Patrick Muller

Senior Data Engineer @ AWS | 20+ yrs building data pipelines powering AI/ML | Hadoop, Spark, Python | Sharing my AI journey | Side project:

[Patrick Muller @patrickmullerAI]

One prompt. 30 minutes. A whole portfolio website.

I gave Opus 5.5 a single prompt describing the site I wanted, and half an hour later https://t.co/FQsKytfMwn existed — layout, copy, project pages, everything.

The progress AI is making is amazing. And honestly? A little scary too.

[Patrick Muller @patrickmullerAI]

You can now describe an Airflow pipeline in plain language and have your AI coding assistant build it. MWAA just got AI agent skills.

Amazon Managed Workflows for Apache Airflow now integrates with the Agent Toolkit for AWS to provide built-in AI agent skills that let you author, debug, and test MWAA workflows — and upgrade MWAA environments — using any AI coding assistant that supports agent skills.

Four skills ship:

- authoring-mwaa-workflow
- debugging-mwaa-workflow
- testing-mwaa-workflow
- upgrading-mwaa-environments

Together they let your assistant interact directly with MWAA in natural language. The canonical example: describe an ETL pipeline in plain language, and the authoring skill generates a working DAG complete with task dependencies and error handling.

They work with both MWAA Provisioned and MWAA Serverless, in all Regions where MWAA is available. To get started, install the aws-data-analytics plugin and configure your AI coding assistant to use the MWAA agent skills.

Here's my take: DAG authoring has always been the tedious part of Airflow — the boilerplate, the dependency wiring, the error-handling patterns you rewrite every time. Having the assistant generate a working DAG from a description is nice, but the real value is the full loop: author, debug, test, upgrade. Debugging a failing DAG is where data engineers actually burn hours — tracing a failed task through logs, figuring out whether it's the task, the dependency, or the environment. If the debugging skill can do that investigation against live MWAA state, that's the skill that earns its keep. And the upgrade skill matters more than it sounds: MWAA version upgrades are the kind of work everyone postpones because they're fiddly and risky, and an assistant that knows the upgrade path removes the excuse.

If you run Airflow on AWS, this is worth trying on your next DAG.

Full announcement: https://t.co/tsnGjlKA2S
Find me on LinkedIn: https://t.co/qKCx8WxLb0

[Patrick Muller @patrickmullerAI]

EMR now checks data quality natively in Spark jobs — no separate validation infra. Same DQDL + APIs as Glue, 30+ rule types, ML anomaly detection, fail / filter / quarantine bad rows. https://t.co/n83Yn07OqM https://t.co/wpcMG8RvuT

[Patrick Muller @patrickmullerAI]

TwelveLabs Pegasus 1.5 is on Bedrock — video-to-text that generates structured, time-coded metadata. Define what matters (editorial segments, speaker changes, sports plays, brand appearances) and it finds the moments with timestamps. Single pass, no pre-indexing pipeline. Video in S3 you can't search just became a candidate for this.
https://t.co/w6Z4ig3Qfr
https://t.co/qKCx8WxLb0

[Patrick Muller @patrickmullerAI]

Bedrock now supports reasoning summaries for OpenAI models — a human-readable summary of how the model approached the task, returned alongside the answer via the Responses API. Reasoning traces are where agent debugging actually happens: wrong answer + visible approach = fixable problem. Available in all regions with OpenAI GPT models.
https://t.co/QZ8XnycH9s
https://t.co/qKCx8Wxdls

[Patrick Muller @patrickmullerAI]

@steezyasfck @condzxyz @ottodoti E qual o intuito de postar mais ? Por curiosidade mesmo, tenho pensado nisso tmb, mas não sei qual os ganhos

[Patrick Muller @patrickmullerAI]

@binhara https://t.co/UFgixLoznr

[Patrick Muller @patrickmullerAI]

Cheguei atrasado na eleição da Bolha Dev https://t.co/SVAPWW3KUQ, mas cheguei com presente: fiz o Commit Fighter com um único prompt no Opus 5.5 🥊

O processo foi simples: um prompt pra construir o jogo + algumas trocas de ideia com o Grok pra criar os personagens e as características deles.

🎮 Jogue: https://t.co/wF8osIvNua
💻 Código: https://t.co/EQhXqMmjmN

Confere o repo e sinta-se à vontade pra dar feedback!

Parabéns ao presidente eleito @manodeyvin e a todos os candidatos: @RealGalego @oraulsena @samsantosb @lucas_montano @sseraphini @AkitaOnRails @oviniciuslana @filipedeschamps @claudeai @gabrielbuzziv

[Patrick Muller @patrickmullerAI]

@binhara Esqueceu o “e façam eles lutar”

[Patrick Muller @patrickmullerAI]

Pedi pro Grok olhar as contas dos top 10 da eleição, e montar um personagem, abaixo a descrição que o Grok montou:

8. Vini Lana (@oviniciuslana)

Fundador da AI Coders Academy, canal com dezenas de milhares de inscritos.

Normal: Aula ao vivo. Um cursor gigante aponta o ponto fraco. O próximo hit nessa hitbox dá dano dobrado, mas só se ele narrar (“ó, repara aqui”).

Special: Academia. Invoca três alunos. Eles copiam o último golpe dele, com delay e um bug cada.

[Google Labs @GoogleLabs]

🚨NEW EXPERIMENT 🚨

Playground is an experimental gaming platform that lets you create your own games with zero coding experience. If you can think it, you can play it.

Go to https://t.co/fRUukpuoqM to learn more! Available to users 18+ in the US.

[Vice President JD Vance @VP]

Vice President JD Vance breaks down the data:

"If you are working as an H-1B visa in the United States of America, you’re earning $20,000 less than an American citizen hired in the same position. If you were brought in by one of these foreign outsourcing firms, you’re making $48,000 less than an American citizen in a similar position. So I’m sure that everybody here understands the game that is being played. You bring in indentured servants from outside the country, you lay off American workers and, if you’re a corporation, you make a ton of money by under cutting the wages of American workers, replacing them with people who frankly shouldn’t be in the United States of America to begin with."

[Breaking911 @Breaking911]

VP VANCE: "There has been no company in the United States, unfortunately, that has abused this system more than Microsoft."

Microsoft laid off 6,000 American workers in 2025 after certifying that it could not find qualified American employees — yet the company was approved for more than 6,000 H-1Bs and filed 3,682 PERM applications, including nearly 1,000 applications for the same positions as the laid off American workers.

[Patrick Muller @patrickmullerAI]

@samsantosb Olha ali no screenshot que tirei, na parte de baixo. Vou pegar os feedback e implementer depois

[Patrick Muller @patrickmullerAI]

Cheguei atrasado na eleição da Bolha Dev https://t.co/SVAPWW3KUQ, mas cheguei com presente: fiz o Commit Fighter com um único prompt no Opus 5.5 🥊

O processo foi simples: um prompt pra construir o jogo + algumas trocas de ideia com o Grok pra criar os personagens e as características deles.

🎮 Jogue: https://t.co/wF8osIvNua
💻 Código: https://t.co/EQhXqMmjmN

Confere o repo e sinta-se à vontade pra dar feedback!

Parabéns ao presidente eleito @manodeyvin e a todos os candidatos: @RealGalego @oraulsena @samsantosb @lucas_montano @sseraphini @AkitaOnRails @oviniciuslana @filipedeschamps @claudeai @gabrielbuzziv

[Patrick Muller @patrickmullerAI]

New role, new homework.

I just joined AWS as a Senior Solutions Architect – Redshift Specialist, so I read everything Redshift shipped in the last two months. Top 3 first, then all 14:

**Top 3**
• Cross-Region data lake queries (GA, Oct 1) — S3 tables in another Region, no replication, traffic stays in VPC
• Agent Toolkit for AWS (Aug 27) — Redshift skills for AI-assisted warehouse work via MCP, no extra charge
• Apache Iceberg v3 read/write (GA, Aug 31) — default column values, row lineage, deletion vectors

**The rest**
• DEBUG permission for secure logging (GA, Sep 30)
• Cluster patch 205 (Sep 16) — Iceberg MVs (incremental refresh), post-quantum TLS
• ChatGPT Work analytics plugin (~Sep 10)
• RG instances in Zurich (GA, Sep 10)
• rg.large single-node clusters (GA, ~Sep 2)
• Identity Center auth + enhanced VPC routing (GA, Aug 31)
• KDS 10 MiB ingestion (GA, Aug 27)
• Patch 203 (Aug 25) — SHOW GRANTS FOR GROUP; Python UDF end-of-support begins
• System tables to S3 Tables (GA, ~Aug 21) — 25+ SYS views as Iceberg
• rg.large + rg.12xlarge in GovCloud (Aug 13)
• Patch 204 (Aug 11–Sep 22) — login lockout, zero-ETL metrics

Full details with sources: https://t.co/5AFbJbZx7L

https://t.co/qKCx8Wxdls
https://t.co/MSIveDdeNE
https://t.co/LhREgTEQxD

[Patrick Muller @patrickmullerAI]

The Iceberg V3 wave on AWS keeps rolling — and this one fills the operational gaps.

The AWS Glue Data Catalog now supports table optimization, statistics, and crawlers for Apache Iceberg V3 tables. In practice: you can auto-compact V3 tables with binpack, sort, or z-order strategies, expire old snapshots, and clean up orphan files to cut storage costs. The optimizations understand V3's new data types — variant, geospatial, nanosecond-precision timestamps.

Two things worth calling out. First, Glue can now generate number-of-distinct-values (NDV) statistics for V3 tables, which is what query engines use to build good plans — without statistics, even the best engine is guessing. Second, Glue crawlers can discover V3 tables in S3 and register them in the catalog automatically, so they're queryable from any V3-compatible engine.

This lands right after S3 Tables added full Iceberg V3 data type support. The pattern is clear: AWS is making V3 a first-class citizen across the whole lakehouse stack. If you're evaluating a V3 migration, the tooling is catching up fast: https://t.co/hv6lMRV8f2
https://t.co/qKCx8WxLb0

[Patrick Muller @patrickmullerAI]

Fall is my month of change.

Every September/October, the trees turn their colors and let go of their leaves — and somehow, every big turn in my life has happened in this same season.

It's when I moved to the US. It's when I started at Amazon — 8 years ago this month. It's when I moved from Virginia to Washington. And now it's when I'm moving back to Virginia.

Today, I'm returning to AWS as a Senior Solutions Architect – Redshift Specialist. I'm really excited about what's ahead, and honestly a little sad to leave the team I've been part of for so long.

But that's the thing about fall — the leaves change colors because they're meant to. Every ending makes room for the next season.

Here's to new beginnings. 🍂

[Patrick Muller @patrickmullerAI]

S3 Tables now supports all Apache Iceberg V3 data types — and the practical ones are geometry, geography, and nanosecond timestamps.

What that means in plain terms: you can store geospatial coordinates (points, lines, polygons) natively instead of encoding them as strings, so fleet tracking and asset mapping workloads can filter on location at query time. And telemetry/financial workloads get nanosecond-precision timestamps at source precision instead of shoehorning them into integers.

Plus column default values: add a column and existing rows get populated automatically, no backfill job. Anyone who's run a backfill over billions of rows will appreciate that one.

This builds on the existing V3 support (Variant type, deletion vectors, row lineage), and S3 Tables keeps doing automatic maintenance and compaction so the V3 tables stay fast as they scale.

Available everywhere S3 Tables is. Small launch, big quality-of-life improvement for anyone on Iceberg V3.

https://t.co/yvYu8CvQYH
https://t.co/qKCx8Wxdls

[Patrick Muller @patrickmullerAI]

This one's for the data engineers: Aurora PostgreSQL can now query Iceberg and Parquet data directly.

You create PostgreSQL foreign tables pointing at your Iceberg or Parquet data in S3, S3 Tables, or the Glue Data Catalog — and query it with your existing PG apps and BI tools. No ETL pipelines, no data duplication. Aurora runs the query through an embedded DuckDB engine under the hood.

Why this matters: the classic pattern was copying lake data into Aurora to join it with operational data — more pipelines, more cost, more schema-drift pain. Now the join happens in place. And for latency-sensitive workloads, you can materialize the lake data into native Aurora tables with plain SQL, still no ETL pipeline.

It also federates external Iceberg REST Catalog-compatible catalogs through Glue, so you're not locked to one catalog.

GA on Aurora PostgreSQL 17.11 and 18.6+, all commercial regions plus GovCloud, no additional charge. If you've been maintaining copy-pipelines just to join lake data with transactional data, this is worth a serious look.

https://t.co/mxUAGVNjD6
https://t.co/qKCx8WxLb0

[Patrick Muller @patrickmullerAI]

@spacedevin This is for my daughter’s company. The website was built in just one day using Opus 5.5.

https://t.co/FdAKXzdF6e