New trend: programming by kicking off parallel AI agents
As Gergley said, parallel agentic coding is new territory. We’ll have to wait and see if there will be actual gains or will it be shitloads of more code to review.
Bookmarks RSS feed
-
-
Why I Spent $10,000 Learning How to Talk
When preparing to speak (for a talk or presentation) or when put on spot, we tend to focus on delivering the entire thing correctly. Instead, work on your first sentence (message and tone), then nail its delivery and let the rest flow. Similar to Steve’s cultivating confidence advice from last month but more specific to speaking skills.
-
The Evolution of LinkedIn’s Generative AI Tech Stack
Other than illustrating LinkedIn’s GenAI adoption journey, the article reinforces my belief in the notion that we should stop regarding AI as a massive job terminator. It’s a tool, just like any other, to supercharge one’s abilities and help move faster. A good case study on cleverly aligning new bolder initiatives with tech debt management, refactoring systems incrementally, and upskilling via pair programming. Personal take (🌶️): Despite the fancy system design, LinkedIn’s AI features for candidates feel average at best.
-
Under Pressure: Benchmarking Node.js on a Single-Core EC2
~2500 reqs/sec. That’s what a lowly 1-core 1-GiB server (Node.js & Postgres) can handle on its own. Let that sink in. Horizontally scaling is not always the answer and comes with its own complexity/tradeoffs. See if increasing the machine size will be enough before throwing in more servers and load balancers.
-
What caused the large AWS outage?
A stark reminder about over-dependence on one vendor for all infra needs. Amazon uses DNS tricks to load-balance its DynamoDB services. Due to a race condition, DNS records were emptied causing an outage of this massively popular distributed NoSQL database. That didn’t just impact applications using DynamoDB but also EC2 which depends on it for droplet provisioning configuration. Double whammy!
-
The 3 Levels of Influence (And Why You're Stuck on Level 1)
My key takeaway comes from the last part. Start small to build influence. Identify a small pain point and create content to fix it (RFC, code, presentation). Then use your earned trust to propose another small, related change. Once you have sufficient influence built from small wins, use it to propose more ambitious changes you always wanted to push.
-
The Problem with this Humanoid Robot - YouTube
Another problem with humanoid robots is that they take creativity and fun out of chores. Remember all the times you found ways to make your chores bearable or even enjoyable. Washing dishes? It’s the perfect time to catch up with a friend over phone. Ironing clothes? The perfect time to reflect on a difficult problem. Cleaning your house? The perfect time to rearrange things and get rid of stuff so there’s less to clean.
-
What is good software architecture?
A bold yet practical stance against formal architect roles. Every engineer should be skilled and empowered to influence system design. This cumulative engineering sense, combined with experience in the trenches, should shape the overall architecture rather than a full time thinker proposing imaginary ideas.
-
How Pinterest Transfers Hundreds of Terabytes of Data With CDC
Reading about how massive platforms use certain technologies is a great way to learn about them (and retain that learning!). This Pinterest case studies highlights change data capture (CDC) in all its glory - and the inefficient methods it replaces like DB triggers and batch jobs to detect inserts, updates, and deletions. Yet another instance of control plane-data plane design. Cool to know Debezium exists and is free.
-
What Actually Happens When You Press ‘Send’ to ChatGPT
Easy to digest, although nothing earth-shatteringly new. My biggest insight is how self-attention works, a mechanism an LLM model uses to decide while parts of the input are most relevant to predict the next token.