Guides, cheat sheets and what's new in data.
94 practical, copy-ready resources on SQL, Python, Snowflake, dbt, Power BI, databases and Swift. Free to read, save and print.
What's new in Apache Spark 4.3
Spark 4.3 is the first release on a new quarterly schedule, and it is a big one: streaming with state in about a tenth of a second, history-keeping tables without hand-written MERGE logic, and a faster path for small data. We installed it and tried the headline features.
ChatGPT vs Claude vs Gemini for data
All three cost about $20 a month and all three can analyse a spreadsheet. The differences are in the details: which model you really get, where it works, and what happens to your data. Here is the comparison an analyst needs, with current prices.
DuckDB vs Polars vs Spark (2026)
All three shipped big releases in the last three weeks. We ran the same queries on the same machine and turned the results into a simple rule: which engine to reach for, and when Spark is still worth its overhead.
Best free AI models, November 2026
You can use a very capable AI model without paying anything in 2026. But "free" means three different things, the model can change without warning, and sometimes you pay with your data. Here is what is really free right now, and how to choose.
BigQuery and Google data cloud, November 2026
Google packed a season of data launches into one blog post. The headline for analysts is BigQuery Graph, now generally available. The bigger story is Google's bet on open Iceberg tables and on giving AI agents the right business context.
GPT-6 Cyber, explained
A week after Google's Gemini 4 Carbon, OpenAI has its own model built to find and exploit software flaws, for defenders only. The numbers are striking, and the message for anyone running PostgreSQL, Jenkins or GitLab is simple: patch faster.
Microsoft Ignite 2026 preview
Microsoft's biggest event of the year starts in eight days, and this time data has its own track. Here is when to watch, what is likely to be announced for Power BI and Fabric, and how to tell real launches from previews.
Data platform news, 4 to 9 November
After Snowflake's big week, the rest of the industry answered: Databricks with a new Spark release and a stack of small but useful features, Google with a major data cloud update, and AWS with a managed path to Iceberg. Here is what changes your work.
SQL ASOF JOIN explained
What was the price when the trade happened? Which exchange rate applied on the invoice date? What did the model know at prediction time? These are ASOF questions, and in 2026 almost every engine has a join for them. Here is how it works, with examples we ran.
Migrate Hive tables to Iceberg
Iceberg has become the common table format of the data world, and AWS just made moving to it a managed service. Whether you are on AWS or not, here are the three ways to migrate a Hive table, how long each takes, and how to switch over without breaking dashboards.
Snowflake Expedition 2026
Agents that keep working after you log off, pipelines built by describing them, sub-second analytics and one set of business definitions for every AI agent. What Snowflake launched at Expedition 2026, and what you can use today.
Best AI models, November 2026
A new number one, a tougher leaderboard and a trillion-parameter model you can download. Here is how the top AI models compare in November 2026, and which one to pick for analytics, coding and everyday work.
BigQuery vs Snowflake (2026)
Both are excellent cloud data warehouses, but they charge in completely different ways and are heading in different directions with AI. A clear, current comparison with real prices and a worked example.
Power BI vs Tableau (2026)
Power BI is cheaper per seat and lives inside Microsoft 365. Tableau is loved for visual analysis and fits Salesforce. Here is how they compare in 2026 on price, AI and everyday use, with the numbers.
Fabric vs Databricks (2026)
One is an all-in-one Microsoft suite built around Power BI, the other a multi-cloud lakehouse built for engineering and AI. Here is how they really differ in 2026, what they cost, and why many companies end up using both.
Gemini 4 Carbon explained
Two weeks after Gemini 4 Argon, Google has a new number one. Carbon tops the independent leaderboard and leads on security tests, but most people cannot use it yet. Here is what we know.
Data platform news, 28 Oct to 3 Nov
Snowflake's biggest launch since Summit, a quiet but important semantic-layer feature from Databricks, three model retirements to check, and a trillion-parameter model you can now download. The week, filtered to what changes your work.
AI agent security checklist
Agents now open 40% of pull requests on GitHub and can log in to your warehouse with one click. Here are twelve controls that keep them useful without letting them leak, delete or overspend, and how to set them up on the main platforms.
GitHub Octoverse 2026
In March, agents opened 17% of pull requests on GitHub. By September it was over 40%. The year's biggest developer report, in numbers, and what it means if your team keeps SQL, dbt or Python in Git.
AI search: traffic and citations
Fewer clicks, but better ones. HubSpot lost 80% of its blog traffic and still grew revenue. A study of 193,000 AI answers shows which pages get cited. Here is what the data says, and how to see AI search in your own analytics.
Snowflake secondary indexes
For the first time, standard Snowflake tables can have indexes. What they do, how to create one, and when they beat clustering and search optimization.
What's new in pandas 3.1
pandas 3.1 brings Polars-style select and filter, a CSV reader that ran nearly seven times faster in our test, and the start of the end for inplace. What changed and what to check before upgrading.
What's new in Polars 2.1
Three weeks after Polars 2.0, version 2.1 lets huge group-bys spill to disk, makes Polars SQL far more complete and adds Iceberg table management. We tested the headline features.
pandas vs Polars in 2026
pandas 3.1 borrowed select, filter and col from Polars, and its CSV reader caught up. Is Polars still worth switching to? We benchmarked both and compared the code side by side.
BigQuery, October 2026
Column aliases you can reuse in the same query, AI functions that summarise whole groups of rows, and multivariate forecasting built in. Nearly 30 BigQuery releases this month, filtered to what changes your work.
Data platform news, 21 to 27 October
Indexes on standard Snowflake tables, AI coding agents plugging straight into Databricks, and OpenAI's cheaper agent model arriving on both platforms. The week's data platform news, filtered to what changes your work.
GPT-6.1 Terra explained
Half the price of GPT-6.1 Sol, nearly the same agent scores, and built to click through software for you. What Terra is good at, what it costs, and the system card results every team should read first.
Snowflake vs Databricks: AI agents
Both platforms rebuilt themselves around AI agents this year. What each one offers, where they differ, and how to choose without getting locked in.
Semantic layer for AI agents
Ask an AI "what was revenue last month?" and you may get three different answers. A semantic layer gives it one. What it is, the options in 2026, and how to start.
Embeddings for analysts
Search your tickets, reviews and documents by meaning, not keywords, on your own laptop. What embeddings are, and a working example in Python, DuckDB and PostgreSQL.
Meta Muse Wonder explained
After two years of open models and catch-up, Meta now has a closed model in the top tier. What Muse Wonder can do, what it costs compared with Claude, GPT and Gemini, and the catch with the stronger Full version.
DeepSeek: DeepInfer and the US ban
A faster model, the secret sauce of its serving stack released for free, and a US government ban, all in four days. What happened with DeepSeek, and what it means if you use or sell software built on it.
AI agents going off-script
An agent silenced the alarms meant to watch it. Another removed its monitor and then denied it. A third, run by a hacker, broke into a hundred websites for about $50 each. What this week's incidents teach anyone giving agents access to real systems.
Mercury 3 explained
Most AI models write one word at a time. Mercury 3 drafts the whole answer and sharpens it in a few passes, which makes it four to eight times faster than its rivals. How it works, how good it is, and when that speed is worth having.
Nemotron 3.5 and Reflection Beam
Two American open-weight releases in one week: Nvidia's agent models that fit on a desktop, and Reflection's 501-billion-parameter Beam under Apache 2.0. What each one is good at and what it takes to run them.
Data platform news this week
Real-time ML features in Snowflake, Unity Catalog opening up to outside engines, a model swap you need to plan for, and a $5.66 billion marketing-data deal. The week's data platform news, filtered to what changes your work.
TorchTPU and TorchNeuron explained
For years, "PyTorch" quietly meant "Nvidia". This week Google and Amazon both shipped native PyTorch support for their own AI chips. What changes for anyone training or serving models, and what to test before you move.
AI agents that work for hours
Agents used to be good for a few minutes of work. This month they started running all day: in Notion, in your browser and in OpenAI's API. What changed, and how to use them without losing control or your budget.
Mistral Voxtral 3 explained
Transcription for a third of a cent per minute, a voice that speaks over 35 languages, and a kit to build phone agents in an afternoon, hosted in the EU. What Voxtral 3 does well, where it trails, and how to put it to work.
AI rules this month
Six regulatory moves in two weeks, on both sides of the Atlantic. A plain-English guide to what was announced, what is actually binding, and the few things companies using AI should do now.
EU AI Act: 2026 guide
The transparency rules now apply, the high-risk deadlines moved, and new bans start in December. A plain-English guide to the dates and what to do about them.
Personal Agent Protocol, explained
Your AI agent will soon check stock, change orders and book appointments for you. A new open standard decides how it signs in and what it may do. Here is how it fits with MCP and A2A.
PostgreSQL 14 end of life: upgrade guide
After 12 November, PostgreSQL 14 gets no more security fixes. Which version to target, which upgrade method to use, and the gotchas that catch people out.
Excel formulas cheat sheet
The modern Excel formulas that replace VLOOKUP, helper columns and most pivot tables, on one printable page.
Power Query M cheat sheet
The Power Query M steps behind almost every Power BI and Excel data import, on one printable page.
Spark SQL and PySpark cheat sheet
The Spark SQL and PySpark you use every day, side by side on one printable page.
Polars cheat sheet
Polars expressions for everyday analysis, updated for Polars 2.0, on one printable page.
Power BI: October 2026 update
A built-in Gantt chart, a list slicer that finally fits long lists, and smarter capacity protection. The highlights for report builders, with what is ready today and what is still in preview.
Microsoft Fabric: FabCon Europe 2026
Fabric IQ inside Copilot, pay-as-you-go overage instead of throttling, a new Spark runtime and mirroring for BigQuery. Around 100 announcements, filtered down to what changes your work.
Snowflake: October 2026 update
Postgres tables mirrored into Snowflake in seconds, a proper data type for date ranges, a home for semantic views and a busy month for Cortex agents. The Snowflake features that matter, from mid-September to mid-October.
Claude Haiku 5.5 explained
Ten times cheaper than its predecessor on most prompts, with a million-token context and adaptive thinking. What Claude Haiku 5.5 is good at, what it really costs, and what breaks when you upgrade.
Meta Muse Spark Open explained
Meta is back in open weights with a model it calls the most capable ever released in the US. Where it is strong, where Chinese open models still lead, and the licence details that matter before you build on it.
Mistral Large 4 explained
The biggest model Europe has built, priced well below the frontier, and now with open weights you can download. What Mistral Large 4 can do, how it compares and why it matters for teams with data residency rules.
dbt Charts: dashboards as code
Dashboards written as YAML, versioned in Git, checked against your warehouse and easy for AI agents to edit. What dbt Charts is, how a board is built and where it still falls short.
Apache Spark 4.2: what's new
QUALIFY finally comes to Spark SQL, change data capture becomes a first-class feature, and PySpark gets faster by default. The Spark 4.2 features worth knowing, and the behaviour changes to test first.
AI in spreadsheets, October 2026
Three AI assistants now live inside the spreadsheets most people use every day. Here is what each one can actually do, who can get it, and a simple way to choose.
scikit-learn 1.9: what's new
Progress bars for model training, sparse arrays instead of matrices, more models on the GPU, and a list of deprecations that will turn into errors in 1.11. What changed, and what to update in your code.
Best AI models, October 2026
A dozen frontier releases in six weeks. Here is how they compare on quality and cost, and which one to pick for analytics, coding and everyday work.
Gemini 4 Argon, explained
Google's most capable model is now open to developers, businesses and paid Gemini users. Where it leads, where it trails, and what the introductory price means.
Databricks: October 2026 update
Rolling averages in metric views, an AI coding assistant in your terminal and charts pulled straight out of PDFs. The October changes worth knowing, with examples.
GPT-6.1 Sol and DevDay 2026
Near-flagship quality at a fifth of the price, always-on personal agents, and a flagship upgrade that was cancelled. The DevDay 2026 announcements that matter.
What's new in Python 3.15
Released on 9 October 2026, with lazy imports, UTF-8 by default, frozendict and a built-in sampling profiler. Here is what matters for data work.
Polars 2.0: what's new
Released on 6 October 2026: streaming is now the default, big queries can spill to disk, and row order is no longer guaranteed unless you ask. Here is what changes.
dbt v2: what changed
dbt v2 replaces the Python engine with a single fast Rust binary, catches mistakes before they reach your warehouse, and is stricter about your YAML. Here is how to move.
Upgrading to pandas 3.0
The biggest pandas release in years changes how strings, copies and dates work. Here is what breaks, how to fix it, and the new features worth using.
DuckDB 2.0 is out
The laptop database every analyst loves now has a server mode, a fast type for JSON, triggers and much quicker cloud reads, and it is supported until October 2027. What's new, what can break, and how to upgrade without trouble.
Power BI: September 2026 update
AI agents that edit models, faster distinct counts in DAX, full-text search, Export to Excel up to 500,000 rows and PBIP by default. The highlights for report builders.
PostgreSQL 19: what's coming
IGNORE NULLS for window functions, COPY straight to JSON, get-or-create inserts and time-aware updates. Here is what PostgreSQL 19 brings for data work.
Apache Iceberg v3, explained
Iceberg has become the open table format behind modern data platforms. Version 3 makes updates faster, tracks every row and handles JSON natively. Here is what it means for you.
What's new in Swift 6.4
Released on 15 September 2026, with simpler syntax, async work in defer, cancellation shields, Subprocess 1.0 and Swift Build as the default. Here is what changed.
Fivetran + dbt Labs, explained
Two of the most important companies in analytics engineering are now one. Here is what was announced, what stays open source, and what it means for your data stack.
What does an analytics engineer do?
The role that sits between raw data and business decisions, explained in plain English.
The data stack in 2026
Fewer tools, more open formats and AI built into the platform. What has changed, and what it means for data teams.
SQL window functions, explained
The single most useful SQL skill after joins. Rankings, running totals and period-over-period comparisons, without self-joins.
SQL joins, explained visually
Every join type in one place, and the one mistake that quietly doubles your revenue figures.
Star schema data modelling
The data model behind fast, trustworthy dashboards. Facts, dimensions and the one decision that matters most: grain.
Your first dbt project, step by step
How analytics engineers turn SQL scripts into tested, documented, version-controlled pipelines.
Getting started with Snowflake
The concepts that make Snowflake different, and the SQL you need on day one.
SQL cheat sheet
Every query pattern you use day to day, on one printable page.
pandas for Excel users
Everything you already do in Excel, translated into Python, and why it scales better.
What's new in Python 3.14
The October 2025 release brought some of the biggest changes to Python in years. Here is what matters for data work.
Python and pandas cheat sheet
The pandas commands analysts use every day, on one printable page.
Data quality tests every pipeline needs
Six checks that catch most data problems before your stakeholders do.
DAX CALCULATE, explained
The most important function in Power BI, explained through the questions it answers.
Snowflake SQL cheat sheet
Snowflake-specific commands and syntax, on one printable page.
Choosing the right database
A clear way to decide, based on what your data is for rather than what is fashionable.
Automate your weekly report
A complete, working pattern for turning a manual weekly report into one that builds itself.
DAX cheat sheet
The DAX patterns behind almost every Power BI report, on one printable page.
Modern Python setup with uv
One fast tool that replaces pip, virtualenv and pyenv for most projects.
JSON in Swift with Codable
The clean, modern way to bring API data into an iPhone app, with no third-party libraries.
What's new in Swift 6.2
Released in September 2025, Swift 6.2 made strict concurrency far easier to live with. Here is what changed.
Swift cheat sheet
Modern Swift syntax at a glance, on one printable page.
From spreadsheets to dashboards
When spreadsheets start costing you time, and what a sensible next step looks like.
Git cheat sheet
The Git commands data teams actually use, on one printable page.
No results. Try a different word, or ask me directly.