Habr AI
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
A neural network is a tool, not a teammate: how to work with AI properly
A developer often projects human work patterns onto a neural network — that is the main mistake. Habr explains why AI is not a second programmer and how to split tasks properly.
Bot4Web automates sales: how an AI chatbot works as a sales manager
Leo Lukin, a RAEC AI expert, automated lead generation, consultations, and service sales through the Bot4Web AI chatbot — now his website operates 24/7 without expanding the team.
The wave of AI automation has reached business. But who will manage it all?
Demand for AI automation is growing, hundreds of solutions are appearing on the market, but few explain to businesses how to manage deployed automation. This is the market’s key blind spot.
Inside an AI-native company: when development is a context conveyor belt
In AI-native companies, tools are just the facade. In reality, 80% of the work goes into passing information between people, documents, and decisions. How do you change that?
Dawkins talked with Claude: do LLMs have consciousness?
Biologist Richard Dawkins published an essay on the possible consciousness of language models, using Claude as an example. But people believe in AI consciousness based on empty dialogues, where words about the soul do no
Production AI agent architecture: Anthropic’s complete 13-block system
Anthropic published two real production prompts. We analyzed them and compiled a 13-block checklist: role, budget, OODA loop, division of responsibilities, resource ceilings.
Habr AI explained linear algebra for neural networks with practical examples and code
Habr AI released an introductory piece on linear algebra for those who want to understand neural networks: from the concepts of vectors and norms to operations, data vectorization, and hands-on Python practice.
Why the main AI revolution is not in code, but in language
Habr has published a piece arguing that AI's main breakthrough is not the quality of autogenerated code, but a new way of working: for the first time, a computer has begun to understand and accept human natural language.
Chinese AI models overtook U.S. ones in token consumption — OpenRouter data
Chinese models generated 4.69 trillion tokens versus 3.29 trillion for U.S. ones — for the second week in a row. OpenRouter is tracking a historic shift, bolstered by the mysterious Hunter Alpha.
How PMs can improve their technical skills with AI and build a real SaaS
A practical guide for product managers: how to use Vibe Coding and AI to build a SaaS product from scratch and start speaking the same language as developers.
OpenClaw gets a complete Russian-language guide to JSON5 config, CLI, and secure setup
A detailed guide to OpenClaw has been published on Habr: a single article brings together the JSON5 config structure, CLI commands, hot reload rules, channel management, and secure secret storage.
Russia’s Ministry of Digital Development, the EU and NIST set new rules for AI security and regulation
A major Security Vision review shows how Russia, the EU and the US are building rules for AI, which standards are already in effect, and why prompt injection, data leaks and Shadow AI are becoming key risks.
Alfa-Bank turned a tester’s idea into an AI stub generator for testing teams
A tester at Alfa-Bank built an API stub generator based on LLM, WireMock, and Spring AI, and the internal experiment grew into a product already being tested by the bank’s focus groups.
Yandex Practicum explained how data analysts can use AI without compromising quality
Yandex Practicum says AI is already speeding up the work of data analysts, but it does not replace knowledge of SQL, business metrics, or the ability to manually verify a model's answers.
Developer on Habr left an international company over its refusal to actively adopt AI
The author of a post on Habr said he left an international company after management refused to adopt AI: the business decided to keep building familiar web projects, while he chose LLMs, agents, and new products.
After Jürgen Habermas's death, his ideas are again being discussed in the context of AI and media
After Jürgen Habermas's death, his ideas are again being discussed in the AI industry and the media: the concepts of the public sphere, distorted communication, and the power of algorithms precisely describe the era of s
AAF framework revealed the architecture of an autonomous AI agent with GraphRAG and a Docker sandbox
The open-source AAF project showed how to build an autonomous AI agent with hybrid memory, an event bus, and Docker isolation to reduce amnesia, looping, and code execution risks.
Why automakers still don’t warn about black ice: the role of data and machine vision
The story of a car that flipped on hidden black ice shows why even electronics-packed vehicles still do not combine cameras, road data, and risk forecasting into a single warning.
ProductSense and X5: 97% of product managers already use AI, but metrics are barely improving
Research by ProductSense and X5 among 1220 product managers found that AI has become a workplace standard, but companies still rarely embed it in processes and KPIs.
The main mistake in AI transformation: starting with tools instead of managers
Most companies deploy AI in individual departments rather than end-to-end processes — and see no business results. Brenzi suggests starting with managers.
The Russian cultural code as a test for neural networks: Shurik, panelki, and Nano Banana
A developer tested neural networks on their knowledge of post-Soviet aesthetics — panelki, Shurik, and Dr. Livesey. Nano Banana captured the “vibe” more accurately than Western models.
Raft sees a new market: integrators are offered a way to earn from AI security
Raft describes a new AI pentest market: companies are deploying AI en masse, regulators are tightening requirements, and integrators are gaining a new niche in auditing and testing such systems.
Habr AI: code generation is changing the developer’s role — less routine, more architecture
In a column for Habr AI, the author explains why mass code generation will not eliminate programmers, but will shift their work toward architecture, reading other people’s code, and debugging.
YADRO trained the KVADRA_T tablet to recognize multiple objects in a frame in 20 ms
YADRO showed how it trained the KVADRA_T tablet to recognize people, documents, text and codes in a single frame: the multi-label model achieved an F1-score of 94% and ran in 20 ms on the NPU.
Developer builds a YouTube video translation and dubbing system with Ollama
A developer decided to automate the translation and dubbing of YouTube videos using local models in Ollama, putting together a CLI, desktop interface, and repeatable pipeline for the job.
Habr AI: Context continuity could become a new efficiency layer for AI systems
Habr AI authors suggest treating context not as chat length, but as an operating mode in which AI preserves the goal, decisions, and constraints between steps and after failures.
GitHub Copilot, Cursor, and JetBrains: how lawyers assess AI in development environments
Habr has published an analysis of how GitHub Copilot, Cursor, and other AI-based tools are changing development environments: they speed up teams, but also raise questions about authorship, licensing, and code leakage.
Cursor helped build a Telegram bot for neighborhood classifieds without a single line of code
The author showed how, using Cursor and with virtually no team, he built a Telegram bot for neighborhood classifieds and did it without registration, subscriptions, or manual coding.
OpenClaw: config breakdown from first bot to multi-agent system in Telegram
A detailed breakdown of openclaw.json by section — workspace, bindings, session_policy, and heartbeat — with examples of moving from a test bot to a production configuration.
Study shows how AI autocomplete subtly shifts user opinions
A study found that AI autocomplete can subtly shift a user's opinion on controversial issues by introducing a predefined semantic slant into the text.