October 2025
GosuCoder reviews Cursor 2.0 and its new Composer 1 model versus SWE 1.5, arguing that owning models/infrastructure and the editor’s faster, multi-agent workflow could reshape AI coding agents.
Alex compares Abacus AI’s Deep Agent Desktop using Claude Sonnet 4.5 versus Kimi K2 to one‑shot build a real Next.js link shortener app, showing Sonnet’s smoother end‑to‑end success while Kimi requires more manual fixes.
The video argues that Amazon’s 30,000 layoffs are driven by reallocating operating costs to buy GPUs and expand AWS capacity, while traditional firms cut jobs because AI automation is already delivering measurable productivity, showing two distinct forces behind widespread layoffs.
A live, hands‑on walkthrough of Cursor 2.0 that shows which features truly change developer workflows—Composer (the new agentic coding model), the Agents view and Plan Mode, multi‑agent/background builds, browser tools, sandboxed terminals, and a stacked‑diffs workflow with Graphite—while shipping a real multi‑image upload feature and debugging issues along the way.
Theo explains new evidence that AI adoption is shrinking junior roles while increasing demand and pay for seniors, why managers replace junior headcount with AI, and how juniors can adapt by building trust and networks.
Theo explains why OpenAI’s eye‑popping losses are largely upfront training and infrastructure investments against rapidly growing subscription and API revenue, arguing the business could become highly profitable as model revenues outpace escalating costs.
Max explains that Meta’s reported 600 layoffs largely target the underperforming FAIR team as leadership consolidates AI efforts under the new TBD lab led by Alexander Wang—signaling a restructure and renewed focus, not the end of the AI hype.
A practical walkthrough of the best AI coding tools from free to $300 per month, comparing IDEs, agent features, API plans, and value at each budget tier.
Scott Tolinski and Wes Bos showcase a curated set of S‑tier MCP servers—including Sentry, SpotlightJS, Context7, Svelte’s official MCP, Cloudflare’s suite, Chrome DevTools/Playwright, Stripe, and Mastra—explaining how they supercharge modern dev workflows for debugging, docs, browser automation, infra, and orchestration.
Overview and hands-on demo of Opencode, an open-source, terminal-based coding agent with a model router (Zen), custom agents and commands, MCP integrations, session sharing, and Neovim integration.
A roundtable with Ray Fernando, Adam (GosuCoder), and Eric Provencher explores the hidden costs of “free” AI agents (data privacy, sustainability), compares models like Haiku 4.5 vs Sonnet 4.5, and dives into planning, retrieval, and community-driven practices for building and debugging AI-powered coding workflows.
A hands-on review of Claude Haiku 4.5 finds it a fast, cost-effective model that performs well in Claude Code and on coding evals (often near Sonnet 4.5) but struggles with prompt-based tool calling in client/root code, making it best for simpler to mid-level tasks and planned refactors.
Theo analyzes the “vibe coding” boom and recent traffic dips, arguing the initial novelty is fading for non‑dev users while some businesses still grow, and suggests these AI app builders may mostly serve as gateways that inspire a small share to learn real coding.
Review and hands‑on comparison of plan modes in Cursor, Claude Code, and Droid, concluding that Cursor with Claude Sonnet 4.5 delivers the most complete plans and best execution with minimal iteration.
Scott and Wes explain Chrome’s new DevTools MCP server—how it automates the browser (à la Puppeteer), surfaces console/network/performance data, takes screenshots, and enables debugging and performance feedback loops directly from editors and agents, along with current limitations and security considerations.
Web Dev Cody demonstrates a practical agentic coding workflow using tools like Claude and Cursor to fix a real bug, iteratively add features (avatars, timestamps, profile routing), and share prompting and context-engineering strategies for faster CRUD development.
CJ from Syntax argues that today’s AI coding tools are unpredictable, goal-seeking wrappers around foundational models that erode the joy and reliability of programming, and outlines disciplined workflows (plans, specs, small steps, tests, agents) before deciding to take a month off from AI to code manually.
Review and hands-on tests of Cursor’s “Cheetah” stealth coding model: blazing-fast web/UI generation with mid-level reliability, weaker Rust support, high token usage, and performance landing between Sonnet 4.5 and Grok Code Fast.
Theo argues that exposing many MCP tools degrades agents and that Cloudflare’s “code mode” approach—letting LLMs write TypeScript to call APIs in a sandbox—handles complexity better, reduces context bloat, and points to a future where code-first configs beat MCP-style patches.
Theo reviews Zhipu's GLM‑4.6, showing it can rival Claude Sonnet on coding and reasoning tasks with strong practical agent performance and much lower cost, while discussing benchmarks, token efficiency, and real-world coding agent runs.
How Claude Sonnet 4.5 and Claude Code 2.0 enable long‑running agent workflows using GitHub Actions and MCP integrations (e.g., Slack, Supabase), plus improved IDE features for practical coding and background tasks.
GosuCoder compares current AI coding agents and models—like Sonnet 4.5, GLM 4.6, Grok Code Fast, and GPT‑5 Codex—using instruction‑following tests and LLM‑as‑a‑judge scoring, highlighting performance, iteration behavior, and where each model works best.