David Jayatillake
@jayatillake.bsky.social
Writer @ davidsj.substack.com
How big should your data team be? Data teams are often oversized. A company of 200 people rarely needs 15+ data staff, usually 5% of org size is enough dataactionmentor.com/knowledge-ba...
How big should your data team be?
Founders and CEOs are wondering if their data function is bloated and if they should replace everyone with AI agents. Data Leaders are scrambling to defend why they need a 15-people data team in a 200...
dataactionmentor.com
The amount you love someone is proportional to how often you Ghiblify their pictures.
This week I look at agents. I think this is a new way to build where we don’t intentionally build code-based software. open.substack.com/pub/davidsj/...
BERT and ERNIE! 😂 tracking.tldrnewsletter.com/CL0/https:%2...
China's biggest public AI drop since DeepSeek, Baidu's open source Ernie, is about to hit the market
Chinese internet search giant Baidu will open source its Ernie gen AI large language model as soon as this week, with uncertain consequences for the market.
tracking.tldrnewsletter.com
I don't usually share photos of my family on social media for good reason, but I'm happy to share these ones!
This post encapsulates how I feel about the current state of LLMs and doomers etc. Really great read: fly.io/blog/youre-a...
My AI Skeptic Friends Are All Nuts
My smartest friends have bananas arguments about LLM coding.
fly.io
It is possible to build machine learning systems which punch up instead of punching down.
nitasha tiku@nitasha.bsky.social · last yr.A lot of people say generative AI shouldn't infringe on copyright. These researchers actually tried to do it. The result: an 8 terabyte dataset of text that's openly licensed or in the public domain & 7 B parameter model that performs as well as Meta's Llama 7B www.washingtonpost.com/politics/202...
Got a cool story about something in the data engineering space? You should 💯 submit it as a talk to Current 2025 in New Orleans 😁 Do it! Now! CfP is open until 15th June. sessionize.com/current-2025... (Pro-tip: you only need an abstract at this point; writing the talk can be later 😅) #dataBS
At the London Data Practitioners Meetup with @pedramnavid.com @jayatillake.bsky.social @rittmananalytics.bsky.social and the London Dagster community
I still think this is the biggest prize in AI. If Siri could actually do most things you do on a phone manually... 9to5mac.com/2025/04/22/s...
Siri’s new boss is already making big internal changes, per report - 9to5Mac
Siri’s new boss at Apple, Mike Rockwell, has reportedly wasted no time making big changes internally to the people building its assistant.
9to5mac.com
Has anyone tried Llama 4 Maverick yet? How big a machine does it need to run locally? @simonwillison.net
Looks like Nintendo became the best at console FPS.
Once again, we've devised a derogatory name for something many of us are doing: "Vibe coding". Just like "Citizen Data Scientist", "Excel Data Analyst", and many other terms made to belittle by the supposed true artisans that came before. open.substack.com/pub/davidsj/...
Vibe coder
Free like a puppy
open.substack.com
I enabled the Cube AI API Slack App today, and I found a little treat 👀 Shout-out @mjirv.bsky.social and @jayatillake.bsky.social! If you're curious how easy this is to set up, check out: cube.dev/blog/make-yo...
A couple of weeks ago I wrote a post about @pola.rs Cloud, and as coincidence would have it, DeepSeek then releases smallpond - a way to use @duckdb.org scaled out. It's pretty cool that an org has managed to build this, but is it the right way? davidsj.substack.com/p/many-ducks...
Many ducks, a big pond
Splitting the query
davidsj.substack.com
🗣️ It was a real honour to speak at the very first Forward Data Conference in Paris last November, not to mention being introduced with a special piano verse just for me! 🎹 www.youtube.com/watch?v=QNQW...
AI on Data - snake oil or actually useful?
YouTube video by ForwardDataConf
youtube.com
🏭 @tobikodata.com released a great new feature for SQLMesh last week called blueprinting. This is a feature that the dbt Labs community has been asking for for ages, and I’ve been wanting it too! open.substack.com/pub/davidsj/...
Model Generation
A big feature release by SQLMesh
open.substack.com
🦆🦆🦆🦆 everywhere these days count.co/blog/announc...
Keep canvases moving with DuckDB on the server
We're super excited to announce the general availability of DuckDB on the server. This is our way of dramatically speeding up query speed while reducing computer cost for customers, helping you drive ...
count.co
New episode: “Semantic Layers: The Missing Link Between AI and Data” with @jayatillake.bsky.social . We discuss how semantic layers bridge raw data and AI, achieving 100% accuracy for natural language queries, and what’s next for LLM-powered data pipelines. 🎧 https://techontherocks.show/14 🎧
Tech on the Rocks | Semantic Layers: The Missing Link Between AI and Data with David Jayatillake from Cube
In this episode, we chat with David Jayatillake, VP of AI at Cube, about semantic layers and their crucial role in making AI work reliably with data. We explore how semantic layers act as a bridge ...
techontherocks.show
Yep! And the new(ish) rust-based data engine also brought the ability to do multi-stage calculations. Bonus - "Tesseract" is an amazing name for a Cube engine 😀 cube.dev/blog/introdu...
Introducing Next-Generation Data Modeling Engine - Cube Blog
Previewing multi-stage calculations and performance leaps in the SQL Planner
cube.dev
pola.rs/posts/polars... This so reminds me of @motherduck.com and @duckdb.org, I guess the open source Polars is as good as it can be on one machine, and Polars Cloud is as good as it can be on many.
Polars Cloud; the distributed Cloud Architecture to run Polars anywhere
DataFrames for the new era
pola.rs
Cancer incidence rises when people live longer and have access to preventive care to diagnose it in time to treat it, and effective treatment increases cancer prevalence because we count the people who are still alive – this is a good thing, and I'm so fucking tired.
🚀 Exciting times in the data transformation space! Recently, dbt Labs acquired SDF Labs, while @tobikodata.com acquired Quary (YC W24)—both promising Rust-based frameworks. So is SDF better than SQLMesh? No. open.substack.com/pub/davidsj/...
SDF
Is it better than SQLMesh?
open.substack.com
One question that comes up for me on calls with companies- why can't I just build a semantic layer in my data platform? My brilliant coworker Morgan wrote a great post on why @cubedev.bsky.social is more beneficial as a semantic layer. Enjoy! cube.dev/blog/why-cub...
Why Cube Store is the Best Choice for Storing Pre-Aggregated Data - Cube Blog
cube.dev
