Habr AI
OpenAI launches GPT-5.4 with PC control and signs a Pentagon contract
OpenAI's new models gained the ability to control computers, Anthropic was deemed a national security threat, and Grok predicts military con
Anthropic is creating a universal currency for the AI economy
The developer of Claude plans to launch a universal digital coin. Anthropic's new financial architecture could become the basis for settleme
How a Russian IT Company Turned Corporate Knowledge into a Working RAG System
Engineers at Tsifra built a RAG pipeline that speeds up search across internal knowledge bases severalfold. Instead of fine-tuning models, t
AI assistants for coding: from autocomplete to autonomous agents
Over three years, AI assistants for programmers have evolved from advanced autocomplete into full-fledged agents capable of reading reposito
97 Hours on a Single GPU: An Experiment With a Self-Learning Neural Network and the Trap of Impressive Metrics
An independent researcher tried to teach a language model to plug in new skills like apps — without retraining. The architecture worked, the
Diagnosis by Algorithm: Who Is Responsible When AI Makes a Mistake in Medicine
AI is already making diagnoses and prescribing treatment. But when the algorithm gets it wrong, the patient is left alone with a system that
Your own cloud LLM: how to fit within 16 GB of VRAM
Spending on language model APIs is rising rapidly. Developers are increasingly looking for an alternative — and finding it in self-hosted so
Architecture as Code: how LLMs accelerate system design
The BCS team showed how to move architectural documentation into code and leave routine tasks to LLM assistants. An approval cycle that used
Self-Aware MCP Server: How to teach AI agents to understand the real world
AI assistants are great at writing code, but they do not know which OS you use or what time zone you are in. The new MCP server solves this
Why language models will never become AGI: Wittgenstein’s century-old lesson
Austrian philosopher Ludwig Wittgenstein formulated a principle that explains the fundamental limit of modern LLMs. We examine why the path
Claude Code and NotebookLM: how to build a free RAG system for deep research
Developers have found a way to combine Claude Code with Google's NotebookLM to create a powerful document analysis system — practically free
Skaro: how one developer built a full-cycle AI orchestrator through sleepless nights
A Russian developer created a tool that turns specifications into finished code with AI, without losing context or compromising quality. Ska
Programmers after AI: who will remain in the profession
A large-scale experiment with AI code generators showed how close we are to the point where machines can replace developers. The results wer
National language model: ambitions, reality, and the cost of sovereignty
Russia is seriously discussing the creation of its own LLM. But between intention and outcome lies a chasm of talent, hardware, and decades
Documents as code: how AI is reshaping the way we work with information
Why keep dozens of copies of the same report if AI can assemble documents from source data on demand? A new paradigm for working with inform
Google released a CLI tool for Workspace: a bridge between enterprise infrastructure and AI agents
Google has made a command-line interface for managing Workspace publicly available. The key point is that the tool is designed so language m
KernelEvo: Russian framework automates GPU kernel generation with AI
The AIRI Institute has introduced KernelEvo, a framework that automatically generates optimized GPU kernels in CUDA and Triton. Instead of w
One developer replaced a product team using AI agents
A developer built an ecosystem of seven websites with SSO, monitoring, and GDPR compliance in a month and for under $1,000, replacing a tech
Cognitive memory for an AI agent: how SQLite replaced vector databases
A developer built a memory system for a local AI agent that can not only remember, but also forget. A graph architecture on top of SQLite pr
AI agent for checking technical specifications: why automate what doesn't work manually
A developer created an AI agent that analyzes technical specifications for errors and contradictions. Her experiment reveals a systemic prob
Edge AI in video cameras: how OpenIPC, an open project, is changing the rules of video analytics
The OpenIPC project has grown from a niche Linux distribution for IP cameras into a full-fledged community that brings AI directly to device
Large language models: why out-of-the-box deployment remains an illusion
There are now so many open LLMs that choosing a workhorse is a quest in itself. But the real problems begin after download: not a single maj
AI starters for business: how AI agents write CRM applications instead of developers
A new approach to building business applications: ready-made AI templates with prompts make it possible to assemble custom CRM modules for p
AI code review without the cloud: how Ollama is changing the approach to local development
Developers showed how to run a full AI code review directly on a local machine—without cloud services, API keys, or subscriptions. Ollama an
Developer built a spine MRI analyzer with Python and Gemini
A programmer with back pain built the desktop application Spine Advisor, which uses the multimodal model Gemini 3 Flash to analyze MRI scans
Decima-8: the architecture seeking to reinvent neuromorphic chips
The new neuromorphic architecture Decima-8 proposes a radical approach: encoding signals with energy levels, digital crossbars instead of un
Juniors vs. neural networks: how beginner developers can survive in the job market
Neural networks already write routine code and solve LeetCode problems — the very work junior developers used to learn on. How can newcomers
Telegram bot builds a VkusVill grocery cart from a single phrase
A Russian developer created an open-source bot that understands requests like “put together breakfast for two” and automatically builds a Vk
Cursor announced the third era of AI software development — and backed it up with data
The Cursor team published a manifesto on the three eras of AI software development. The company's internal statistics show that the paradigm
How to stop guessing and start measuring the quality of AI agents
The Bitrix24 team spent six months building a benchmarking system for its AI agent Marta. Their experience shows why manual testing no longe