How Unify cut its AI agent costs 95% in two weeks

<p>Connor Heggie spent his early career on a fifteen-person self-driving startup run like a research lab, then moved to Scale AI&#39;s mapping team, before becoming the co-founder and CTO at Unify. Unify builds agents for go-to-market teams, and for the last two years, one of AI&#39;s big mainstream narratives has been automating away the sales rep entirely. Connor and his team built the opposite: an agent that gives every sales rep &quot;an engineer in their back pocket.&quot; He walks through how Unify&#39;s harness evolved from million-agent batch jobs to a chat product, and unpacks the engineering that makes it cost-effective to run at scale.</p><p>–</p><p>We also discuss:</p><ul><li>How Unify cut 90-95% of costs two weeks before launch</li><li>The 15-requests-per-second ceiling inside OpenAI&#39;s prompt cache</li><li>Why your LLM judge must be a different model family</li><li>The Speed Audit: why one at a time beats a table of 1,000</li><li>Why Unify&#39;s subagents are just a function call</li><li>What working on self-driving taught Connor about running evals</li></ul><p>–</p><p>Timestamps:<br>00:00 Introduction</p><p>01:30 &quot;Go-to-market is a search problem&quot;</p><p>06:10 The old workflow: drag-and-drop nodes over a million-row table</p><p>09:40 Why the harness is similar to a coding agent&#39;s</p><p>10:50 Running durable agents in the cloud without a full VM</p><p>13:15 Giving models pandas-like superpowers over a live table</p><p>15:50 Why Unify&#39;s subagents are just a function call</p><p>18:25 The 15-requests-per-second limit hiding inside OpenAI&#39;s cache</p><p>24:20 Optimizing for prompt caching hit rates</p><p>28:20 Fork versus child subagents</p><p>32:35 The Speed Audit: why one at a time beats a table of 1,000</p><p>37:50 Locking memory to keys instead of letting the agent freestyle</p><p>44:15 Why AI isn’t taking over sales</p><p>49:00 What working on self-driving taught Connor about running evals</p><p>52:46 Why your LLM judge must be a different model family</p><p>54:45 Ditching full VMs for Monty, a Python REPL that suspends</p><p>58:55 Semantic merge sort: why Connor is obsessed with RLMs</p><p>59:53 How Unify cut 90-95% of costs two weeks before launch</p><p>1:01:15 The case for &quot;semantic linters&quot; over skill files</p><p>1:04:55 Why Unify runs mostly on OpenAI, their first investor</p><p>1:06:12 Why 10x cheaper tokens still lose on tool efficiency</p><p>1:07:24 Why open-source models don’t make economic sense (ye