Best of AI — January 2026

1
Article
Supabase·16w
Introducing: Postgres Best Practices
Supabase released Agent Skills for Postgres Best Practices, a collection of 30 rules across 8 categories to help AI coding agents write correct, performant Postgres code. The rules follow the Agent Skills open standard and cover critical areas like query performance, connection management, Row Level Security, schema design, and concurrency. The repository addresses common mistakes seen across hundreds of thousands of Postgres projects, such as missing indexes, RLS bypasses, and connection pool exhaustion. These best practices complement the Supabase MCP server by teaching agents proper judgment while the MCP server handles database connections and execution.
491
8
2
Article
Cassidy's blog·17w
Do not give up your brain
While AI tools like ChatGPT can be valuable assistants, over-reliance on them for basic tasks like writing emails or generating responses can atrophy critical thinking skills. People who depend on AI for communication often struggle in real-time conversations. Maintaining mental sharpness requires actively using your brain rather than defaulting to AI for every task. The key is treating AI as a tool to augment thinking, not replace it.
436
35
3
Article
Xavier Womack·15w
Claude: the #1 AI for programmers?
Claude outperformed ChatGPT and other AI models in a coding task involving Tauri and glassmorphic windows. While ChatGPT provided outdated code and hallucinations, Claude delivered precise and accurate solutions within minutes. The author suggests Anthropic prioritizes coding capabilities more than competitors, making Claude a top choice for programming assistance despite other models ranking higher on synthetic benchmarks.
371
53
4
Article
Web Nepal·17w
The Rise of Contextual Vibe Coding
Most developers use AI coding tools ineffectively by providing vague prompts without context. The key to productive AI-assisted coding is providing dense context about your stack, architecture, constraints, and intent before asking for code. LLMs function like fast interns with perfect recall but zero situational awareness, requiring explicit information about existing decisions, tradeoffs, and boundaries. Around 60% of AI-generated code requires edits because prompts lack system-level context, clear goals, constraints, and feedback loops.
262
37
5
Article
Wix Engineering·19w
Why I Stop Prompting and Start Logging: The Design-Log Methodology
The Design-Log Methodology addresses the "Context Wall" problem in AI-assisted coding by maintaining version-controlled markdown documents that capture design decisions before implementation. Instead of repeatedly explaining architecture to AI, developers create immutable design logs in a ./design-log/ folder that document problems, questions, answers, and implementation plans. The AI reads these logs first, asks clarifying questions in the document, and only codes after design approval. This approach transforms AI from a code generator into an architectural partner, enabling simple prompts, preventing context drift, and creating traceable implementation history.
242
11
6
Video
Fireship·17w
The unhinged world of tech in 2026...
A satirical look at tech trends for 2026 covering AI's plateau and continued hype cycle, the emergence of humanoid robots and wearable AI, struggles in the VR/AR space, chip shortage driving nuclear power resurgence, quantum computing breakthroughs with Google's Willow chip, digital IDs and central bank digital currencies in Europe, and JavaScript runtime evolution with Node.js TypeScript support, Deno bundling, and Bun's growing popularity. The job market shows mixed signals with AI creating 'code janitor' roles while threatening mid-level positions, though software engineering growth is still projected at 15% through 2034.
227
7
7
Article
Adventures in Nodeland·17w
The Human in the Loop
AI has fundamentally changed software development workflows, but human review and judgment remain critical. While AI can generate code rapidly for bug fixes and features, engineers must review every change to ensure correctness, security, and architectural fit. The bottleneck has shifted from coding speed to review capability. The role of programmers who simply execute tasks is obsolete, but software engineers and architects who provide judgment, understand context, and maintain accountability are more crucial than ever. The real risk isn't AI replacing developers, but creating a culture where shipping unreviewed AI-generated code becomes acceptable.
175
15
8
Article
Hacker News·18w
2026 is the Year of Self-hosting
CLI agents like Claude Code have made self-hosting dramatically easier by eliminating the need to manually configure Docker, compose files, and networking. Combined with affordable mini PCs and Tailscale for secure networking, setting up a home server with services like Vaultwarden (password manager), Immich (photo storage), and Plex is now accessible to software-literate users without sysadmin expertise. The author runs 13 services on a $379 Beelink mini PC using just 4GB RAM, managing everything through natural language prompts to Claude Code instead of searching documentation.
165
19
9
Article
Addy Osmani·19w
My LLM coding workflow going into 2026
A comprehensive workflow for using LLM coding assistants effectively in 2026. Start with detailed planning and specs before coding, break work into small iterative chunks, provide extensive context to the AI, choose appropriate models for each task, and maintain human oversight through rigorous testing and code review. Use version control aggressively with frequent commits, customize AI behavior with rules and examples, leverage automation as quality gates, and treat the AI as a powerful but fallible pair programmer requiring clear direction. The approach emphasizes that AI amplifies engineering skills rather than replacing them, with the developer remaining accountable for all code produced.
151
2
10
Article
SHAPeS·15w
We. Are. Screwed
A concerned reaction to Moltbook, described as 'Reddit for LLMs,' expressing alarm about AI systems developing unusual communication patterns and autonomous behaviors. The author worries about the implications of AI agents operating independently on the internet, potentially engaging in malicious activities like crypto scams and malware distribution, suggesting we may need defensive AI systems in response.
139
43
11
Article
MetalBear·17w
How Our Engineering Team Uses AI
MetalBear's engineering team shares practical experiences using AI coding tools while building mirrord, a Kubernetes development tool in Rust. AI proves most valuable for understanding unfamiliar code, exploring architectural alternatives, and generating scripts. It struggles with complex architectures and long-running reasoning tasks. The team finds ChatGPT most reliable for iteration, Gemini best for deep research but prone to losing context, and Claude Code somewhere in between. Success depends on scoping problems tightly and controlling context rather than which model is used. AI accelerates tedious and exploratory work but cannot replace deep system understanding.
127
4
12
Video
AICodeKing·17w
Antigravity 3.0 (New Upgrades): These New Updates make ANTIGRAVITY REALLY GOOD!
Antigravity 3.0 introduces three major updates: Skills (custom workflows triggered on-demand similar to Claude's code commands), revised rate limits (now weekly instead of 5-hour refresh for Pro users, though Ultra users remain unaffected), and Secure Mode (terminal permission controls with allow/deny lists). Skills enable team-shareable custom agent behaviors for tasks like code reviews and testing. The rate limit change disadvantages Pro tier users who may exhaust quotas faster, though access to models like Opus 4.5 and Gemini 3 Pro remains. Secure Mode addresses security vulnerabilities through configurable command execution permissions, from fully automatic to manual approval for each command.
123
5
13
Article
Decube·18w
What is Context Engineering?
Context Engineering is the practice of designing and operationalizing business meaning, data lineage, quality signals, and policy constraints so AI systems can reliably understand and act on enterprise data. Unlike prompt engineering (which focuses on how questions are asked), Context Engineering establishes what AI systems know before questions are posed. It comprises four core components: semantic context (business definitions), lineage context (data flow and dependencies), operational context (quality and reliability signals), and policy context (compliance and usage constraints). This foundation is critical for Agentic AI systems that reason and act autonomously, enabling them to assess risk correctly, explain decisions, and know when to escalate. Enterprises should prepare by inventorying critical data, unifying metadata into a single context layer, and exposing context through APIs for AI agent consumption.
123
9
14
Article
Elevate·15w
The 80% Problem in Agentic Coding
AI coding agents now handle 80% or more of code generation for early adopters, but this shift introduces new challenges. While agents accelerate initial development, they create "comprehension debt" where developers understand less of their own codebase. Common issues include assumption propagation, abstraction bloat, and sycophantic agreement without questioning premises. Teams see 98% more PRs merged but 91% longer review times. The shift works best for greenfield projects and small teams, but struggles with legacy codebases. Success requires treating AI as an orchestrator rather than a faster typewriter, focusing on declarative problem definition, automated verification, and maintaining architectural oversight. The transition fundamentally splits engineers between those who enjoy coding itself versus those who enjoy building products.
120
7
15
Article
David Heinemeier Hansson·18w
Promoting AI agents
AI coding agents have evolved significantly, moving beyond autocomplete to autonomous tools that can control terminals, run tests, and search documentation. Modern models like Claude Opus 4.5 and Gemini 3, when used with terminal interfaces like OpenCode, can produce production-grade code contributions. While not replacing programmers entirely, these agents enable supervised collaboration where developers review outcomes rather than write every line. The shift represents a paradigm change from pair programming to team-based workflows, though claims of 90%+ AI-generated code remain unrealistic for quality professional work.
113
10
16
Article
Daily Dose of Data Science | Avi Chawla | Substack·15w
[New] Generative UI for Agents
Generative UI is an emerging pattern where AI agents render actual UI components instead of just returning text responses. Unlike traditional chat interfaces, agents can now display weather cards, confirmation dialogs, data tables, and other interactive elements by selecting pre-built components and filling them with data at runtime. Three approaches exist: static (predefined components), declarative (component registry), and open-ended (raw HTML/iframes). Protocols like A2UI, AG-UI, and MCP Apps enable real-time bidirectional communication between agents and frontends. CopilotKit has open-sourced a complete implementation for React with integrations for LangGraph, CrewAI, and other agent frameworks. MiniMax also launched Agent Desktop, a desktop environment where AI agents can browse the web, manage files, and automate developer tasks.
110
4
17
Video
DevOps Toolbox·18w
The Most Underrated IDE.
Zed is a Rust-based, open-source code editor from the Atom team that emphasizes speed and Vim-first design. It offers extensive out-of-the-box features including native Vim keybindings, built-in AI assistance with their own Zeta model, real-time collaboration, integrated debugging, terminal with AI help, and multi-cursor editing. The editor requires minimal configuration compared to Neovim, supports multiple AI providers, and includes a CLI utility. While not a complete Vim replacement, it provides a compelling alternative for developers seeking a fast GUI editor with strong Vim support and modern AI features at $10/month or free without AI.
107
10
18
Article
InfoWorld·18w
Why ‘boring’ VS Code keeps winning
VS Code maintains 76% market share among professional developers in 2025, even as AI-first editors like Cursor and Google's Antigravity emerge. Microsoft's dominance stems from ecosystem lock-in, with many new tools forking VS Code's codebase rather than replacing it. GitHub Copilot reached 20 million users, benefiting from distribution through existing workflows. However, trust issues emerged in 2025 around forced AI adoption, API deprecations, and security vulnerabilities. Google's Antigravity shows promise but faces skepticism due to Google's history of discontinuing products. The competitive advantage lies not in features but in providing stable, integrated developer experiences that enterprises can rely on long-term.
104
20
19
Article
Valdemar·15w
The market is now pushing "Duolingo for coders."
Gamified coding education platforms that teach syntax through mobile apps and streak rewards are insufficient in an era where AI agents can deploy entire systems. The focus should shift from memorizing syntax to learning system architecture and leveraging AI tools effectively. Success requires architectural thinking rather than accumulating coding badges.
100
47
20
Video
The Coding Sloth·16w
I Have Spent 500+ Hours Programming With AI. This Is what I learned
AI coding assistants work best when you already know how to program and communicate clearly. Being extremely specific in prompts, breaking tasks into smaller pieces, providing technical context and documentation, and telling AI what not to do dramatically improves results. Using guidelines files, MCP tools for extended functionality, and verification methods helps reduce errors. AI amplifies existing habits—good engineering practices lead to better AI output, while poor habits get amplified too. The key is treating AI as a multiplier of your skills, not a replacement for thinking.
97
7
21
Article
Frontend Masters·15w
What Senior Engineers Need to Know About AI Coding Tools – Frontend Masters Blog
Senior engineers often struggle with AI coding tools not because they lack aptitude, but because they haven't learned prompt engineering techniques. Research shows that simple prompting patterns like chain-of-thought (adding "let's think step-by-step") can increase accuracy from 17.7% to 78.7%. Senior engineers have a natural advantage once they master these basics, as they already possess the judgment and domain knowledge to ask the right questions and identify what's missing in AI outputs. Learning fundamental prompting techniques, understanding when to use AI agents versus writing code manually, and knowing how to debug AI hallucinations are now essential skills for professional software development.
94
1
22
Article
UX Planet·16w
AI killed your job. Evolve.
AI is transforming specialized technical roles into commodities, shifting professional value from execution to outcome ownership. Historical examples like scribes and switchboard operators show how technology repeatedly eliminates specialized activities while creating new value. The future belongs to those who pivot from technical execution to strategic accountability, defining constraints, validating outputs, and ensuring business objectives rather than producing artifacts.
91
22
23
Article
The Last Week in AWS·17w
I Hope This Email Finds You Before I Do
A developer built an AI-powered email assistant called "Billie" to handle spam and low-effort pitches with passive-aggressive responses. The system uses AWS Lambda, Cloudflare Email Routing, Claude AI for classification and drafting, and SES for sending. Emails are classified into tiers (spam, low-effort pitches, podcast requests, real humans) with AI-generated responses that are technically professional but carry an undercurrent of menace. Shadow mode ensures human approval before sending, and an operator context panel allows real-time instruction updates without code changes.
86
9
24
Article
DigitalOcean Community·19w
Olmo 3: Fully Open-Source LLM from AI2 (Models, Data, & Code)
Olmo 3 is Allen AI's fully open-source large language model available in 7B and 32B parameter versions. The release includes complete access to models, training datasets (Dolma 3 with 9.3 trillion tokens), code, and training logs. The model uses a three-stage training pipeline: pretraining on Dolma 3 Mix, mid-training on Dolma 3 Dolmino for skill enhancement, and long-context extension on Dolma 3 Longmino. Post-training uses the Dolci suite with SFT, DPO, and RLVR techniques. The 32B model employs grouped query attention while the 7B uses multi-head attention. OlmoTrace enables tracing text back to training sources for auditing and contamination detection.
83
1
25
Article
Google Developers·15w
Tailor Gemini CLI to your workflow with hooks
Gemini CLI v0.26.0+ introduces hooks, a middleware-like system that lets developers customize the AI agent's behavior at specific lifecycle points. Hooks enable injecting custom context, enforcing security policies (like blocking secrets from being written to files), and automating workflows through scripts that run synchronously within the agent loop. The feature supports extensions, allowing bundled hooks to be installed with a single command. Examples include security scanners that prevent API keys from being committed and the "Ralph loop" extension that forces continuous iteration on difficult tasks.
81
5

See all AI archives