week over week. If a link is wrong, people can remove it, and that correction feeds back into how future feedback is sorted.
-
So what
What reps and support hear reaches product every week with the accounts attached, instead of getting lost in call recordings and Slack threads.
Under the hood
For GTM engineers
A weekly cron job triggers durable Vercel Workflows, which handle thousands of LLM calls and save intermediate results in batches as checkpoints, so reports can be regenerated from the same data. Vector embeddings narrow the candidate feature requests (300 in his example) to a top 30 per piece of feedback; Brandon found the top 25 usually held about 90% of the relevant matches, which cut cost compared with checking 100. Because product, engineering and field teams disagreed when hand-labeling relevance, he tuned for usefulness to each product team plus an unlink flow rather than a perfect eval set, and used a second agent to review each full run. Built with the AI SDK and AI Gateway; Vercel's agent framework delivers the digest into Slack.
More examples
Other systems GTM engineers have built, or see all of them.
-
WorkOS Gave the whole go-to-market team one shared record of every account and contact, which anyone can ask questions of or act on without help
Jack Perales · GTM Engineer
-
Modal Turned the black box behind Modal's account data into a console the team can use to see what's flowing, what's broken and why
Chris Prinz · GTM Engineer -
Attention Sent each new lead to the rep most likely to win it, based on deal history and shared background; win rates rose 30%
Jacob Fleisher · Head of Sales
Hiring someone who can build like Brandon?
We search for GTM engineers who have shipped systems like this one, and Joe speaks to every candidate before you do.