My Tech Journal - May 2026

My Tech Journal - May 2026 Edition

---

AI Agents

• AI Agent Conference 2026: Alex Gallego presents the Agentic Data Plane
  by Alex Gallego, May 22, 2026
  

  • What is an Agent?
  

  • the prompt
  • set of tools
  • tokens alloted

  

  • In an enterprise, you want the agents to have access to a fixed set of (approved) tools. An agentic data plane focuses on
  

  • trust,
    
  • explainability and
    
  • context
    and offers controls to deploy regulated enterprise agents in production.
    

  • How to unlock private data to AI agents in a safe, auditable manner?
  The proposed architecture of an agentic data plane for the enterprise.
  

  

  • An audit log is an immutable transcript of agent actions along with the return payload in OpenTelemetry (OTel) format
  

  • The desired properties of an Audit Log:

  • immutable, complete, provably correct
    
  • survives ‘f’ node failures, given ‘2f + 1’ nodes
    
  • empirically verifiable
    

C++

• Effective Ranges: A Tutorial for Using C++2x Ranges - Jeff Garland - CppCon 2023
  • C++20 is the first official standard that introduced ranges.
  • Ranges are the largest revamp of the Standard Template Library (STL) in 20 years and impacts daily C++ programming dramatically.
  • This tutorial offers an excellent overview of ranges, views and spans.
  • Ranges algorithms are not fully compatible with STL algorithms.
  • Views are ranges with lazy evaluation and lazy evaluation allows for composition of several processing steps. This allows us to write functional style code.
  • If you are on older releases of C++, say, C++14 or C++17, then Eric Niebler’s range-v3 library allows you to use ranges.
  

eBPF

• XDP-Programmable Data Path in the Linux Kernel by Diptanu Choudhury
  ;login: Spring 2018 Vol 43, No. 1
  • an easy to read, short introduction to eBPF and XDP
  • offers a high level overview of XDP-based Firewall service
  

GenAI

• ghost
  • ghost is a hosted database (Postgres) service for agentic AI use-cases
  • emphasizes a Git like fork model to give every agent its own fork (copy) of the production database
  • after testing and validation, changes from the fork are merged into the production database and the fork is discarded
  • enables several agents to work in parallel without affecting the production database
  • offers Skills and MCP integration to work with several frontline products
  

Golang

• Debugging Go: from zero to Kubernetes - Michele Caci
  by Michele Caci
  • good introduction to debugging Go applications on local node, in docker container and in kubernetes environment
  • GitHub repo: Debugging Go: from zero to Kubernetes • Delve is a debugger for the Go programming language. As per the offical documentation,

    Delve is a better alternative to GDB when debugging Go programs built with the standard toolchain.
    It understands the Go runtime, data structures, and expressions better than GDB.
  

Infrastructure

• Creating our Own Kubernetes & Docker to Run Our Data Infrastructure - Modal
  
• The architecture of VictoriaLogs
  by Aliaksandr Valialkin, CTO, VictoriaMetrics
  • Nice high level overview of various design decisions that make VictoriaLogs really compact, fast and efficient
  • Implemented in Go
  • There are some similarities with ClickHouse e.g.
  
  • uses LSM tree to take advantage of the fact that the (log) file contents never change
    
  • logs are split into blocks of upto 2 MiB (compared to 8 MB - the default granule size in ClickHouse)
    Each block fits into typical CPU cache. This cache friendly design decision probably gives it a huge advantage.
    
  • The sequential disk I/O benefits both the write and read use-cases
    
  • column-oriented storage that allows for better data compression ratios
    
  • use of Bloom Filters
    

Testing

• Testing Distributed Systems the right way ft. Will Wilson
  The Geek Narrator • Jun 12, 2024
  
  • References mentioned in the conversation
    
    • google/AFL: american fuzzy lop - a security-oriented fuzzer
      
    • Antithesis
      
    • So you think you want to write a deterministic hypervisor?
      by Alex Pshenichkin
      
    • FoundationDB - the open source, distributed, transactional key-value store
      
  • Conventional Testing:
    a) Good for catching regressions
    b) Tests are fragile; they are dependent on implementation details of the system under test
    
  • Property Based Testing
    an idea from functional programming
    you always expect some property of the system to hold
    OR
    you never expect certain behavior from the system
    
  • Deterministic Simulation Testing
    a) applying fuzzy or Property Based Testing approach to distributed systems
    b) build mock implementation of all the parts of your system that could be non-deterministic
    
  • Autonomous Testing
    Tests are generated by a computer to cover scenarios that you didn’t think about