This lesson from FDE Interviews covers the unit economics that decide whether an AI deployment ships. The core math: cost per request from input and output tokens, multiplied by volume, compared against the human or manual baseline it replaces. It also explains the API-versus-self-host break-even, noting self-hosted GPUs only win above a real utilization threshold. The page links related course lessons on engagement scoping and pilot constraints, plus practice questions on tokenization, capacity estimation, batch pipelines, and prompt caching.

Key points

• Unit economics gate whether AI deployments ship
• Cost per request = token costs times volume
• Compare against the human baseline being replaced
• Self-hosted GPUs beat APIs only at high utilization
• Includes practice questions on capacity and batch design

• Trading Margin for Moat: Why the Forward Deployed Engineer Is the Hottest Job in Startups

  InsightAndreessen Horowitz2025-06-04EN

  Andreessen Horowitz argues AI startups should embrace implementation-heavy services, like Salesforce, ServiceNow, and Workday did, rather than chase product-led growth and 80% gross margins.

• What I learned analyzing 1K forward deployed engineer jobs

  LearnBloomberry2025-11-18EN

  The author analyzed 1,000 forward deployed engineer job postings from Revealera and found the title covers three distinct roles: Builder FDEs who embed with customers writing production code (60%), rebranded sales…

• AWS invests $1 billion to embed AI forward deployed engineers with customers

  NewsAbout Amazon2026-06-30EN

  AWS is investing $1 billion in a dedicated Forward Deployed Engineering (FDE) organization that embeds engineers and purpose-built AI agents directly inside customer teams.

• Why AI Infrastructure must evolve for Agent Experience — Akshat Bubna, Modal CTO

  InsightLatent Space2026-07-08EN

  In this Latent Space episode, Modal CTO Akshat Bubna argues that cloud infrastructure designed for human developers must evolve for agents, which need fast iteration loops, sandboxes, and programmatic control instead…

• Why AI Compute Comes in Rectangles — Erik Bernhardsson, Modal CEO

  InsightLatent Space2026-10-22EN

  Elastic GPUs, sandbox growth, and the economics of the agent cloud

• What’s the Future of Vertical SaaS in an AGI World? Jamie Cuffe, CEO of Pace

  Podcast / talkSequoia Capital2026-10-28EN

  In this Sequoia Capital podcast, Pace CEO Jamie Cuffe explains how his company acts as an agentic process outsourcer for insurance, replacing traditional BPOs on mission-critical back-office work.