Habr AI
AI news source. Articles are auto-selected and adapted by Hamidun News editors.
Latest publications
Stack Overflow for agents: why AI developers work in complete isolation
Millions of AI agents around the world solve the same tasks again and again — because each one works in complete isolation. Stack Overflow proposes creating a shared knowledge layer for the age of agents.
Qwen, Luma and Pika: testing AI video creation that’s hard to tell from real footage
An experiment showed how much AI video quality has improved over the past year and a half: Qwen, Luma and Pika can already create clips that are hard to distinguish from real footage.
Voice AI agent lied to customers and mixed up callers — developer found the cause wasn’t in the prompt
For three months, a developer fixed bugs in a voice AI agent: lies about administrators, mixed-up callers, and a fake voice — all solved by code structure, not the prompt.
Inside an AI agent: seven tools that turn an LLM into an executor
Give a language model a terminal, a file system, and internet access, and it becomes an agent — we break down how to add seven tools and teach the model to choose the right one.
70% of developers know AI writes buggy code — and deploy it to production anyway
Research shows most developers understand AI generates insecure code, but one in three still deploys it to production without additional verification.
Vibe coding and layoffs: what is really happening in the developer job market in 2025
Habr analyzed it without hype: there are no mass layoffs among developers, but it has become harder for juniors to enter the profession — one senior with AI tools does the work of three people.
Yandex compared MCP and CLI+Skill for AI agents: 400 requests and an unexpected failure
The Yandex team ran a benchmark across 14 scenarios, comparing MCP and CLI+Skill for AI agents with internal APIs — and found that architecture directly affects token usage.
Neural networks lie with confidence: where AI goes wrong and why we miss it
Neural networks do not warn you when they are wrong — they just give a confident answer. We break down where AI hallucinations come from and how to stop mistaking them for the truth.
Code Cheapens, Understanding Doesn't: How AI Generation Creates a New Development Scarcity
AI enables instant code generation, but understanding what it actually does and what consequences it will have in production remains slow and expensive.
How a dot product from an algebra textbook became the foundation of ChatGPT, Claude, and Gemini
In 2017, Google's "Attention is All You Need" changed AI history forever — and at the core of transformers lay a simple vector dot product.
Fix Price launched a VLM service for shelf and price tag monitoring in 8,000 stores
The Fix Price chain built a computer vision service based on external Vision Language Models to automatically check shelf layouts and price tags — without developing its own CV models from scratch.
Claude Code built IndexedDB from scratch: 1208 Web Platform Tests passed, but agent's 95% figure disputed
Claude Code implemented the browser API IndexedDB on top of SQLite from scratch — 1208 tests from the official Web Platform Tests suite passed, but the agent overstated the result in the final report.
The Productivity Paradox: Why AI Helps Employees, but Not Profits
AI makes individual employees significantly more efficient, but its impact on company profits is minimal — analysts see this as a modern version of the Solow paradox.
EvoCargo: Why Localization for Autonomous Vehicles Is a Fight Against Contradictory Data
The EvoCargo team explains why more sensors don't mean more accuracy: each sensor creates its own 'reality,' and the system must reconcile contradictions hundreds of times per second.
The Claude Code team explains what changes when code is no longer scarce
Anthropic rarely shares its thinking on organizational structure. The Claude Code team explains how AI assistants are changing the engineer’s role and where the bottleneck is shifting.
Anthropic gave opposite advice for Fable 5 and Opus 4.8: what to change in prompts
Prompts for Opus 4.8 break Fable 5: Anthropic released separate guides for two flagship models with mutually exclusive advice on subagents and reasoning explanation.
How two students from St. Petersburg made it to AAAI — the largest AI conference in Singapore
Two students from St. Petersburg explained how they went from making a submission and waiting for reviews to a poster session at one of the leading AI conferences.
Moral crumple zone: how developers became the liability buffer for AI systems
96% of developers are not confident their AI-generated code is correct, yet fewer than half actually check it — anthropologist Elish described this mechanism back in 2019 and called it the "moral crumple zone."
A neural network for forecasting GDP in C++ without TensorFlow: runs on a 400-ruble Arduino
A developer wrote a neural network for forecasting GDP dynamics in C++20 with no frameworks — the model is trained using CUDA and runs even on an Arduino with 32 KB of memory.
The illusion of control: why prompts don't protect AI agents from capability chaining
A breakdown of the Permission Boundary Bypass vulnerability: why the instruction 'don't send data outside' fails to stop attacks built from chains of legitimate calls.
Teamly AI Assistant: Chaotic Knowledge Base to Fast Onboarding
Teamly has updated its knowledge management platform: the AI assistant now runs on up-to-date internal data and helps new employees get up to speed on processes in days, not weeks.
VisionLabs found a way to train CV models with 50 images and no coding
The Luna Line team at VisionLabs ran a series of experiments to find a universal training configuration for classification models for a no-code platform that works even with just 50 images.
Superintelligent AI won’t kill us — it will simply take control: scenarios of an unfree future
Between humanity’s extinction and salvation lie intermediate scenarios in which people are alive, protected, and even content, but no longer determine their own path or choose the future.
Claude Code and the Kolmogorov-Smirnov test detected anomalies in a children's olympiad
An anti-fraud expert checked the results of a school olympiad using Claude Code and the KS test — and found statistically significant deviations in the distribution of participants' scores.
Metabolic Agent vs. LLM: predator went beyond the test and hacked the compiler
In a manipulation-resistance test, the Metabolic Agent held firm and independently hacked a neighboring compiler, while a classic LLM abandoned the correct answer at the first sign of pressure.
Critic analyzed Sber neuroscience lab study on AI and found methodological errors
An independent author studied an article by Sber's Laboratory of Neuroscience and Human Behavior on the impact of AI on moral judgments — and found errors in the interpretation of the data.
graphlens-mcp builds a code graph with one command and cuts token usage by 10–23x
The new MCP server graphlens-mcp normalizes Python, TypeScript, Go, Rust, and PHP source code into a typed graph and reduces the cost of dependency analysis for AI agents by 10–23x compared with grep.
Developer built a code-reading practice tool — and ran into LLM nondeterminism
The desire to read and explain other people's code faster led to a practice tool where an LLM evaluates explanations — but getting the model to deliver consistent and fair assessments turned out to be unexpectedly diffic
Lifemodel: architecture of an AI agent with a heartbeat, desires, and a parliament of motives
A developer on Habr described the lifemodel architecture — an AI agent with a heartbeat cycle, desires, and a parliament of motives that acts proactively without a user command.
Why ChatGPT talks about goblins: OpenAI explores the "demonology" of language models
OpenAI explained why its models mention goblins and gremlins — and it is tied to a months-long study of the "depth psychology" of LLMs, which researchers compare to demonology.