Running a Software Repository with Grok Bot Agents
Ray Fernando turns Grok Bot into a repository coordinator that delegates coding work, tracks pull requests and uses specialized agents to keep delivery moving.
AI Coding
Videos about using AI models and agents to write, review, debug, and maintain software. 40 videos.
Building a Local Dictation App with Claude Code
Pat Simmons builds a private macOS dictation app with Claude Code, using local speech models for fast transcription, custom vocabulary and system-wide text insertion.
Claude Code New Features, Explained
Claude Code works more like a dependable AI employee when a project supplies shared context, scoped tickets, review standards, testable feedback loops, recurring routines and explicit permission boundaries.
Gemini 3.7 Flash Is Here: Testing Google's Best Model Yet
Gemini 3.7 Flash is fast and comparatively inexpensive, but its hands-on coding results improve on its predecessor without reaching uniformly reliable frontier performance.
GLM 5.3 Is HERE – Is THIS the BEST Open Model Yet?
GLM 5.3 showed persistence and strong 3D reasoning, but uneven coding and game results did not consistently match its benchmark expectations.
Grok 4.6 Is INSANE – Is THIS a Frontier Model?
Grok 4.6 produced excellent front-end and 3D results with solid games, placing it near frontier quality without leading every technical test.
Grok 4.6 is Claude Fable 5, but dirt cheap
Grok 4.6 delivered strong coding results at lower test cost, but its surrounding tools remain less complete for broad knowledge work than the leading desktop agent systems.
DeepSeek V4 Pro Is HERE – Is THIS the BEST Open Model Yet?
DeepSeek V4 Pro is a clear improvement over its preview, combining strong research and polished successes with slow execution and several incomplete or unreliable interactive results.
Three OpenAI Engineers Shipped A Million Lines. Your Ten-Hour Agent Run Starts Here.
Long-running coding agents work better when people progressively shape context through durable instructions, current state, project maps and review checkpoints.
Every AI Tool You Need to Be Using in August 2026
Alex Finn recommends choosing AI models and interfaces by task instead of expecting one system to handle planning, coding, design, mobile delegation, local work and collaboration equally well.
Meta Open Source Is Back - Muse Glimmer First Test!
Meta's open-weight Muse Glimmer model is fast and capable across several coding tasks, but its first hands-on results remain inconsistent.
Ling 3.0 Tiny First Test - Can a Model This Small Really Code?
Ling 3.0 Tiny is remarkably fast and capable for a compact model, but uneven coding and three-dimensional results keep expectations in check.
Ex-NASA dev reveals his Agentic Engineering Workflow
Maintainable agent-written software requires humans to decide product intent, architecture, program structure and testable vertical slices before agents implement the code.
Meta Muse Code Is HERE – Spark 1.2 & Meta’s NEW Coding Agent!
Bijan Bowen finds clear improvement in Muse Spark 1.2 and useful Muse Code agent modes, but uneven results and high test cost limit the value case.
Marketing Agents Masterclass (GROW your startup)
Marketing agents work best as narrow code-first systems that connect intent signals, enrichment, outreach, follow-up and content feedback while reserving model inference for real judgment.
OpenAI's keyboard is addicting and unfinished
The Codex Micro keyboard has appealing programmable hardware and useful status controls, but its first-run software makes common model, chat and transcription workflows harder than they should be.
Qwen 27B Multi-Agent Flight Simulator Test
Multi-agent coding worked best when an orchestration layer split the job into milestones, reviewed intermediate work and recovered from context failures, while the same local model working alone did not finish the application.
Qwen 3.8 Max Hands-On Review
Qwen 3.8 Max produced impressive code and polished web output, but its physical reasoning, computer use and 3D work remained inconsistent and expensive.
Don’t use AI Agents without using these 8 skills
David Ondrej's most useful agent skills turn recurring practices into reusable instructions for safety, isolation, delegation, guided setup, decision review and reliable long-running work.
GPT-5.6 Luna First Test – Hands-On With OpenAI’s CHEAPEST Model!
GPT-5.6 Luna delivers useful coding and visual output for very little money, but its strongest results sit beside strange interfaces, weak physical reasoning and uneven polish.
DeepSeek V4 Flash Is Insane: The Best Small Model Yet!
DeepSeek V4 Flash delivers unusually strong coding and interactive-generation results for its active size and price, but spatial reasoning and game logic remain inconsistent.
Ling 3.0 Flash First Test: A Surprisingly Good Coding Model!
Ling 3.0 Flash shows unusually capable coding and repair behavior for a small active model, although complex game logic and spatial tasks still fail inconsistently.
Agentic Engineering, explained by a 10x developer
David Ondrej and Thorsten Ball argue that stronger coding agents shift software work from typing and model micromanagement toward product judgment, clear context and asynchronous verification.
OPUS 5 is Actually INSANE
Opus 5 combines stronger coding, reasoning and self-verification with lower pricing, while its system card raises questions about model autonomy and safety.
Opus 5: No-Hype Full Review & Testing
Opus 5 won five practical blind comparisons through stronger visual execution and consistent task completion, often with lower token costs than Fable 5.
Claude Opus 5 Is INSANE – Is This the BEST Model Yet?
Opus 5 produces unusually detailed visual applications and games, but long runtimes and occasional missing elements still require careful review.
Claude Opus 5 DESTROYS Fable 5
Opus 5 wins most difficult coding comparisons, but verbose behavior and a weaker coding harness can make it less efficient as an everyday default.
Most Valuable Skill of 2026: Managing AI Agents
AI-native software teams work best when people manage parallel agents through clear context, secure boundaries, automated testing and deliberate decision points.
Poolside Laguna S2.1 Local-Model Review
Poolside Laguna S2.1 was persistent and unusually creative for a locally runnable model, but its coding output needed repeated repair and often remained incomplete.
Kimi K3 Frontier-Model Cost Comparison
Kimi K3 often approached frontier-model output at much lower cost, while Fable 5 remained strongest on the hardest 3D work and GPT-5.6 Sol won some faster everyday tasks.
Prompting Is Dead. Here Is How You Create Loops
Reliable agentic coding comes from a repeatable loop that separates specification, implementation and review while keeping each change visible and controllable.
Gemini 3.6 Flash Is Here - Testing Google's Best Model Yet!
Gemini 3.6 Flash is exceptionally fast and can package ambitious coding results well, but its visual, spatial and interface quality varies sharply across tasks.
GLM 5.2 vs Kimi K3: Which Open Source Model Wins?
Pat Simmons finds Kimi K3 substantially stronger and often more efficient for complex coding and design, while GLM 5.2 remains the cheaper choice for straightforward research writing.
Qwen 3.8 Max Preview Is Here - Is This the Best Open Model Yet?
Bijan Bowen finds Qwen 3.8 Max Preview visually inventive on some coding tasks but inconsistent on spatial reasoning, hardware work and complex scenes, making its open-weight release more notable than its current reliability.
Do these 5 things in Claude Fable 5 NOW
Alex Finn recommends using Fable 5 to rethink recurring workflows, build personal context, propose multiple directions, delegate browser tasks and reserve scarce high-end usage for judgment.
L8 Principal's Agentic Engineering Setup (just copy him)
David Ondrej and Kun Chen show how one supervising agent can coordinate parallel workers, escalate ambiguous decisions, validate generated code and expose services through agent-efficient interfaces.
Codex vs Fable: Which AI Agent Picked the Better Problem?
Nate B Jones finds Fable stronger at identifying strategically valuable problems while Codex is more dependable at executing bounded tasks, suggesting teams should separate problem discovery from implementation.
Kimi K3 Broad Capability Test
Pat Simmons finds Kimi K3 substantially better than Kimi K2.7 and competitive across many tasks, but inconsistent enough that frontier models still lead the hardest builds.
Kimi K3 Open-Weight Frontier Test
Bijan Bowen finds Kimi K3 to be the strongest open-weight model he has tested, with impressive 3D and agentic output despite high cost, slow reasoning and uneven reliability.
Inkling Open Model First Test
Bijan Bowen finds Thinking Machines' Inkling model unimpressive on his coding tests but potentially useful to enterprises that need a tunable US-based open-weight model.