ssp.sh

Some more related notes mentioned here: - Vim Motions for Writers: www.ssp.sh/brain/vim-mo... - Zen Mode for Writing: www.ssp.sh/brain/zen-mo... - Obsidian: www.ssp.sh/brain/obsidian - AI use: www.ssp.sh/brain/using-... I hope that is useful. Looking forward to your way of wri…

Keep AI Out of Your (Obsidian) Vault ssp.sh

www.ssp.sh/ai/ :)

AI Usage ssp.sh

> Yes. And give them that insight nugget. Now the AI is useful for brainstorming. Find others' opinions over the years, summarized and curated at «Will AI Replace Human Thinking? The Case for Writing and Coding Manually». www.ssp.sh/brain/will-...

Will AI Replace Human Thinking? The Case for Writing and Coding Manually ssp.sh

Obviously, not every task requires a level-three agent with extensive setup and integrated tooling for extended data engineering. But the moment that output touches production, you want the deterministic core underneath it, not just a model that sounds confident. www.altimate.…

Where AI Agents Belong in Data Engineering: The Correctness Layer altimate.ai

> when smartphones were everywhere, we needed to put more interesting things in front of the camera and have more interesting lives. I won't change my writing style (just yet, and maybe I do subconsciously), but I will still use these styles because I just like them or they fi…

Should I Change My Writing Style to Short Stories, because of AI and Low Attention Span? ssp.sh

I wrote more at www.ssp.sh/blog/vector-....

Vector Technologies for AI: Extending Your Existing Data Stack ssp.sh

But the moment you go from one agent to five, the orchestration problem shows up exactly like it did with data pipelines a decade ago. The history of DE is becoming the playbook for agent engineering. Worth paying attention to which patterns transfer and which do not. www.ssp.…

AI Orchestrators ssp.sh

Data engineering spent fifteen years building orchestrators — Airflow, Dagster, Prefect, Kestra. We are now doing the same thing for AI agents. Agor from Max the original creator of Airflow. Agent Teams from Anthropic, native to Claude Code. Omnigent from Databricks. Superset-sh, herdr, Gastown.

Automating my newsletter with the latest updates from my second brain, new essays, books I'm reading, notes, and latest social media post (pulled with DuckDB from Bluesky). No Gen-AI, all direct fetched from local MD files or Firehose via Python (linked), integrated with Listmonk and Neomd (email).

In 2025, AI in data engineering was mostly slop on top of slop. The agents themselves are great. Almost every wrapper above them was not. For 2026, my prediction and hope is the same thing: less hype, more value. Stop putting AI in the name. Treat it as a tool, not the product.

I've asked myself where AI agents belong in data engineering work, specifically. Turns out there are 3 levels of "help" (chat, autonomous, dedicated tools)—but the real unlock is a 4th layer: a deterministic core underneath the LLM that proves correctness. Here's the breakdown.

Don't overrate how you do it; it matters more what you do, especially with LLM models. It's just a tool.

Crazy times we live in. I have manually curated knowledge over the years, and when people see it, some default to it being generated with AI, as they cannot comprehend doing something by hand anymore (or else?). Not sure is the right direction..