Your GTM belongs in git
Four operators showed the go-to-market they run in git, then argued about who owns it. The whole evening, with a timestamp on every talk.
Nobody on stage was selling version control. They each arrived at it from a different failure, and the failures rhyme: go-to-market logic is software now, so it needs a diff, a review, a preview of what changes and a way back.
Cole D'Ambra rebuilt Plain's scoring model five times before the pattern was visible. The logic lived in local folders and in CRM settings, so it had no memory, no rule for change and nothing to revert to: one more signal rewrote the whole thing and the briefs came back worse.
Hear it at 4:42Same logic, different home. The scoring model became one TypeScript file with over 500 automated checks behind it in GitHub Actions, and the Cargo workspace came along as TypeScript. His own line about it: you do not need an engineer to do this.
Hear it at 4:42Karl Rafidimanana's framing: a broken payment system fails loudly and a broken go-to-market workflow keeps running. An agent with a bad scoring model does not make one mistake either, it makes one every day, which is why moving faster raises the value of review rather than lowering it.
Hear it at 23:10The weekly job at Plain opens a pull request when a high intent account is being suppressed, and prices the change against all 55,000 accounts first: roughly 200 move into tier 3, 86 into tier 2, 23 into tier 1. A human still has to say yes.
Hear it at 4:42John Kutay runs this at Rippling scale and still starts in files: versioned, readable and consumable by agents. Graduate to a database or a warehouse when concurrent writers make pull requests the bottleneck, which is a real limit and not the starting condition.
Hear it at 49:55Asked in the fireside what three things a non-engineer should go and learn, both answers were the same one: durable tables, unique keys, system design. Without it the stack becomes the thing nobody on the team wants to touch.
Hear it at 1:09:32Nobody demoed a prototype. Each one opened the go-to-market their company runs on, and answered questions about it from the room.
Account scoring and deep enrichment moved out of local folders into a repo, and then the agents started proposing the changes themselves. A weekly GitHub Action reads pipeline data, asks whether a high intent account is being suppressed and whether a signal was missed, and opens a pull request with an impact preview. The one he walked through loosened an engineering density threshold from 30% to 20%, and a few hundred accounts re-tiered overnight once it was merged.
Watch this talk
Three layers: Cargo for the data and the hosting, the ICP and the scoring model in GitHub, and a custom app as the surface reps work from. One client dropped a prioritization tool costing around $50,000 a year because it could not version a rule or tell a rep why an account scored the way it did. The CRM stays the process layer; this is the activity layer on top of it.
Watch this talk
The leak he went after is a positive reply sitting unanswered. Leads land in a Cargo model, a play researches, writes and sends across email and LinkedIn, and when a reply arrives a seven step classification decides whether to draft an answer, which goes to Slack for approval. Built from a single prompt with the Cargo CLI and Claude, and it mailed the room live during the talk.
Watch this talk
Growth OS as he runs it: 2,000 internal monthly users, around a million AI prompts a month, over a billion rows processed a day. Signals are the lifeblood of outbound and raw signals need verification, so a validation layer built around coding agents resolves an event to an account and a contact before it ships as a pipeline. Every signal carries a weight and a decay: three to six months for a job change, fresher for hiring.
Watch this talkWhether the person who owns this arrives as an engineer or as a marketer, and why both routes are working. Why Rippling's outbound volume fell as its best accounts went back to humans, with the systems chasing what those humans find. And when to build the interface yourself rather than bend Salesforce around it.
Watch this talk