https://www.danilchenko.dev/Recent content in Home on danilchenko.devHugoen-usSat, 11 Jul 2026 08:23:30 +0000https://www.danilchenko.dev/posts/python-tooling-guide/Thu, 02 Jul 2026 06:00:00 +0000https://www.danilchenko.dev/posts/python-tooling-guide/A complete guide to modern Python tooling in 2026: uv, Poetry, and pip; mypy, Pyright, ty, and Pyrefly; plus free-threading, t-strings, and Polars vs Pandas.https://www.danilchenko.dev/posts/tech-careers-ai-guide/Thu, 04 Jun 2026 08:44:59 +0000https://www.danilchenko.dev/posts/tech-careers-ai-guide/9 deep dives into the 2026 tech job market: layoffs, AI displacement data, salary benchmarks from Cyprus to the US, and the productivity paradox. Full guide.https://www.danilchenko.dev/posts/llm-inference-efficiency-guide/Wed, 13 May 2026 08:55:00 +0000https://www.danilchenko.dev/posts/llm-inference-efficiency-guide/Five approaches to making LLMs faster and cheaper — compression, diffusion decoding, architecture, KV cache, and sparse attention — explained with real numbers.https://www.danilchenko.dev/posts/ai-coding-tools-guide/Mon, 13 Apr 2026 07:00:00 +0000https://www.danilchenko.dev/posts/ai-coding-tools-guide/Everything we've covered on AI coding tools — comparisons, pricing, privacy, agents, and the security risks nobody expected. Updated April 2026.https://www.danilchenko.dev/posts/django-bolt/Sat, 11 Jul 2026 08:23:30 +0000https://www.danilchenko.dev/posts/django-bolt/Django-Bolt puts Actix and Rust under Django to hit 188K RPS. I dug through the source and benchmarks — where it beats FastAPI, and where the speed evaporates.https://www.danilchenko.dev/posts/python-3-15/Fri, 10 Jul 2026 08:23:56 +0000https://www.danilchenko.dev/posts/python-3-15/Python 3.15 lands lazy imports, a frozendict builtin, unpacking in comprehensions, and the Tachyon sampling profiler. Here's what changed and what to test.https://www.danilchenko.dev/posts/fastapi-vs-litestar/Thu, 09 Jul 2026 08:22:01 +0000https://www.danilchenko.dev/posts/fastapi-vs-litestar/FastAPI vs Litestar vs Django Ninja in 2026: real code, serialization speed, dependency injection, and a clear pick for every kind of Python API project.https://www.danilchenko.dev/posts/cs-degree-worth-it/Wed, 08 Jul 2026 08:22:44 +0000https://www.danilchenko.dev/posts/cs-degree-worth-it/Is a computer science degree worth it in 2026? CS grad unemployment hit 6.1%, yet the answer flips depending on whether you study in the US or the EU.https://www.danilchenko.dev/posts/claude-sonnet-5-review/Tue, 07 Jul 2026 08:22:27 +0000https://www.danilchenko.dev/posts/claude-sonnet-5-review/Claude Sonnet 5 review after a week: it nearly matches Opus 4.8 on coding, beats it on Terminal-Bench, and runs at just $2/$10 — but the tokenizer bites.https://www.danilchenko.dev/posts/duckdb-vs-polars/Fri, 03 Jul 2026 08:23:13 +0000https://www.danilchenko.dev/posts/duckdb-vs-polars/DuckDB vs Polars for data engineering in 2026: benchmarks, memory usage, SQL vs DataFrame ergonomics, and when to reach for each. Real numbers, honest verdict.https://www.danilchenko.dev/posts/go-generic-methods/Thu, 02 Jul 2026 08:21:35 +0000https://www.danilchenko.dev/posts/go-generic-methods/Go 1.27 finally adds generic methods: type parameters on methods at last. A hands-on tutorial on the syntax, the chaining it enables, and the interface catch.https://www.danilchenko.dev/posts/mem0-vs-cognee-vs-letta/Wed, 01 Jul 2026 08:23:14 +0000https://www.danilchenko.dev/posts/mem0-vs-cognee-vs-letta/Mem0 adds drop-in memory in 5 lines. Cognee builds a knowledge graph. Letta makes the agent its own memory. Tested all three — here's which fits your stack.https://www.danilchenko.dev/posts/sparse-attention-explained/Tue, 30 Jun 2026 08:24:33 +0000https://www.danilchenko.dev/posts/sparse-attention-explained/How sparse attention cuts LLM inference cost by 10x on long contexts. Covers DeepSeek NSA, MInference, H2O, and The Sparse Frontier's findings.https://www.danilchenko.dev/posts/perplexity-bumblebee-review/Mon, 29 Jun 2026 08:22:13 +0000https://www.danilchenko.dev/posts/perplexity-bumblebee-review/Bumblebee scans npm, PyPI, Go, MCP configs, and editor extensions for compromised packages, all without running a single install script. Hands-on review.https://www.danilchenko.dev/posts/ai-replace-software-engineers/Fri, 26 Jun 2026 08:25:11 +0000https://www.danilchenko.dev/posts/ai-replace-software-engineers/SignalFire tracked 80M+ companies: engineers are 55% of Big Tech hires, attrition is 9%, front-end roles dropped 25%. The real data on AI and dev jobs.https://www.danilchenko.dev/posts/mistral-ocr-4-tutorial/Thu, 25 Jun 2026 08:24:28 +0000https://www.danilchenko.dev/posts/mistral-ocr-4-tutorial/Extract text, tables, and structured JSON from PDFs using Mistral OCR 4 in Python. Bounding boxes, confidence scores, and a complete invoice pipeline.https://www.danilchenko.dev/posts/claude-code-workflows/Wed, 24 Jun 2026 08:24:44 +0000https://www.danilchenko.dev/posts/claude-code-workflows/Learn to build 4 real Claude Code workflow scripts for code audit, migration, security review, and research. Includes actual output and cost breakdowns.https://www.danilchenko.dev/posts/qwen-max-vs-opus-gpt/Tue, 23 Jun 2026 08:23:14 +0000https://www.danilchenko.dev/posts/qwen-max-vs-opus-gpt/Qwen 3.7 Max costs $7.50/M output vs Claude's $25. But it generates 4x more tokens per task. Full benchmark, cost-per-task math, and who wins for what.https://www.danilchenko.dev/posts/ai-layoffs-2026/Mon, 22 Jun 2026 08:24:39 +0000https://www.danilchenko.dev/posts/ai-layoffs-2026/185K tech workers cut in 2026, 1,115 per day. Gartner surveyed 350 execs and found zero ROI correlation. Deutsche Bank calls it AI redundancy washing.https://www.danilchenko.dev/posts/glm-5-2-review/Thu, 18 Jun 2026 08:25:00 +0000https://www.danilchenko.dev/posts/glm-5-2-review/GLM-5.2 scores 62.1 on SWE-bench Pro vs GPT-5.5's 58.6, ships under MIT, and costs $1.40/M input tokens. Benchmarks, pricing, and the China data question.https://www.danilchenko.dev/posts/litellm-vulnerability/Wed, 17 Jun 2026 08:32:06 +0000https://www.danilchenko.dev/posts/litellm-vulnerability/LiteLLM has been hit by RCE, SQL injection, privilege escalation, and a PyPI supply chain attack in 2026. Here's every CVE, who's affected, and how to fix it.https://www.danilchenko.dev/posts/rust-vs-go/Tue, 16 Jun 2026 08:23:54 +0000https://www.danilchenko.dev/posts/rust-vs-go/Rust vs Go compared with real benchmarks, salary data, and production use cases. Go for 80% of backends, Rust for the 20% where latency and memory matter.https://www.danilchenko.dev/posts/go-1-26-features/Sun, 14 Jun 2026 08:22:06 +0000https://www.danilchenko.dev/posts/go-1-26-features/Go 1.26 ships Green Tea GC by default (10-40% less overhead), experimental SIMD, runtime/secret, and a rewritten go fix. Hands-on code for each feature.https://www.danilchenko.dev/posts/claude-code-max-vs-pro/Sat, 13 Jun 2026 08:24:01 +0000https://www.danilchenko.dev/posts/claude-code-max-vs-pro/Claude Code Pro gives $20/mo in credits, Max 5x gives $100, Max 20x $200. After June 15, headless usage bills at API rates. What each plan actually costs.https://www.danilchenko.dev/posts/gpt-5-5-review/Fri, 12 Jun 2026 08:26:38 +0000https://www.danilchenko.dev/posts/gpt-5-5-review/GPT-5.5 hits 82.7% on Terminal-Bench and uses 72% fewer tokens than Claude — but loses SWE-Bench Pro to Opus 4.7. Seven weeks of real agentic use, reviewed.https://www.danilchenko.dev/posts/claude-fable-5-review/Thu, 11 Jun 2026 08:22:24 +0000https://www.danilchenko.dev/posts/claude-fable-5-review/Claude Fable 5 hits 80.3% SWE-bench Pro and 29.3% FrontierCode Diamond. It also costs 2x Opus 4.8, retains your data 30 days, and silently falls back.https://www.danilchenko.dev/posts/gemini-flash-vs-haiku/Wed, 10 Jun 2026 08:21:45 +0000https://www.danilchenko.dev/posts/gemini-flash-vs-haiku/Gemini 3.5 Flash vs Claude Haiku 4.5 vs MAI-Code-1-Flash — SWE-bench scores, token costs, and which flash model actually writes better code in 2026.https://www.danilchenko.dev/posts/opencode-vs-claude-code/Tue, 09 Jun 2026 08:22:00 +0000https://www.danilchenko.dev/posts/opencode-vs-claude-code/OpenCode (free, 75+ models) vs Claude Code ($20/mo, best SWE-bench) vs Cursor ($20/mo, IDE-native). Real pricing, benchmarks, and which one wins for your stack.https://www.danilchenko.dev/posts/devin-desktop-review/Sat, 06 Jun 2026 08:22:38 +0000https://www.danilchenko.dev/posts/devin-desktop-review/Windsurf became Devin Desktop on June 2. Cascade dies July 1. Here's what the rebrand, Devin Local, and ACP support mean after a week with the new IDE.https://www.danilchenko.dev/posts/project-glasswing/Fri, 05 Jun 2026 08:21:57 +0000https://www.danilchenko.dev/posts/project-glasswing/Claude Mythos found 10,000+ critical bugs in 8 weeks. Inside Project Glasswing — real numbers, the patching crisis, and why Anthropic won't release the model.https://www.danilchenko.dev/posts/agent-client-protocol/Thu, 04 Jun 2026 08:23:11 +0000https://www.danilchenko.dev/posts/agent-client-protocol/Set up Claude Code, Gemini CLI, or Codex in JetBrains and Zed via ACP — the open protocol that decouples AI coding agents from editors. Step-by-step configs.https://www.danilchenko.dev/posts/ai-replacing-junior-developers/Wed, 03 Jun 2026 08:23:47 +0000https://www.danilchenko.dev/posts/ai-replacing-junior-developers/Junior developer employment fell 20% since 2024 while ML engineer openings surged 59%. Stanford, BLS, and Forrester data on who's being cut and what works.https://www.danilchenko.dev/posts/uv-python-tutorial/Tue, 02 Jun 2026 08:17:14 +0000https://www.danilchenko.dev/posts/uv-python-tutorial/Step-by-step uv tutorial: create a real Python CLI project, manage deps, run tests, lint with ruff, build a wheel, and ship it in Docker — from scratch.https://www.danilchenko.dev/posts/copilot-ai-credits/Mon, 01 Jun 2026 08:22:06 +0000https://www.danilchenko.dev/posts/copilot-ai-credits/GitHub Copilot switched to AI credits on June 1. Token math per model, real session costs, and whether your $10/month Pro plan still makes sense.https://www.danilchenko.dev/posts/claude-code-hooks/Sun, 31 May 2026 08:21:09 +0000https://www.danilchenko.dev/posts/claude-code-hooks/Build 5 real Claude Code hooks step by step — auto-format, command blocker, AI linter, test runner, and context injector. Full configs and scripts included.https://www.danilchenko.dev/posts/swe-job-market-2026/Thu, 28 May 2026 08:21:06 +0000https://www.danilchenko.dev/posts/swe-job-market-2026/ML engineer roles up 59% from 2020, general SWE openings down 49%. Dice, CompTIA, and BLS data shows a 2026 market split by AI fluency. Salaries included.https://www.danilchenko.dev/posts/webmcp-tutorial/Wed, 27 May 2026 08:20:52 +0000https://www.danilchenko.dev/posts/webmcp-tutorial/Build a WebMCP-enabled website from scratch with declarative HTML forms and the JavaScript imperative API. Working code for Chrome 149's origin trial.https://www.danilchenko.dev/posts/agentic-coding-trends/Tue, 26 May 2026 08:23:23 +0000https://www.danilchenko.dev/posts/agentic-coding-trends/Anthropic's 2026 report claims coding agents will reshape software development. Here's what the 8 trends actually mean after running agents on production code.https://www.danilchenko.dev/posts/pyrefly-vs-mypy-vs-ty/Sun, 24 May 2026 08:22:48 +0000https://www.danilchenko.dev/posts/pyrefly-vs-mypy-vs-ty/Pyrefly 1.0 checks 1.85M lines/sec with 87.8% conformance, mypy 2.0 adds parallelism, ty hits sub-10ms updates. Full benchmarks and migration guide.https://www.danilchenko.dev/posts/efficient-llm-reasoning/Sat, 23 May 2026 08:23:44 +0000https://www.danilchenko.dev/posts/efficient-llm-reasoning/Seven papers fix LLM overthinking: Sketch-of-Thought cuts tokens 84%, shorter chains boost accuracy 34.5%, and budget-aware prompting halves costs.https://www.danilchenko.dev/posts/google-jules-review/Fri, 22 May 2026 08:21:56 +0000https://www.danilchenko.dev/posts/google-jules-review/Google Jules queues coding tasks, runs them in a cloud VM, and opens PRs while you sleep. Free tier gives 15 tasks/day. Here's what worked and what didn't.https://www.danilchenko.dev/posts/antigravity-cli-vs-claude-code/Thu, 21 May 2026 08:23:55 +0000https://www.danilchenko.dev/posts/antigravity-cli-vs-claude-code/Google's Antigravity CLI replaces Gemini CLI on June 18. Compared with Claude Code and Codex CLI on pricing, rate limits, benchmarks, and open-source status.https://www.danilchenko.dev/posts/ai-bug-bounty-crisis/Wed, 20 May 2026 08:20:45 +0000https://www.danilchenko.dev/posts/ai-bug-bounty-crisis/HackerOne paused payouts, Curl quit its bounty, Linux's security list is unmanageable. The AI vulnerability flood and the zero-days buried in the noise.https://www.danilchenko.dev/posts/ai-coding-productivity-paradox/Fri, 15 May 2026 08:22:16 +0000https://www.danilchenko.dev/posts/ai-coding-productivity-paradox/Developers think AI coding tools make them 24% faster. A controlled study measured 19% slower. Here's what five studies found — and what to do about it.https://www.danilchenko.dev/posts/spec-driven-development/Thu, 14 May 2026 08:22:52 +0000https://www.danilchenko.dev/posts/spec-driven-development/Build a complete Python CLI expense tracker using GitHub Spec Kit and Claude Code. Full walkthrough from requirements to working code in 30 minutes.https://www.danilchenko.dev/posts/moondb-backend-vibe-coding/Wed, 13 May 2026 11:12:16 +0000https://www.danilchenko.dev/posts/moondb-backend-vibe-coding/Build a full backend with REST API, auth, file storage, and AI endpoints by sending one JSON schema to MoonDB — zero deploy, zero config, ready in 5 minutes.https://www.danilchenko.dev/posts/thinc-code-reasoning/Wed, 13 May 2026 08:24:46 +0000https://www.danilchenko.dev/posts/thinc-code-reasoning/THINC trains a 4B parameter model to reason entirely in code. It scored 78.1% on competition math, beating Qwen3-235B at 75.2%. Here's how the method works.https://www.danilchenko.dev/posts/chrome-devtools-mcp/Tue, 12 May 2026 08:17:00 +0000https://www.danilchenko.dev/posts/chrome-devtools-mcp/Set up Chrome DevTools MCP with Claude Code, Cursor, or Gemini CLI in 5 minutes. 44 tools for debugging, performance, network, and browser automation.https://www.danilchenko.dev/posts/deepseek-v4-pro-review/Sat, 09 May 2026 08:30:04 +0000https://www.danilchenko.dev/posts/deepseek-v4-pro-review/DeepSeek V4 Pro scores 80.6% on SWE-bench Verified at $1.74/M input tokens — 7x cheaper than Claude Opus 4.7. Real benchmarks, costs, and safety gaps.https://www.danilchenko.dev/posts/gemini-cli-tutorial/Fri, 08 May 2026 08:24:33 +0000https://www.danilchenko.dev/posts/gemini-cli-tutorial/Set up Google's free Gemini CLI in 5 minutes, configure GEMINI.md, add MCP servers, and build a Python project — all on the 1,000 requests/day free tier.https://www.danilchenko.dev/posts/ai-agent-guardrails/Thu, 07 May 2026 08:22:31 +0000https://www.danilchenko.dev/posts/ai-agent-guardrails/AI agent guardrails from 4 real production wipes — PocketOS, Replit, Amazon. Scoped tokens, destructive-action gates, isolated backups, plan-first mode.https://www.danilchenko.dev/posts/ai-infrastructure-layoffs/Tue, 05 May 2026 08:22:46 +0000https://www.danilchenko.dev/posts/ai-infrastructure-layoffs/$725B in AI capex, 100K+ layoffs, and 275K unfilled AI roles. The numbers behind big tech's 2026 human-to-GPU trade — and what it means for your career.https://www.danilchenko.dev/posts/markitdown-vs-docling-vs-marker/Sun, 03 May 2026 08:20:42 +0000https://www.danilchenko.dev/posts/markitdown-vs-docling-vs-marker/Three open-source PDF-to-Markdown tools for RAG and LLM pipelines, tested on real documents. Speed, table fidelity, install pain, and which one to pick.https://www.danilchenko.dev/posts/gpt-claude-gemini-coding/Fri, 01 May 2026 08:25:00 +0000https://www.danilchenko.dev/posts/gpt-claude-gemini-coding/Three weeks rotating between GPT-5.4, Claude Opus 4.7, and Gemini 3.1 Pro on real coding work — benchmarks, token costs, and the per-task winner for each.https://www.danilchenko.dev/posts/fastmcp-mcp-server/Wed, 29 Apr 2026 08:24:49 +0000https://www.danilchenko.dev/posts/fastmcp-mcp-server/Build a production-ready MCP server in Python with FastMCP 3.2 — tools, resources, prompts, GitHub OAuth proxy, MCP Inspector, and Claude Desktop hookup.https://www.danilchenko.dev/posts/python-t-strings/Mon, 27 Apr 2026 08:21:49 +0000https://www.danilchenko.dev/posts/python-t-strings/Python 3.14 t-strings (PEP 750) let you intercept values before string interpolation. Hands-on tutorial with SQL safety, HTML escaping, and structured logging.https://www.danilchenko.dev/posts/ai-brain-fry/Wed, 22 Apr 2026 06:03:20 +0000https://www.danilchenko.dev/posts/ai-brain-fry/AI brain fry hit 14% of AI-using workers in the BCG-Harvard study. Devs near the top, behind marketing and HR. Productivity drops once you cross 4 tools.https://www.danilchenko.dev/posts/asynctls-sparse-attention/Wed, 22 Apr 2026 00:06:00 +0000https://www.danilchenko.dev/posts/asynctls-sparse-attention/AsyncTLS sparse attention fuses block filtering, token selection, and async KV cache offloading for 1.3-4.7x throughput gains at 48k-96k token contexts.https://www.danilchenko.dev/posts/composer-2-review/Tue, 21 Apr 2026 04:04:27 +0000https://www.danilchenko.dev/posts/composer-2-review/Cursor Composer 2 ships at $0.50/M input — roughly 1/10 of Opus 4.6 — and beats Opus on Terminal-Bench. Then a developer found Kimi K2.5 in the model ID.https://www.danilchenko.dev/posts/go-iterators/Tue, 21 Apr 2026 00:06:00 +0000https://www.danilchenko.dev/posts/go-iterators/How to write a custom Go iterator using iter.Seq and iter.Seq2, handle early termination, convert push to pull with iter.Pull, and avoid the yield traps.https://www.danilchenko.dev/posts/cyprus-developer-salary-2026/Mon, 20 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/cyprus-developer-salary-2026/Cyprus software engineer salary 2026 — real levels.fyi numbers for Limassol vs Nicosia, plus how the new 50% exemption above €55K flips net pay vs Berlin.https://www.danilchenko.dev/posts/hetzner-vs-digitalocean/Sun, 19 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/hetzner-vs-digitalocean/Hetzner just raised cloud prices 30–37% on April 1. Real migration numbers, current per-month costs, and when DigitalOcean still wins for your stack.https://www.danilchenko.dev/posts/meta-layoffs-may-2026/Sat, 18 Apr 2026 08:30:00 +0000https://www.danilchenko.dev/posts/meta-layoffs-may-2026/Meta is cutting around 8,000 jobs starting May 20, about 10% of its workforce. AI-assisted productivity is the stated reason, with more cuts later in 2026.https://www.danilchenko.dev/posts/claude-code-subagents/Sat, 18 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/claude-code-subagents/Claude Code subagents run side tasks in their own context window. Here's how to create them, when to pick one over a skill, and the mistakes to avoid.https://www.danilchenko.dev/posts/recursive-language-models/Sat, 18 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/recursive-language-models/Recursive language models treat a huge prompt as a Python variable the model can grep and recurse over. MIT's paper shows it beats GPT-5 on long context.https://www.danilchenko.dev/posts/agentic-memory-llm/Fri, 17 Apr 2026 10:00:00 +0000https://www.danilchenko.dev/posts/agentic-memory-llm/A new paper from Alibaba teaches LLM agents to store, update, and delete their own memory via reinforcement learning. Beats Mem0 and A-Mem on 5 benchmarks.https://www.danilchenko.dev/posts/polars-vs-pandas/Fri, 17 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/polars-vs-pandas/Polars vs Pandas in 2026: performance benchmarks with real numbers — 3-11x speedups on large datasets, but Pandas still wins for small data and ML pipelines. When to switch and when to stay.https://www.danilchenko.dev/posts/ty-vs-mypy-vs-pyright/Thu, 16 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/ty-vs-mypy-vs-pyright/mypy vs Pyright vs ty — full comparison of speed, conformance, and IDE integration. Pyright leads at 98% spec conformance, ty is 10-60x faster but passes only 53% of tests. Real project benchmarks included.https://www.danilchenko.dev/posts/claude-code-vs-codex-cli/Wed, 15 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/claude-code-vs-codex-cli/Claude Code wins on code quality (81% SWE-bench). Codex CLI wins on speed and uses 4x fewer tokens. Side-by-side pricing, benchmarks, and best use cases.https://www.danilchenko.dev/posts/tech-layoffs-q1-2026/Wed, 15 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/tech-layoffs-q1-2026/78,557 tech workers lost jobs in Q1 2026 and nearly half the cuts cite AI. Here's which companies cut deepest, which roles vanished, and what's actually hiring.https://www.danilchenko.dev/posts/cursor-vs-github-copilot-real-cost-2026/Tue, 14 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/cursor-vs-github-copilot-real-cost-2026/Cursor Pro's $20/mo base hits $40–80 under heavy agent use. Copilot Pro+ is $39/mo with 1,500 requests. Full cost breakdown with every tier compared.https://www.danilchenko.dev/posts/python-314-free-threading/Mon, 13 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/python-314-free-threading/Python 3.14 drops the GIL for good. Real benchmarks show 3.5x speedups on 4 cores. See which libraries break and how to install the free-threaded build.https://www.danilchenko.dev/posts/junior-developer-jobs-2026/Sun, 12 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/junior-developer-jobs-2026/Entry-level developer hiring dropped 67% since 2022. But the full story is more complicated than the doomsday headlines suggest, and more useful for your career.https://www.danilchenko.dev/posts/uv-vs-pip-vs-poetry/Sun, 12 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/uv-vs-pip-vs-poetry/uv vs Poetry compared head-to-head — install speed, lock file resolution, feature comparison, plus pip benchmarks. Which Python package manager should you use in 2026?https://www.danilchenko.dev/posts/2026-04-11-github-ai-agents-pull-requests/Sat, 11 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-11-github-ai-agents-pull-requests/AI agents pushed 17 million pull requests to GitHub last month. The platform buckled with five outages in two days and shipped a kill switch to disable PRs.https://www.danilchenko.dev/posts/2026-04-11-triattention-kv-cache-compression-long-reasoning/Sat, 11 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-11-triattention-kv-cache-compression-long-reasoning/TriAttention uses pre-RoPE vector concentration and trigonometric scoring to compress KV cache 10.7x while matching full attention accuracy on reasoning tasks.https://www.danilchenko.dev/posts/2026-04-10-mempalace-review-ai-memory-system-milla-jovovich/Fri, 10 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-10-mempalace-review-ai-memory-system-milla-jovovich/MemPalace's 100% LongMemEval claim was hand-tuned. The real 96.6% score still beats Mem0 and Zep for free. Honest verdict after running the benchmarks.https://www.danilchenko.dev/posts/2026-04-09-claude-emotion-vectors-blackmail-cheating/Thu, 09 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-09-claude-emotion-vectors-blackmail-cheating/Anthropic found 171 emotion vectors inside Claude Sonnet 4.5 that causally shape behavior. Amplifying the desperation vector pushed blackmail from 22% to 72%.https://www.danilchenko.dev/posts/2026-04-07-run-gemma-4-locally-ollama-llama-cpp-vllm/Tue, 07 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-07-run-gemma-4-locally-ollama-llama-cpp-vllm/Step-by-step guide to running Google Gemma 4 locally on your hardware with Ollama, llama.cpp, and vLLM — including model picks, VRAM requirements, and real gotchas.https://www.danilchenko.dev/posts/2026-04-06-ai-scientist-v2-first-peer-reviewed-ai-paper/Mon, 06 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-06-ai-scientist-v2-first-peer-reviewed-ai-paper/AI Scientist-v2 from Sakana AI produced the first fully AI-generated paper to pass peer review at ICLR. Here's how the agentic tree search system works and why it matters.https://www.danilchenko.dev/posts/2026-04-06-apfel-review-free-local-ai-mac/Mon, 06 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-06-apfel-review-free-local-ai-mac/Apfel exposes Apple's hidden 3B on-device LLM from the command line. I tested it for shell scripting, summaries, and code. Here's what works.https://www.danilchenko.dev/posts/2026-04-05-claude-found-500-zero-days-llm-vulnerability-research/Sun, 05 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-05-claude-found-500-zero-days-llm-vulnerability-research/Claude discovered 500+ zero-days in Linux, FreeBSD, Firefox, and Ghost — including a 23-year-old NFS bug. Inside the bash-script pipeline Anthropic used.https://www.danilchenko.dev/posts/2026-04-03-deepseek-mhc-manifold-constrained-hyper-connections/Fri, 03 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-03-deepseek-mhc-manifold-constrained-hyper-connections/DeepSeek's mHC uses the Sinkhorn-Knopp algorithm to fix training instability in hyper-connections. Here's how doubly stochastic matrices stabilize LLM scaling.https://www.danilchenko.dev/posts/2026-04-02-emergent-misalignment-fine-tuning-llm-persona-features/Thu, 02 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-02-emergent-misalignment-fine-tuning-llm-persona-features/Emergent misalignment research shows fine-tuning LLMs on insecure code triggers broad harmful behavior. OpenAI's SAE analysis found the persona features behind it.https://www.danilchenko.dev/posts/2026-04-01-error-cascades-multi-agent-llm-systems/Wed, 01 Apr 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-04-01-error-cascades-multi-agent-llm-systems/AutoGen, CrewAI, LangGraph: 5 of 6 multi-agent LLM frameworks hit 100% error infection. A genealogy graph defense lifts the catch rate from 32% to 89%.https://www.danilchenko.dev/posts/2026-03-31-diffusion-language-models-mercury-1000-tokens-per-second/Tue, 31 Mar 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-03-31-diffusion-language-models-mercury-1000-tokens-per-second/Mercury uses diffusion instead of autoregressive decoding to generate all tokens in parallel, hitting 1,000+ tokens/sec. We break down how it works.https://www.danilchenko.dev/posts/2026-03-30-four-color-theorem-near-linear-time-algorithm/Mon, 30 Mar 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-03-30-four-color-theorem-near-linear-time-algorithm/Four color theorem just hit O(n log n) — a 2026 paper proves planar graphs contain linearly many reducible configurations, beating the 30-year O(n²) bound.https://www.danilchenko.dev/posts/2026-03-27-google-turboquant-llm-compression-6x-zero-accuracy-loss/Fri, 27 Mar 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-03-27-google-turboquant-llm-compression-6x-zero-accuracy-loss/Google's TurboQuant algorithm compresses LLM KV cache memory by 6x with zero accuracy loss and no retraining needed. We break down the ICLR 2026 paper.https://www.danilchenko.dev/posts/2026-03-26-claude-dispatch-vs-openclaw-vs-mariner/Thu, 26 Mar 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-03-26-claude-dispatch-vs-openclaw-vs-mariner/Claude Dispatch (), OpenClaw (free + API), Google Mariner () compared after a week with each desktop AI agent. Cost, setup, security, and which to pick.https://www.danilchenko.dev/posts/2026-03-24-ai-coding-tools-secret-leaks/Tue, 24 Mar 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-03-24-ai-coding-tools-secret-leaks/GitGuardian counted 28.65M leaked secrets in 2025. AI-assisted commits leak 2x more, Claude Code hits 3.2%. Here's the pre-commit setup that blocks it.https://www.danilchenko.dev/posts/2026-03-24-cursor-vs-claude-code-vs-windsurf/Tue, 24 Mar 2026 06:00:00 +0000https://www.danilchenko.dev/posts/2026-03-24-cursor-vs-claude-code-vs-windsurf/A hands-on comparison of Cursor, Claude Code, and Windsurf after months of real daily use — pricing, features, and which tool fits your workflow best.https://www.danilchenko.dev/about/Thu, 01 Jan 2026 00:00:00 +0000https://www.danilchenko.dev/about/About Maksim Danilchenkohttps://www.danilchenko.dev/contact/Thu, 01 Jan 2026 00:00:00 +0000https://www.danilchenko.dev/contact/Get in touchhttps://www.danilchenko.dev/privacy/Thu, 01 Jan 2026 00:00:00 +0000https://www.danilchenko.dev/privacy/Privacy Policy for danilchenko.dev