Alex Strick van Linschoten
@strickvl.bsky.social
ML Engineer (@ ZenML), researcher (& author of a few books).
Pinned
🤔 Do you ever wonder how companies are putting LLMs and GenAI apps into production? What stacks do they use? What architecture did they go with?
I put together a database of known public technical writeups with summaries of the key technical features.
I put together a database of known public technical writeups with summaries of the key technical features.
These two are at the top of my reading pile for the coming weeks. Happy to get them in the post today from Manning after a little wait!
January 30, 2026 at 3:01 PM
These two are at the top of my reading pile for the coming weeks. Happy to get them in the post today from Manning after a little wait!
Yesterday I was skeptical about dynamic pipelines. Today: where they actually make sense.
January 30, 2026 at 10:00 AM
Yesterday I was skeptical about dynamic pipelines. Today: where they actually make sense.
I've been thinking about dynamic pipelines lately, where the DAG structure emerges at runtime instead of being defined upfront.
January 29, 2026 at 3:31 PM
I've been thinking about dynamic pipelines lately, where the DAG structure emerges at runtime instead of being defined upfront.
As promised yesterday — here's the demo of the two ZenML MCP apps I built the day Anthropic launched MCP Apps.
What you're looking at:
What you're looking at:
January 28, 2026 at 1:44 PM
As promised yesterday — here's the demo of the two ZenML MCP apps I built the day Anthropic launched MCP Apps.
What you're looking at:
What you're looking at:
I spent part of today building two ZenML MCP apps on the day Anthropic launched MCP Apps — the first official MCP extension. Interactive UIs that render inside the conversation.
Day-one notes on what it's actually like to build one:
Day-one notes on what it's actually like to build one:
January 27, 2026 at 5:39 PM
I spent part of today building two ZenML MCP apps on the day Anthropic launched MCP Apps — the first official MCP extension. Interactive UIs that render inside the conversation.
Day-one notes on what it's actually like to build one:
Day-one notes on what it's actually like to build one:
ZenML just shipped our first Agent Skill: Quick Wins.
I originally wrote our Quick Wins docs page as a collection of 15 things you can do in ~5 minutes to make your ZenML pipelines more robust. It's always been a sleeper hit.
I originally wrote our Quick Wins docs page as a collection of 15 things you can do in ~5 minutes to make your ZenML pipelines more robust. It's always been a sleeper hit.
January 26, 2026 at 1:23 PM
ZenML just shipped our first Agent Skill: Quick Wins.
I originally wrote our Quick Wins docs page as a collection of 15 things you can do in ~5 minutes to make your ZenML pipelines more robust. It's always been a sleeper hit.
I originally wrote our Quick Wins docs page as a collection of 15 things you can do in ~5 minutes to make your ZenML pipelines more robust. It's always been a sleeper hit.
Sunday. Sentimental Value. Fifth time’s a charm.
January 25, 2026 at 9:48 AM
Sunday. Sentimental Value. Fifth time’s a charm.
Apparently time for my yearly repost of this line from 'Halt and Catch Fire'. Perhaps more relevant in 2026 than last year.
January 24, 2026 at 11:51 AM
Apparently time for my yearly repost of this line from 'Halt and Catch Fire'. Perhaps more relevant in 2026 than last year.
YouTube used to allow you to search/filter the videos you received when searching by upload date. i.e. I want to see the most recent videos over 20 minutes on a particular topic or keyword. No way to do that any more.
😓
😓
January 23, 2026 at 3:02 PM
YouTube used to allow you to search/filter the videos you received when searching by upload date. i.e. I want to see the most recent videos over 20 minutes on a particular topic or keyword. No way to do that any more.
😓
😓
Two case studies (both now in the ZenML LLMOps Database) came out this month that caught my attention. AutoScout24 (Europe's largest car marketplace) and Thomson Reuters both published detailed write-ups on how they're building internal platforms for AI agents.
January 23, 2026 at 2:13 PM
Two case studies (both now in the ZenML LLMOps Database) came out this month that caught my attention. AutoScout24 (Europe's largest car marketplace) and Thomson Reuters both published detailed write-ups on how they're building internal platforms for AI agents.
I spent last night messing around with PyO3 to see if I could run Rust code inside ZenML pipelines.
ZenML only officially supports Python, but PyO3 compiles Rust into a native Python module. ZenML doesn't know or care that there's Rust underneath.
ZenML only officially supports Python, but PyO3 compiles Rust into a native Python module. ZenML doesn't know or care that there's Rust underneath.
January 22, 2026 at 1:43 PM
I spent last night messing around with PyO3 to see if I could run Rust code inside ZenML pipelines.
ZenML only officially supports Python, but PyO3 compiles Rust into a native Python module. ZenML doesn't know or care that there's Rust underneath.
ZenML only officially supports Python, but PyO3 compiles Rust into a native Python module. ZenML doesn't know or care that there's Rust underneath.
Last year I wrote a chapter for a Packt book on DeepSeek covering rationale distillation for legal contract classification.
The problem: law firms can't send confidential contracts to external APIs. But they still need sophisticated clause classification across 41 legal categories.
The problem: law firms can't send confidential contracts to external APIs. But they still need sophisticated clause classification across 41 legal categories.
January 21, 2026 at 3:39 PM
Last year I wrote a chapter for a Packt book on DeepSeek covering rationale distillation for legal contract classification.
The problem: law firms can't send confidential contracts to external APIs. But they still need sophisticated clause classification across 41 legal categories.
The problem: law firms can't send confidential contracts to external APIs. But they still need sophisticated clause classification across 41 legal categories.
Humans are bad at picking schedules. Shopify proved it at scale.
When you're running 10,000+ DAGs and 150,000 runs per day, patterns emerge. Shopify noticed that engineers were clustering their cron schedules at "nice" times: midnight, 9am, top of the hour.
When you're running 10,000+ DAGs and 150,000 runs per day, patterns emerge. Shopify noticed that engineers were clustering their cron schedules at "nice" times: midnight, 9am, top of the hour.
January 20, 2026 at 12:56 PM
Humans are bad at picking schedules. Shopify proved it at scale.
When you're running 10,000+ DAGs and 150,000 runs per day, patterns emerge. Shopify noticed that engineers were clustering their cron schedules at "nice" times: midnight, 9am, top of the hour.
When you're running 10,000+ DAGs and 150,000 runs per day, patterns emerge. Shopify noticed that engineers were clustering their cron schedules at "nice" times: midnight, 9am, top of the hour.
ZenML MCP server now supports the full control plane.
We added 12 new read-only tools covering entities AI assistants actually need to understand your ML infrastructure:
→ Deployments: Discover what's running, check status, pull bounded logs
We added 12 new read-only tools covering entities AI assistants actually need to understand your ML infrastructure:
→ Deployments: Discover what's running, check status, pull bounded logs
January 19, 2026 at 9:57 AM
ZenML MCP server now supports the full control plane.
We added 12 new read-only tools covering entities AI assistants actually need to understand your ML infrastructure:
→ Deployments: Discover what's running, check status, pull bounded logs
We added 12 new read-only tools covering entities AI assistants actually need to understand your ML infrastructure:
→ Deployments: Discover what's running, check status, pull bounded logs
Something I really miss in Raycast that was great in Alfred is the ability to trigger the command bar, then start typing and then immediately have the folders show up as you type.
January 18, 2026 at 10:02 AM
Something I really miss in Raycast that was great in Alfred is the ability to trigger the command bar, then start typing and then immediately have the folders show up as you type.
I could move faster on side projects if I stuck with tools I already know. Lately I've been doing the opposite.
January 17, 2026 at 8:03 PM
I could move faster on side projects if I stuck with tools I already know. Lately I've been doing the opposite.
Every FastAPI server gets beautiful documentation for free. MCP servers get... nothing.
January 16, 2026 at 2:01 PM
Every FastAPI server gets beautiful documentation for free. MCP servers get... nothing.
Building a "simple" abstraction is never simple.
Schedules are a good example. At first glance: cron expressions, start times, done.
Here's how ZenML's approach to schedules evolved:
Schedules are a good example. At first glance: cron expressions, start times, done.
Here's how ZenML's approach to schedules evolved:
January 16, 2026 at 9:59 AM
Building a "simple" abstraction is never simple.
Schedules are a good example. At first glance: cron expressions, start times, done.
Here's how ZenML's approach to schedules evolved:
Schedules are a good example. At first glance: cron expressions, start times, done.
Here's how ZenML's approach to schedules evolved:
I've been tracking my reading since 2012. Over 1000 books. This evening I wondered: how often am I picking books I end up satisfied with?
Built a quick tool to find out—swipe right if satisfied, left if not. Rated all 1000+ books in one sitting.
Built a quick tool to find out—swipe right if satisfied, left if not. Rated all 1000+ books in one sitting.
January 15, 2026 at 3:18 PM
I've been tracking my reading since 2012. Over 1000 books. This evening I wondered: how often am I picking books I end up satisfied with?
Built a quick tool to find out—swipe right if satisfied, left if not. Rated all 1000+ books in one sitting.
Built a quick tool to find out—swipe right if satisfied, left if not. Rated all 1000+ books in one sitting.
I ran across a story from LinkedIn's engineering team recently that seems to illustrate the kinds of problems platform teams face when monitoring scheduled pipelines.
January 15, 2026 at 10:03 AM
I ran across a story from LinkedIn's engineering team recently that seems to illustrate the kinds of problems platform teams face when monitoring scheduled pipelines.
A couple of days ago I asked what local LLMs people were using for agentic work. I've now done the research and talked to people I know who run these models daily. Here's what I found.
January 14, 2026 at 3:35 PM
A couple of days ago I asked what local LLMs people were using for agentic work. I've now done the research and talked to people I know who run these models daily. Here's what I found.
Stale models hurt users. Twitter proved it.
In 2018, Twitter's Cortex team built ML Workflows on Airflow to replace the ad-hoc scripts teams were using to retrain models.
Before: "manually triggering and waiting for a series of jobs to complete."
After: scheduled retraining that ran reliably.
In 2018, Twitter's Cortex team built ML Workflows on Airflow to replace the ad-hoc scripts teams were using to retrain models.
Before: "manually triggering and waiting for a series of jobs to complete."
After: scheduled retraining that ran reliably.
January 13, 2026 at 12:31 PM
Stale models hurt users. Twitter proved it.
In 2018, Twitter's Cortex team built ML Workflows on Airflow to replace the ad-hoc scripts teams were using to retrain models.
Before: "manually triggering and waiting for a series of jobs to complete."
After: scheduled retraining that ran reliably.
In 2018, Twitter's Cortex team built ML Workflows on Airflow to replace the ad-hoc scripts teams were using to retrain models.
Before: "manually triggering and waiting for a series of jobs to complete."
After: scheduled retraining that ran reliably.
It's been a while since I was last using local models in an agentic harness (think Toad / OpenCode etc).
I want it more for documentation tasks + other local filesystem work rather than for coding per se.
I want it more for documentation tasks + other local filesystem work rather than for coding per se.
January 12, 2026 at 1:59 PM
It's been a while since I was last using local models in an agentic harness (think Toad / OpenCode etc).
I want it more for documentation tasks + other local filesystem work rather than for coding per se.
I want it more for documentation tasks + other local filesystem work rather than for coding per se.
I built a physical e-ink dashboard for monitoring ZenML pipelines.
TRMNL is a small e-ink display—pulls data via webhooks and shows it on an always-on screen. Perfect for keeping an eye on ML pipelines without constantly checking the dashboard.
github.com/zenml-io/tr...
TRMNL is a small e-ink display—pulls data via webhooks and shows it on an always-on screen. Perfect for keeping an eye on ML pipelines without constantly checking the dashboard.
github.com/zenml-io/tr...
January 12, 2026 at 9:57 AM
I built a physical e-ink dashboard for monitoring ZenML pipelines.
TRMNL is a small e-ink display—pulls data via webhooks and shows it on an always-on screen. Perfect for keeping an eye on ML pipelines without constantly checking the dashboard.
github.com/zenml-io/tr...
TRMNL is a small e-ink display—pulls data via webhooks and shows it on an always-on screen. Perfect for keeping an eye on ML pipelines without constantly checking the dashboard.
github.com/zenml-io/tr...
I've been using Claude Code's slash command system to build little productivity tools, and yesterday I shipped one that scratches a long-standing itch: Detour.
The problem: I'm deep in implementation work, and a tangential question pops up—"wait, how does this other module handle auth?"
The problem: I'm deep in implementation work, and a tangential question pops up—"wait, how does this other module handle auth?"
January 11, 2026 at 11:00 AM
I've been using Claude Code's slash command system to build little productivity tools, and yesterday I shipped one that scratches a long-standing itch: Detour.
The problem: I'm deep in implementation work, and a tangential question pops up—"wait, how does this other module handle auth?"
The problem: I'm deep in implementation work, and a tangential question pops up—"wait, how does this other module handle auth?"