Your AI agent can now build your data pipelines.
Introducing Flights: agent-native data pipelines, built into MotherDuck. Point your agent at the MCP server and describe what you want:
→ deployed Python pipelines in minutes
→ ingest, transform, call an LLM, replicate from a
We're excited to co-host the Real-Time Lakehouse & Agentic AI Meetup on June 15, 2026, together with our partners MotherDuck and Lenses to discuss the future of real-time AI and data infrastructure.
Our speakers include:
Rayees Pasha, CPO, RisingWave
Tun Shwe, AI Lead, Lenses
"Vibe coding is very dangerous and irresponsible." Agentic engineering is not.
@sspaeti interviewed @wesmckinn (creator of pandas, co-creator of Apache Arrow) on how he ships with AI: hours speccing, sub-agents watching for drift, every commit auto-reviewed.
And the open question: how do we grow senior engineers when the hard labour we used to learn from is gone? Wes bets on learning by osmosis, design patterns, architecture
His sharpest line: "When code is free, saying no is our last defense." Every feature is cheap to create and expensive to maintain. Scope, design, and taste are what's left.
Wes runs Claude Code, Codex, and Gemini CLI daily. ~$20k/month at API rates. Trust comes from the process: the superpowers skill framework for specs, and his own AI reviewer Roborev catching what slips through.
Vibe coding = one-prompt it, never look at the code, ship it.
Agentic engineering = hours in the spec phase, the architecture comes from you, sub-agents watch for drift, and every commit gets auto-reviewed before it lands.
"Vibe coding is very dangerous and irresponsible." Agentic engineering is not.
@sspaeti interviewed @wesmckinn (creator of pandas, co-creator of Apache Arrow) on how he ships with AI: hours speccing, sub-agents watching for drift, every commit auto-reviewed.
Is your Postgres slow at answering analytics questions?
It's a familiar wall: Postgres ships the product, then absorbs the dashboards, then the customer-facing analytics tab. Then one day it stops keeping up. Ten-second page loads. API timeouts. Read replicas that don't help.
Multiple DuckDB processes can now write to the same lakehouse table without Spark.
Daniel Beach calls catalog commits "a big deal nobody's talking about" on the new Explain Analyze.
Also: ADBC, Cloudflare's Skipper, and a 40x speedup from an agent in a loop.