Why did Ramp build Laminar instead of using a managed agents API?

Three reasons in Kronick's account: the harness had to be model and provider agnostic, and a single lab's managed API would lock them in; building alongside product teams in the same company, Slack, and codebases ships faster than any vendor; and they wanted to own observability, down to a CTO who prints agent traces on paper to find waste. His caveat is that you still need good engineers on call and keeping the platform healthy.

What does the build-versus-buy memo leave out?

The landing column. It compares engineer-weeks against license fees, two cells in the build column, and the Landing Ledger shows that column collapsed while landing held flat. The memo never prices who adopts the platform, who reads its traces, or who is paged when it breaks, which is what decided Laminar's outcome.

What does Ramp's own sequence say about buying first?

Ask Ramp, the support chat agent, ran on a vendor solution until it outgrew it, when the team wanted tighter integration and lower latency. Buy first, learn what you need from the vendor's constraints, then build when the vendor is the constraint rather than the budget. The three lines can be written at any moment, including while you are still on the vendor.

Build or Buy an AI Agent Harness? Three Lines First

Should you build or buy your AI agent harness?

THE SHORT ANSWER

Decide it from the landing column, not the build column. Sam Kronick's October 20, 2026 post on Ramp Engineering says Ramp built its harness, Laminar, in six weeks, that five months later every agentic product at Ramp runs on it, and that coding agents made building and maintaining a platform marginal next to a vendor dependency. True, and that is the column of the Landing Ledger that collapsed. Laminar landed because Ramp's product teams were its customers from day one, in the next chair. So write three lines before anyone estimates engineer-weeks: the first internal customer by name and where they sit, the person who reads the agent traces every week, and the on-call roster. If any line is blank, buy. If all three are filled, build. My April audit of 39 agents found 13 orphans nothing read and a named owner as the biggest predictor of survival, which is the same lesson at fleet scale.

Six weeks is the build number, and the build number stopped deciding anything. The number that matters is how many products run on the harness five months later, and Ramp's answer to that one is all of them, for a reason the memo would never show: the customers sat in the next chair.

The three lines, the ledger argument, and the fleet audit are in Ramp Built Its Harness in Six Weeks. Kill the Build-Versus-Buy Memo.. The twelve-row comparison of build cost and landing cost is The Landing Ledger: What It Costs to Build vs What It Costs to Land, and the orphan count comes from 39 PM AI Agents Deployed: What Stuck, What Died, and Why.

Last reviewed 2026-10-21 · 1 min read

Should you build or buy your AI agent harness?

SOURCES

THE LONG VERSION

RELATED ANSWERS