One AI to Paint, One to File, One to ... Burn Your House Down (Maybe)

Adobe flexes Firefly, Microsoft builds an Agent army, and a robot that can do the laundry

Apr 28, 2025

Fellow Data Tinkerers!

Before we get to this week’s round up of all things data, just wanted to mention the referral program with the new subscribers. Based on the number of people you share Data Tinkerer with, you can unlock the following rewards:

1 share - a PDF of top 15 articles on Data Tinkerer

2 shares - 100+ cheat sheets for all things data. From Python, SQL, Power BI to Tableau, Spark, Git and many more

4 shares - 100+ resources to learn all things data (science, engineering, analysis). It includes videos, courses, projects and can be filtered by skill or whether it’s free or paid

So if you enjoy reading Data Tinkerer, share with friends and earn referral rewards!

Refer a friend

Now, with that out of the way, let’s get to this week’s news round up on all things data and AI

The Buzz 🐝

Adobe just leveled up Firefly at MAX London 2025: one platform to crank out images, videos, audio, and vectors without juggling different tools. New Image Models 4 and 4 Ultra bring sharper realism and more control, and they threw in Firefly Boards for team brainstorming. It’s Adobe's way of saying, "You’re stuck with us but hey, at least we are more slick now".

Microsoft just gave Copilot a full upgrade kit. There’s Copilot Researcher to dig through Slack, Jira, and your mess of files, Copilot Analyst to analyse your messy spreadsheet and Copilot Notebook for project brainstorming. Plus, they opened an Agent Store to find and use a range of different agents.

Physical Intelligence, a robotics company, introduced their new model (π0.5). It can now handle doing tasks in new spaces without melting down. The goal is to create a robot that doesn’t need a three-hour orientation to find your dishwasher i.e. a machine that's less high-maintenance than a new hire.
(Ah yes, put the towel on the stove buddy. What could possibly go wrong?)

Hey maybe after all, this is coming true:

Data Science & AI

How Walmart Automated 400+ Forecasts and Cut Runtime by Half
No model works for every SKU. Walmart's solution? An autotuning framework with rolling validation, automated feature engineering, and real-time model selection. If you want to learn more about it, check the article!

Claude Code: Best practices for agentic coding
Anthropic shared best practices for using Claude Code effectively, like adding a CLAUDE.md file to explain your project and writing tighter, goal-focused prompts.
From Junior to Senior Data Scientist: 5 Traits That Accelerated My Career Growth
Andres Vourakis
shares some good tips on how to be progress in your career by being more valuable to the business

Data Engineering

Apache Airflow 3
Apache Airflow 3.0 just landed with a React-based UI, DAG versioning, and event-driven scheduling. It’s now built to handle everything from ETL to MLOps to GenAI workflows
Now you can curse at this more slick version when your pipeline fails (Source: Airflow)
How Airtable Made Archive Validation Work at Petabyte Scale
Learn how Airtable archived half of its storage without validation problems

I spent 5 hours understanding how Uber built their ETL pipelines
Vu Trinh
gives a great breakdown of Uber’s ETL pipelines

Data Analysis and Visualisation

The Dashboard Fallacy: When Seeing Everything Means Understanding Nothing
Dashboards look smart. Most don’t help you think. This post breaks down the dashboard trap and why more data ≠ better decisions.