Rost Glukhov | AI Systems & Infrastructure

Spec-Driven Development entered 2026 as the serious developer’s answer to vibe coding drift.

Decision records are the missing memory layer in AI-assisted software development.

Go error handling is easy to complain about.

Testing concurrent Go code has always required a bit of discipline.

Google’s Agent2Agent protocol, usually shortened to A2A, had a strange first year.

Polling agents are one of the least glamorous parts of AI assistant architecture, but they are also one of the most useful.

The A2A Protocol, short for Agent2Agent Protocol, is an open standard for communication between independent AI agent systems.

AI agent architecture is starting to split into two layers.

CQRS is one of those patterns that gets oversold, overcomplicated, and occasionally misdiagnosed as a cure for plain old CRUD boredom.

Mermaid is a text-based diagramming tool for people who would rather write diagrams than drag boxes around a canvas.

Organizing notes by topic sounds logical until you have notes on PostgreSQL in five different folders and cannot find the one that matters for today’s problem.

Most engineering notes are written once and forgotten.

The dominant model for publishing knowledge online has not changed much since the early 2000s: write something, polish it, publish it, move on.

Single-model systems are simple.

Running a 70B parameter model to summarize a 200-word email is wasteful.

LLMs are unpredictable.

LLM costs scale linearly with usage.

Memory turns assistants from reactive to persistent, but it is also where many systems quietly rot.

A production AI assistant is not “an LLM with a prompt”.

AI is not replacing knowledge management; it is changing the shape of it for both individuals and teams.

Developers do not usually suffer from a lack of information.

Open-source AI agent frameworks are exploding in popularity on GitHub.

I tested Speculative decoding (Multi-Token Prediction, MTP) performance in Qwen 3.6 27B and 35B on an RTX 4080 with 16 GB VRAM.

llama.cpp router mode is one of the most useful changes to llama-server in years.

Most modern knowledge systems optimize retrieval, and that is understandable.

The premise is simple: compiled knowledge is more reusable than retrieved fragments.

PKM, RAG, wikis, AI memory systems, and now practical AI-assisted workflows are often discussed as if they solve the same problem.

Information overload is less about sheer volume than about unresolved inputs.

Most LLM “structured output” tutorials are unserious.

This page is a practical reference for agentic LLM inference tuning (temperature, top_p, top_k, penalties, and how they interact in multi-step and tool-heavy workflows).

Idempotency in distributed systems is the property that saves you after the network lies, the queue retries, the client panics, and the operator hits replay.

You already chat to Hermes Agent from your phone with text.

Hermes Agent ships with a Kanban-style board and the Hermes Gateway that can saturate your self-hosted LLM if too many tasks are dispatched at once.

Hermes Agent treats skills as the default way to teach repeatable workflows.

Hermes Agent from Nous Research is a model-agnostic, tool-using assistant you run locally or on a VPS.

MinIO Community Edition is no longer a safe default for new production systems.

Most AI agent stacks still treat security as a post-demo fix.

Modern assistants still forget everything when you close the tab unless something persists beyond the context window.

You know the drill.

OpenClaw did not fail as a product.

For a long time, llama.cpp had a glaring limitation: you could only serve one model per process, and switching meant a restart.

Most teams misuse Claude Skills in one of two ways.

Batch files and shell scripts often need a short wait so a double-clicked window or installer log stays visible.

Hermes AI assistant, officially documented as Hermes Agent, is not positioned as a simple chat wrapper.

OpenClaw has two extension stories, and they are easy to mix up.

This article is about OpenClaw plugins — native gateway packages that add channels, model providers, tools, speech, memory, media, web search, and other runtime surfaces.

OpenClaw looks simple in demos.

The real argument is not whether PostgreSQL can search text or whether Elasticsearch can store documents.

Alerting gets described as a monitoring feature far too often.

Slack integrations look deceptively easy because you can post a message in one HTTP call.