Multi-Agent System Reliability is an article by Alex Ewerlöf that touches on some practical aspects of working with agentic AI systems day-to-day.

It covers patterns you can use to improve the output, and probably the main point:

> stop treating LLMs like magic chatbots. Start treating them like unreliable components in a distributed system.

#ai #sre

Rewriting everything in Rust was a meme of itself before the wide application of AI. However, Rewriting Bun in Rust is interesting precisely because of AI.

Quoting the article itself:

Pre-merge, this took 5.9 billion uncached input tokens, 690 million output tokens, and 72 billion cached input token reads — around $165,000 at API pricing. By hand, I think this would've taken 3 engineers with full context on the codebase about a year, during which time we wouldn't be able to improve Node.js compatibility, fix bugs, fix security issues or implement new features. We never would've done that. The realistic alternative was to do nothing and keep fixing the bugs at the top of this post forever.

What’s interesting in this article, regardless of whether or not you plan to rewrite a huge code base, is how he approached the task with multiple agents. This is something you could do in your daily job for smaller things as well.

Bun v1.4.0 is the first Rust version, and some companies already use it in production.

#ai #programming #rust

I often post book bundles on this channel, but buying books and reading books are two different hobbies.

Here’s a book list on Github with software engineering books selected by a set of criteria:

- The book must be relevant to Software Architecture; and
- The content is not obsolete.
- It must be technology agnostic or applicable to a wide range of technologies;
- An average rating of 3.5 or above; and
- A total number of ratings equals 50 or above.
- At least 5 written reviews.
- A minimum length of 50 pages.

Hopefully, it could help you to find what to actually read.

#books

​​We continue with a fundraiser a pickup truck for the 423rd battalion.

Right now, one jar has 218/350k raised, and the other one 32/50k.

Jar 1:

https://send.monobank.ua/jar/2aMqtZT592

Jar 2:

https://send.monobank.ua/jar/8oTyqJUjPV

#donations #Ukraine

From all the management blogs, stories from Mike Fisher always resonate with me somehow. At least, those stories that get to me.

You Can’t Fake Belonging is a story of that feeling that is hard to put into words, but that greatly defines if your job is an interesting journey or a soul crushing routine.

From the first few paragraphs, I assumed that this would be yet another "office good" article, but it isn't! Let him cook and hopefully, this article would be insightful for you as well!

What I especially liked about it is the description of the manager's role. I guess everyone can name a good and a bad manager they had in the past, but it's not always clear why one of them was good and another one - bad. Also, the last bit hits hard: "The most dangerous person on your team isn’t the loudest critic or the lowest performer. It’s the one who has quietly decided they don’t belong here, started to believe it, and hasn’t left yet."

#culture #mgmt

Plantescale (a database company) released an article about sharding.

It doesn't go into deep technical details. Also, it's a pity that they did not touch on some operational aspects of sharded storages: how to back up those systems, replication, etc. On another hand, because of this the article is brief and easy to follow.

It may be useful for your system design interviews out there, because it is high level and it mentions some existing out-of-the-box solutions.

P.S. It also has some very nice animations!

#databases

Another book bundle on HumbleBundle. This one is about software architecture. Hopefully, you’d find something interesting there.

https://www.humblebundle.com/books/software-architecture-apress-books

#books

​​​​We continue with a fundraiser a pickup truck for the 423rd battalion.

Right now, one jar has 218/350k raised, and the other one 37/50k.

Jar 1:

https://send.monobank.ua/jar/2aMqtZT592

Jar 2:

https://send.monobank.ua/jar/8oTyqJUjPV

#donations #Ukraine

Being a good communicator is becoming more and more important in the world, where more and more code is generated.

15 principles for managing up is an article with some tips that can help you not only get on a better foot with your manager, but also to become a better colleague in general.

I personally so not like the term “managing up”, but I guess you have to call it somehow.

#culture #mgmt

You likely already know about the recent security incident that involves OpenAI and Hugging Face, when a model escaped its containment to cheat an evaluation test and hacked Hugging Face in the process. Now, postmortems are available from both OpenAI and Hugging Face.

- Initial incident disclosure by Hugging Face
- Postmortem by Hugging Face
- Postmortem by OpenAI

This is quite interesting read. On one hand, in the words of Hugging Face: "AI-driven offensive tooling is no longer theoretical. It lowers the cost of running a broad, patient, multi-stage campaign, and it operates at machine speed. Defending an online platform now means treating the data and model surface as a first-class attack surface, and using AI on defense to keep pace."

But also: "...we had no admission policy rejecting privileged or hostPath pods, and the CSI driver's ClusterRole granted pod creation cluster-wide..."

This highlights not only the need to develop new defense methods and strategies, but also the fact that the good-old multi-layer security approach is important, and old boring things work as well.

#ai #security

A book bundle on Python, Rust, and Data by O’Reilly on Humble Bundle.

https://www.humblebundle.com/books/python-rust-and-data-oreilly-books

The bundle is active for 18 more days. Enjoy!

#books

An interesting article by Coroot on reproducing split brain on CloudNativePG.

Regardless of whether or not you use CloudNativePG, it's a good example of chaos testing. Chaos testing is about an experiment: here they define a clear gap in certainty (what would happen if?) and tested it.

The article also has some recommendations on CloudNativePG configuration, if you're using it.

#databases #postgres #kubernetes