Joe ReisJune 25, 202646m

From Airflow to AI Agents: Maxime Beauchemin on Building Agor and Running a Company with AI Agents

TL;DR

Agor started as Maxime's fix for terminal chaos: After realizing Claude Code made him 2x to 10x faster, Beauchemin built Agor to manage parallel agent sessions, shared environments, and team collaboration instead of juggling tmux tabs, Docker conflicts, and isolated local workflows.
About 80 percent of Preset now uses agents daily: The company runs shared agents for analytics, deal desk, legal, competitive intelligence, and outbound sales, including a Slack-based "data galore" agent connected to Airflow, dbt, source code, and data ops context.
Bug fixing became a visual agent workflow: In Agor, a scheduled bug-bashing agent wakes up hourly, pulls issues from GitHub or Shortcut, creates a branch, tries to reproduce the bug, writes a fix, opens a PR, and hands off only the final review to humans.
Beauchemin passed on an Airflow startup because he was done with data engineering: Even with VCs pushing him toward an Airflow company, he chose Superset because he was tired of ETL after more than a decade and wanted to build colorful, interactive UI and UX products instead.
The new constraint is human context sharing: Beauchemin argues that coding is no longer the main slowdown. Handoffs, reviews, and coordination are. His answer is smaller "one-pizza" teams, shared branch-based environments, and agent sessions that travel with the work from product to engineering to QA.
He thinks AI enablement may be the highest-impact job in companies right now: If one person finds a 10x workflow, their biggest contribution is helping everyone else get there through better tools, MCP setup, governance, and shared systems rather than just chasing their own extra gains.

The Breakdown

Maxime Beauchemin says roughly 80 percent of his company now uses an internal agent command center every day, with AI agents handling everything from analytics and legal review to bug triage and QA. The Airflow and Superset creator explains why he skipped building an Airflow company, how Agor emerged from his own "Claude Code moment," and why the real bottleneck now is human coordination, not coding.