Summary

Simon Späti interviews Maxime Beauchemin—creator of Airflow and Superset—about how he uses AI agents in data engineering and running Preset. The central artifact is Agor, a 'Figma for agents' canvas that gives teams a shared, visual workspace for coordinating AI sessions instead of everyone running isolated local prompts. Beauchemin argues that the real bottleneck in agentic workflows is no longer model capability but context quality, coordination overhead, and security—framed as the need for 'Okta for Agents.' He also revisits the semantic layer as a structured context mechanism for agents, and introduces db-agents as a way to embed AGENTS.md documentation directly inside databases. The interview concludes that Amdahl's Law and human review cycles are now the primary speed constraints on agentic development teams.

Key Insight

The limiting factor in agentic data engineering is no longer model capability but the infrastructure around agents—shared context, identity/security, and human review throughput—and teams that solve these coordination problems first will capture the real productivity gains.

Spicy Quotes (click to share)

• 5

  The goal is to automate most of the automatable CEO-stuff.

• 7

  I haven't written a function by hand for a long time, and I might not anymore — except when I feel nostalgic.

• 6

  There are more agents than humans nowadays. Everyone has a Claude Max plan, and agents handle almost all code writing.

• 5

  I trust agents the same way as I would an employee.

• 4

  I let the workers run in god mode, but using dedicated environments/sandboxes, hooked to a dedicated git worktree repo, it can run autonomously and solve problems on an initial prompt, visualized in a shared canvas style.

• 3

  I see two different semantics: the semantic layer and the YAML. There are the hard constraints — not every area needs that strictness — and then the softer ones with Markdown and Agentic Skills, good for 80-90% but with no guarantees.

• 7

  A high Yap-to-Ship ratio would mean lots of human interaction, which is the clear new bottleneck.

Tone

practical and exploratory, grounded in real production experience, with cautious optimism about agentic workflows tempered by hard-won awareness of security and coordination limits