# Service First, Software Later

**Track:** Discovery & Validation at AI Speed — AI for Entrepreneurship — complete (29)
**Framework / surface:** venture strategy
**Level:** Intermediate
**Prerequisites:** Validate Before You Build

**In one line:** Services-shaped revenue is not failure — if every engagement leaves behind workflows, data, evals, and software.

## Theory, aesthetics & inspiration

Much of what works in AI entrepreneurship does not begin looking like software at all: it begins as a human expert plus an AI workflow plus a proprietary process, embedded in a customer's operation — the pattern institutionalized by the forward-deployed engineer, the Palantir-coined role now common across AI startups. Investors learned to discount "services-shaped revenue" for good reason — Casado and Bornstein's 2020 a16z analysis showed AI companies' margins dragged toward services economics — but the discount misses a distinction that decides ventures: whether each engagement compounds into reusable assets. A service delivery that leaves behind workflows, evals, fine-tuning corpora, and eventually product is a software company being assembled in the field; one that leaves behind only invoices is consulting with extra steps. The exponential makes the first kind a strategy rather than a compromise, because the manual expertise you sell this year defines the job, generates the data, and writes the spec for the software that delivers it next year. The test is not the revenue's shape but its residue.

**Founder question:** Does each service engagement leave behind software, data, and evals — or just invoices?