This Week I Learned - Week #31 2025

This Week I Learned - 

* GitHub released its latest coding agent to its Enterprise and Pro+ users: GitHub Copilot Coding Agent. Pipeline/headless agents like GitHub Copilot Coding Agent are built for scale. They integrate with CI/CD pipelines or scheduled jobs, enabling tasks like enforcing standards, migrating code, or updating infrastructure across repositories. While they demand stricter guardrails and detailed specifications upfront, they excel in delivering consistent and repeatable changes without requiring constant oversight for every prompt.

* CrewAI is a Python framework for orchestrating role-playing, autonomous AI agents. It is built entirely from scratch—completely independent of LangChain or other agent frameworks.

* "LLM system design is its own engineering discipline. We’re past the point where prompt engineering and model fine-tuning are enough. What matters now is composition; how models, tools, memory, and decision logic are structured and orchestrated. How you monitor them. How you debug them. How you ship them." - Armand Ruiz

* "Vibe coding and its AI hallucinations are creating a huge "technical debt" and future work for every skilled human programmer and software developer, to fix the mess at high fees.

AI-generated apps should display a privacy warning: "Vibe coded app! May publish your data, submit at your risk - building in PUBLIC".

Next time, ask in the AI prompt: "Make the app compliant with GDPR and privacy laws, too." - Good luck with that: you need to hire someone skilled to check, but it's expensive." - Fabio Ciucci

* Northern Virginia is the largest data center market in the world (13% of global capacity and 25% of capacity in the Americas). It has 352 data centers consuming four gigawatts (GW). Natural gas fuels 55% of Virginia’s electricity generation.  Northern Virginia hosts a high concentration of data centers due to its proximity to the landing points of numerous transatlantic undersea cables.

The European Union (EU) has 1830 data centers, around 15% of the world’s total, and a third of these are in Germany. 

Despite significant technological advancements over the decades, undersea cables have remained largely unchanged since their inception in 1850. The primary reason for this is speed. While satellite data transmission records around 200 

gigabits per second, undersea cables are at least five times faster, exceeding 200 terabits per second. However, these cables are prone to breaks, with approximately 150 incidents occurring annually due to aging, accidents or geopolitical conflicts.

Currently, both the United States and China have complete value chains for undersea cables. The world relies on a fleet of only 46 cable repair ships, 30 of which belong to NATO members.  - Franklin Templeton Institute

* Google Sheets has introduced a Timeline view, an interactive visual layer for some Google Workspace subscription types.

* Satellite basemaps from Google Maps, Esri, and Mapbox look different though they often use imagery from the same providers like Maxar, Airbus, Planet, and ICEYE. The differences come from how each platform processes that imagery. The final product involves extensive post-processing: stitching tiles, removing clouds/shadows, color correction, georeferencing, and sometimes manual editing.

Key insight: You're not just seeing raw satellite photos - you're seeing each company's decisions about data processing, visual design, and user priorities.

* People Analytics Handbook at GitLab uses data-driven insights to improve talent decisions and workforce processes at GitLab.  

* Google re:Work shares proven practices, real examples, and research from Google and other institutions to foster a people-first approach that drives success for your organization. 

* Google has five major office sites and more than 13,000 employees in India.

* U.S. accounts for 20% of India’s exports.

* U.S contributes only 5% to revenues of Nifty 500 companies.

* India has approximately 300 airfields in total, while the United States has around 16,000.

* The Maharashtra government has approved a proposal permitting nearly 60,000 public trusts, including temples, educational institutions, and charities, to invest up to 50% of their funds in instruments such as mutual funds and bonds.

* Approximately 10 AI-powered surveillance cameras, costing Rs 4.5 crore, will be installed at every entry and exit point of Secunderabad station to monitor criminal activities. Facial recognition systems will also be implemented at railway stations in Mumbai, New Delhi, Howrah, Sealdah, Patna, and Chennai.

* People+AI, an initiative by Nandan Nilekani’s EkStep Foundation, envisions the future of Artificial Intelligence (AI) in India through its "Adbhut India" (Amazing India) vision, which includes the Open Cloud Compute (OCC) project. As Nilekani puts it, "Adbhut India is becoming the AI use case capital of the world."

* Swiggy invested in Rapido in 2022 and owns around 12 percent in the company, which translates to around $120 million at the current valuation of slightly over $1 billion. Rapido is aiming to break the duopoly of Zomato and Swiggy in food delivery.

* The Maharashtra government has approved a proposal permitting nearly 60,000 public trusts, including temples, educational institutions, and charities, to invest up to 50% of their funds in instruments such as mutual funds and bonds.