Prompting LLMs with rich first-party context

Sylvain details building prompts: define goals, add product and intent data, and provide an offer library to improve outputs.

Play episode from 31:10

chevron_right

Transcript

chevron_right

Transcript

Episode notes

Listen on Spotify

Sylvain Giuliani started his career in lifecycle marketing, where he learned to code and build systems without relying on engineers. He then joined Pusher as a marketing manager where he worked his way up to CRO, before joining Census in 2020 to rebuild their GTM motion from the ground up. After leaving Census at the end of 2024, Sylvain joined Augment Code as employee number 50. When he joined, Augment was finishing its research phase, and had no self-serve product, signup flow, or billing system. Over the past year, he’s built the entire GTM motion from scratch, and helped scale Augment to hundreds of thousands of users with a lean team by hiring technical generalists who build systems that compound on top of one another.

Subscribe for weekly updates on top GTM Engineering content, open roles & more

In this podcast, we discuss:

* Why AI companies need leaner teams that use AI to eliminate inefficiencies due to AI cost compressing margins

* Hiring technical generalists for Growth from companies with great cultures like Rippling, Datadog, and Ramp, who can write code and build systems that compound over time

* Examples of how Syl and the team are prioritizing and pushing the boundaries of what each individual can accomplish

* Centralizing all data in your warehouse and only buying tools that can flow data back to it

* The advantages of analytics and ops reporting into GTM

Episode highlights:

* Most sales bottlenecks stem from reps lacking knowledge at their fingertips and having to do manual work to multi thread into the account. Augment automates account-level ABM campaigns directly from Salesforce, then uses AI to prep reps with key context before calls. This gives reps everything they need to answer questions without relying on solutions engineers, collapsing five meetings into one and shortening large deal cycles to sixty days.

* During interviews, Sylvain borrows Datadog’s approach of presenting candidates with seemingly impossible challenges the company hasn’t solved yet. Rather than seeking the answer, he watches for people whose eyes light up and whose gears start spinning when faced with ambitious problems. This filters for candidates who challenge conventional thinking rather than defaulting to how things were done at their previous company, which is critical as Augment scales while remaining lean.

* Augment’s data architecture prioritizes control and speed by centralizing all data in their warehouse and not hesitating to buy software instead of build in-house. They ensure every tool they use can send data back to their data warehouse, allowing the team to swap email providers or AI personalization tools as they see fit without the risk of losing historical context. This prevents vendor lock-in, so when a tool doesn’t support a growing need or costs get too high, Augment can easily swap vendors or build critical workflows in-house.

* Sylvain positions analytics directly under GTM rather than finance because of the cross pollination benefits. When data teams report elsewhere, they often prioritize accuracy over speed, but when the data team sits close to GTM, they naturally identify learnings and proactively flag opportunities. This creates a tight feedback loop where data teams discover actionable patterns like usage spikes, and reps learn which signals actually convert to pipeline.

Where to find Sylvain:

* LinkedIn

Transcript details:

(00:00) Introduction and background

(02:52) Joining Augment and early challenges

(04:12) Building the team and scaling

(05:47) Structuring teams for AI-native pace

(08:51) Why predictable revenue is broken

(12:05) The $10M per rep challenge and efficiency gains

(15:59) Hiring people who challenge conventional thinking

(17:15) Finding talent at culture-fit companies

(20:18) Warehouse-first data architecture