---
title: 'In the News: September 16, 2026'
description: "DAIR.AI founder Elvis Saravia says agent coordination begins to collapse after one orchestrator and one executor because agents repeat one another's verification work."
canonical_url: 'https://darkfactory.dev/news/2026-09-16-midday'
markdown_url: 'https://darkfactory.dev/news/2026-09-16-midday.md'
collection: news
date_published: '2026-09-16T13:20:00-04:00'
date_modified: '2026-09-16T13:20:00-04:00'
---

# In the News: September 16, 2026


Elvis Saravia, founder of DAIR.AI and a former Meta AI researcher, says one orchestrator and one executor subagent work best in his harnesses. Adding another subagent causes the setup to collapse, even when he mixes model families.

## 1. Elvis Saravia says agent coordination collapses after one orchestrator and one executor

**[If you build agent harnesses, this is important](https://x.com/omarsar0/status/2100219606405431391)** · Elvis Saravia (@omarsar0), founder, DAIR.AI, formerly Meta AI · X, September 16, 2026

Saravia quote-tweeted a post from Eric Provencher (@pvncher), who wrote: "if you're running more than 2 sub agents at time, you're almost certainly burning tokens for 0 quality gain," adding that "agents don't trust each other enough to avoid double-checking everyone's homework." Saravia said his own experience matches it: "combining one orchestrator agent and one executor subagent typically works best right now," and "if you try, for instance, to add another subagent to the mix, things start to collapse." He said the pattern holds "even when mixing model families." He was also blunt about inflated claims elsewhere on the platform: "when you see someone on X bragging bout their 100+, 2+ levels deep multi-agent system, you almost certainly know it's made up." Read directly about three and a half hours after posting, the thread had drawn 10.2K views, 44 replies, 11 reposts, 71 likes, and 84 bookmarks.

**Why it matters:** Saravia and Provencher both attribute the failure to agents rechecking one another's work. Their accounts give harness builders a specific boundary to test: before adding another executor, measure its effect on result quality and redundant verification.