STOP Using Claude Code RIGHT NOW (Here's Why)

Summary (technological concepts, product features, reviews/guides)

Main thesis / update

• The speaker says they haven’t used Claude Code in months because performance has been getting slower.
• They attribute the shift to the idea that OpenAI’s Codex has improved significantly, while Claude Code has declined.

---

Codex vs. Claude Code (core comparison)

Codex improvements highlighted

• Codex quality: The speaker describes Codex as “really, really good,” crediting the change primarily to quality improvements.
• Agent-style expansion: OpenAI is extending Codex beyond pure coding into daily task automation and agent-style workflows.
• Work-type selection: Users can choose the type of work, and Codex attempts to adapt to match that intent.
• Workflow portability/import: If you use Claude Workbench / Claude Co-work heavily, Codex offers a workflow feature that can make it easier to import existing workflows (given the “right architecture”), effectively moving workflows from Claude’s ecosystem into OpenAI’s.

What’s still missing for Claude (speaker’s analysis)

• The speaker suggests Claude may be declining long-term due to lack of compute resources, framing AI development as constrained by a supply/compute bottleneck—“AI is a supply problem.”
• They reference a prior explanation from last week (a “five layer cake” model).

---

Claude ecosystem progress (not ignored)

Even while claiming Claude Code has declined, the speaker highlights Claude’s ongoing momentum:

• Autodesk Fusion integration for 3D modeling, especially praised by architects who previously felt AI wasn’t yet helpful for their specific modeling workflows.
• Blender connectors / integrations, along with links to creative and design tooling, including:
  • Adobe Creative Cloud
  • Canva
  • Affinity
• These are positioned as enabling more advanced design workflows that Claude can plug into.

---

Cursor (acquisition + developer platform/SDK)

• Cursor, previously acquired by Elon Musk’s company (xAI / Rock AI mentioned), is described as improving.
• The speaker reports using Cursor heavily for coding and suggests many developers are moving from Claude Code to Cursor as Claude worsens.
• Cursor released its own SDK, described as enabling others to:
  • take the Cursor “harness”
  • build AI agents that behave like coders on your machine
  • treat it as becoming more commodity-like across model/agent ecosystems

---

Practical guide: switching models in VS Code

The speaker provides a workflow approach for power users using Visual Studio Code:

• Instead of relying on the Claude Code app UI, set up your environment so you can:
  • type Claude in the VS Code terminal (after installation)
  • run Claude from the terminal as part of your developer workflow
• Model switching is framed as easy:
  • If one provider falls behind (Claude or Codex), you can swap within the same code base by launching the other via terminal commands (example: launching Codex with Codex).
• The speaker contrasts this with standalone apps (Claude/OpenAI), noting their interfaces may be nicer, but argues the VS Code/terminal setup makes switching smoother.

---

Gemini usage notes

• The speaker claims Google Gemini is pushing forward strongly in creating websites, potentially positioning it as best-in-class for web creation.
• They also mention Gemini doing well with office/presentation workflows, reducing reliance on PowerPoint and moving toward more capable tooling.
• Tip: For companies using Google accounts, Gemini is suggested as a good way to manage and use token budgets.

---

Repo for design files / style prompting

• The speaker references a GitHub-style repository described as a “repo for getting a lot of design files,” intended to:
  • provide design/style examples to Claude (e.g., “design in this style”)
  • generate UX output aligned to a desired style
• Claimed benefit: helping web/UX designers train AI to act like a designer for specific UX styles.

---

OpenAI image generation lead (examples + use case)

Why OpenAI is leading (speaker’s view)

• The speaker says OpenAI is leading image generation, especially with recent ChatGPT model upgrades that improve design/image-related tasks.

Example use cases mentioned

• Converting a raw image into a menu
• Generating a visual brand teardown from a product (example: Olipop)

Reusability

• The speaker emphasizes the outputs are now good enough for marketing use, and can be reused later with other AI models for new contexts.

Relative strengths summary (speaker’s ranking)

• OpenAI: best for coding, agent use, and now images/design
• Gemini: previously strong/leader in website creation (and earlier design-related areas)
• Claude: said to be weaker for coding/image design compared to others, though still improving via integrations like Fusion/Blender

---

GitHub “skills” for coding agents

• The speaker recommends exploring coding resources on GitHub, credited to Matt Pocock (as mentioned).
• They claim these “skills” can improve AI agents for coding by making small but useful enhancements.

---

Main speakers/sources

• Primary speaker: the YouTube creator (unnamed in subtitles)
• Referenced sources/organizations:
  • OpenAI (Codex, ChatGPT image/design model)
  • Anthropic / Claude (Claude Code, integrations)
  • Cursor (and its SDK)
  • xAI / Rock AI (mentioned in relation to Cursor’s acquisition)
  • Google (Gemini)
  • Matt Pocock (GitHub coding “skills” resource)
• Community/tooling mentioned: Visual Studio Code, plus creative tools (Autodesk Fusion, Blender, Adobe Creative Cloud, Canva, Affinity)