Habr AI
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
VTB explained why AI pilots stall before production and how architecture can fix it
At Data Fusion on April 8–9, VTB acknowledged that many AI projects do not make it to production deployment: the cause may be accumulating errors, and the solution is an architecture with a human in the loop.
Why AI-generated texts annoy Habr readers and how to edit them manually
Habr AI published a column arguing that what bothers readers is not AI text itself, but its hollow confidence, formulaic structure, and lack of original thought — and how editing fixes that.
Singapore introduces AI in primary school, and Asia debates its benefits
After Singapore’s decision to introduce primary school students to AI, the debate shifted from the hype around education technology to whether such tools undermine memory, attention, and learning itself.
Five Free AI Image Generators: An Honest 2026 Test
We examine the five best free AI image generators in 2026: how they differ, where they break on complex tasks, and which one to choose for a specific use case.
OpenAI, Qwen and GigaChat: Why Choosing AI Models Is Getting Harder for Russian Business
The author examines how Russian businesses are being squeezed between OpenAI and Anthropic access restrictions, costly local inference, and the compromise of relying on fine-tuned Qwen models.
OWASP SAMM gains Agentic SAMM extension for secure development with AI agents
The Agentic SAMM project extends OWASP SAMM for AI agents: it adds controls for context, tool calls, trust in MCP servers, and the risk of long autonomous sessions.
Habr AI published the Rozitta Parser case study: how vibe coding eats up sleep and time
An architect with no development experience used DeepSeek, Claude and Gemini over six months to turn a Telegram parser into Rozitta Parser — and described how much sleep, money and personal time that kind of vibe coding
Senar: why AI agents do not replace programmers and are changing the development process
In the second part of the series on Senar, the author shows that an AI agent writes code quickly, but does not know how to doubt, remember the project context, or take responsibility for the consequences.
SimpleOne: AI in ITSM has become the corporate standard for support and service desk
SimpleOne described how companies are embedding AI into support: automating up to 80% of routine requests, speeding up SLA fulfillment, and keeping data within a secure perimeter.
Jitsi Meet: why transcription for electronic medical records requires Jigasi and Vosk
The analysis showed that transcription in Jitsi Meet is built around Jigasi, XMPP/SIP, and external speech recognition, and that for electronic medical records this is not enough without LLM post-processing and manual re
Google, Tambo, and GenUI: three approaches to AI interfaces built on the fly
A breakdown of Generative UI shows when AI should not only respond with text, but also assemble an interface: from HTML generation to Google’s A2UI and prebuilt components in Tambo.
SimpleOne showed why Claude and AI reviewers can’t distinguish AI-generated code from human-written code
SimpleOne proposed a test built around pairs of functions in which even an AI reviewer could not reliably distinguish human-written code from generated code, and explained why the issue comes down to context rather than
Pyotr Tripolsky on Habr showed an AI agent for analyzing news and Bitcoin signals
Pyotr Tripolsky published on Habr a diagram of an AI agent that looks for recent market events, filters out noise, and issues a short-term BUY, SELL, or WAIT signal.
Major AI companies are driving up memory prices and changing the rules for the gaming industry
The AI data center boom has driven up memory and storage prices, raised the cost of PCs and consoles, and at the same time pushed game studios toward more aggressive automation.
Suno 5.5 improves voice copying and melody preservation in Cover mode
In the Suno 5.5 update, Your Voice now produces recognizable vocals, and pairing it with Cover makes it possible to carry over your own melody, but quality still depends heavily on the dataset, range, and style.
Alfa-Bank described “vibe leadership”: how AI took over routine work and accelerated the growth of the A-Token platform
A senior Alfa-Bank executive explained how the in-house AI AlfaGen automated technical documentation, saved millions, and freed product teams to focus on strategy and growth.
X5 Tech added AI skills assessment to developer interviews
X5 Tech has integrated an AI block into developer interviews and evaluates not whether tools are banned, but how a candidate formulates prompts, checks the model’s answers, and takes responsibility for the code.
Why AI makes errors in requirements and architecture more costly despite speeding up development
AI has accelerated the path from idea to prototype, but that has made mistakes in requirements and architecture more costly: the wrong decision now scales before anyone can stop it.
Habr AI: why creating sentient AI may be more dangerous than enhancing the human brain
In a Habr AI column, the author argues that developing sentient AI carries greater ethical and existential risks than neural interfaces and enhancing humans themselves.
Anthropic, OpenAI, Google, and Microsoft have launched an AI platform war instead of a model race
The AI market is shifting from a model race to a platform battle: it now matters to control not only the models themselves, but also agent execution, data, security, and workflows.
OpenGrall introduces “Engineer” mode: robot writes drivers and configures modules on its own
OpenGrall has added an “Engineer” mode: via text command, the robot can create plugins, write code for new hardware, and calibrate modules, but the final integration decision remains with a human.
SD Studio turns local Stable Diffusion into “its own Midjourney” with an LLM assistant
SD Studio combines local Stable Diffusion, an LLM, and ready-made presets into a single pipeline to produce illustrations faster on your own GPU and avoid overpaying for external services.
Habr explained how to protect smart home voice control from leaks and hacks
Habr explained why accurate recognition alone is not enough for a smart home: without encryption, speaker verification, auditing, and local processing, voice commands are easy to intercept or spoof.
BentoML showed how to turn Grounding DINO into a production service with a web API
Using Grounding DINO as an example, the author showed how BentoML packages a computer vision model into a service with parameter validation, Swagger UI, a Docker build, and ready-to-use endpoints.
Habr details an AI framework for Claude with Clean Architecture and a TDD cycle
An author on Habr presented a framework that guides Claude through stories, test plans, and quality gates to produce code in the style of Clean Architecture and TDD instead of chaotic vibe coding.
Falcon Tech explained how machine vision for cities grew into a network of 4,000 systems
Falcon Tech described how, over eight years, it turned parking monitoring into a large-scale video analytics platform: in Moscow, it already runs on more than 4,000 hardware-software systems.
Falcon Tech showed how Moscow's video monitoring system grew out of parking enforcement
Falcon Tech explained how it built a video monitoring system for Moscow: from parking enforcement to analyzing the load on urban infrastructure using data from thousands of hardware-software systems.
MTS: the first architectural decisions in AI made today set constraints for decades ahead
In a column on Habr AI, the MTS team compares AI development to multilayered technical debt: decisions in code, data, and processes made now will constrain products for many years.
red_mad_robot engineer showed how to build an NER service for résumés: from annotation to API
A practical breakdown of NER for résumés was published on Habr: the author shows how to define entities, build and annotate a dataset, compare BERT models, and package the result in FastAPI.
Mo Gawdat of Google X: why AI's main threat is not code, but ethics
Former Google X executive Mo Gawdat believes AI already steers our attention, will accelerate a painful reshaping of the labor market, and will run into humanity's main limit — a lack of ethics.