Lambros Petrou

Lambros Petrou

@lambrospetrou.com

Bootstrapping a business to simplify HTTP API testing ๐Ÿป skybear.net Principal Software Engineer @cloudflare. Past: DDOG, WhatsApp, AWS/Amazon ๐Ÿ  lambrospetrou.com

438 followers200 following254 posts

Lambros Petrou reposted

> We've been profitable from month one โ€“ every single month for 20 years. That consistency has given us the freedom to make decisions based on what's right for our customers and our family. The dream. Hugely inspired by your journey, Ian. Itโ€™s something Iโ€™d love to have for myself one day.

Ian Landsman@ianlandsman.com

I wrote up a few thoughts on HelpSpot turning 20 today ianlandsman.com/helpspot-tur...

Next project: AI agent for investing. Based on my own preferences and criteria for company fundamentals, combined with some trend following technical analysis and breakout detection from pivot points. An MCP server to allow me to chat with the agent through LLMs, and web UI.

Lambros Petrou reposted

I started reading software blogs before I got my first programming job 20 years ago. At this point, Iโ€™ve read thousands of blog posts and essays about software, but only a small handful stuck in my mind and changed the way I think. refactoringenglish.com/blog/softwar...

refactoringenglish.comThe Software Essays that Shaped MeI started reading software blogs before I got my first programming job 20 years ago. At this point, I've read thousands of blog posts and essays, but a small handful stuck in my mind and changed the w...

I have been listening to the Market Wizards books by Jack Schwager and I loved them. It does make investing and trading seeming like a skill that needs a lot of work, consistency, patience, and risk management. However, not a single strategy is best. It depends on your character.

Share your best resources, articles, publications, talks, about "AI in Databases and Storage products". I imagine some applications are for query optimization. What else is already being explored or would benefit from AI? Infrastructure or application level.

BlueSky wants me to verify my age (seems to be a new UK law), and the provider they use failed at least 4 times today in different times. If you force a check on your users, make sure it works first.๐Ÿ˜ฎโ€๐Ÿ’จ

Future in 20XX. Custom UI (websites, apps) doesn't matter anymore, at all. All devices directly handle multimodal input, and based on "intents" the service's APIs are called, and the response is converted to visual/audio/feeling (e.g. vibrations). Hardware + APIs + Data, wins.

Assuming you have an initial capital of 200K USD. What gets you to at least 2 million USD in the bank faster and more probable? A) Investing/Trading vs B) Building a business (3y+ horizon)?

Lambros Petrou reposted

I finally got some time to implement the "Per Request" insights for #SkybearNET. You can now drill down at the request level and figure out how your API is behaving, latency timings (DNS, TCP connect, transfer, and total duration), alongside response body sizes and status.

Also implements a URL grouping mechanism so that you can get a holistic view of an endpoint. Useful when your tests are doing hundreds of requests with different resources of the same type. For example, in the screenshot above, I am doing operations on hundreds of D1 databases.

"As AI expands, the real opportunity is to find a way to use human effort to create more value. [...] What we do have is agency over how weโ€™ll thrive in a world where human work is being redefined. Either you work for an AI or AI works for you." seths.blog/2025/07/prod...

seths.blogProductivity, AI and pushbackTypesetters did not like the laser printer. Wedding photographers still hate the iphone. And some musicians are outraged that AI is now making mediocre pop music. One group of esteemed authors is dโ€ฆ

"Logistician - ISTJ" sounded wrong initially, but the more I read about it, the more it fits me. I expected the "Architect - INTJ" type, and in hindsight I can even trace back to the questions that probably tipped me over. I find this personality test not to be very bullshity.๐Ÿ˜…

๐Ÿ’ก Shower idea, "ATTACHITE: SQLite files coordinator across servers". You like servers (VPS), and want to be able to use more than 1 machine. Shard your dataset across N SQLite files (N>=120). ฮ‘TTACH them to the app to query across them. Grow by moving files to new servers.

The ATTACHITE tooling will make it easy to figure out which server has each SQLite file, and handle the moving of the file to new servers. Mostly acting like a "migrator of SQLite files". Litestream integration for backups. It's a bad idea for most, but could be lots of fun!

There are many drawbacks, but the benefits are: 1. Super fast SQLite access. Everything local, no network. 2. Reduce blast radius*. 1 machine downtime is not 100% downtime. 3. Scale out. * Availability of a single SQLite file still depends on a single machine.

"Always use Postgres" ๐Ÿ˜… The database where removing a JOIN leads to 250x query execution time improvements. Mostly half joking here, it does its job well, but it always needs extra work at scale.

Lambros Petrou reposted

I have been using @kagi.com with Claude 4 Opus for several days now to implement and iterate on several investing strategies using Trading View PineScript. Over time I extended my system prompt and now it spits out scripts exactly in my coding style, explaining them perfectly. #ToTheMoon ๐Ÿš€

Lambros Petrou reposted

๐Ÿ”„ Just revamped Litestream! Our open-source SQLite tool now features faster point-in-time restores, conditional writes for S3, and lightweight read replicas. Perfect for modern deployments without changing your code. Even AI coding agents can benefit! โšก ๐Ÿ”— โžก๏ธ fly.io/blog/litestr...

Lambros Petrou reposted

โœจ It's new samwho visual essay o'clock! โœจ Reservoir sampling is one of my favourite algorithms. It allows you to sample from a set without knowing the size of that set. I try to assume no prior knowledge, and use no scary math notation. Just beautiful visuals. Let me know what you think! โค๏ธ

New article out: "Cloudflare Durable Objects are Virtual Objects"๐Ÿš€ This pops up a lot and is a mindset shift from the traditional way of thinking on "creating" and "destroying" resources, so I wrote a few thoughts how to think about it. www.lambrospetrou.com/articles/dur... #DurableObjects

lambrospetrou.comCloudflare Durable Objects are Virtual Objects | Lambros PetrouHow to think about Cloudflare Durable Objects as Virtual Objects and the mindset shift it brings.
View full thread

**2. Automatic instantiation:** Orleansโ€™ runtime automatically creates in-memory instances of an actor called activations. At any point in time an actor may have zero or more activations. An actor will not be instantiated if there are no requests pending for it.

New article out: "Cloudflare Durable Objects are Virtual Objects"๐Ÿš€ This pops up a lot and is a mindset shift from the traditional way of thinking on "creating" and "destroying" resources, so I wrote a few thoughts how to think about it. www.lambrospetrou.com/articles/dur... #DurableObjects

lambrospetrou.comCloudflare Durable Objects are Virtual Objects | Lambros PetrouHow to think about Cloudflare Durable Objects as Virtual Objects and the mindset shift it brings.

Kudos to the Microsoft Orleans folks that described Virtual Actors in their publication "Orleans: Distributed Virtual Actors for Programmability and Scalability" back in 2014.

**1. Perpetual existence:** actors are purely logical entities that always exist, virtually. An actor cannot be explicitly created or destroyed and its virtual existence is unaffected by the failure of a server that executes it. Since actors always exist, they are always addressable.

Starting a new project. What are the best AI models for semi real time audio/text processing? I need fast responses (less than 2 seconds), and I will handle providing the right context and how often to call them. Send me things to read. ;)