Nate Jones is a respected developer and YouTube creator knowledgeable about AI and related issues.
Here he discusses his own use of AI as of early November 2025. While these are his personal choices which may or may not align with others’ choices, they do give at least an overview and some guidance about what versions of LLMs etc one might want to use for a specific purpose. See more on Nate B. Jones here.
AI is the mother of all power laws and rewards those who are willing to jump in and learn by doing.
1. ChatGPT (GPT-5 / GPT-5 Thinking Mode)
Use: Analysis and idea development (“thinking, not writing”). Handles long context sessions, brainstorming, and logical structuring.
Pros:
• Excellent for analysis and conceptual clarity.
• Large context window for deep conversation.
• “Thinking mode” maintains reasoning quality; “fast mode” digests large text quickly.
Cons/Pitfalls:
• Weak “default voice” for finished writing.
• Output tone often generic without detailed prompting.
• Limited for PowerPoint and Excel beyond simple CSV tables.
• Best for reasoning—not final deliverables.
2. Claude Sonnet 4.5 (Anthropic)
Use: Primary writing assistant, plus Excel analysis and moderate PowerPoint/Word work.
Pros:
• Excellent at following user tone and voice when given samples.
• Strong instruction-following; good for iterative drafting.
• Useful for editing existing Excel files and doing analytic breakdowns.
• Handles Word formatting well if prompted precisely.
Cons/Pitfalls:
• Context-window limits cause breakdowns on large files or decks.
• Requires “chunking” (divide work into 5–8-slide or modular tasks).
• Less visual polish than specialized PowerPoint AIs.
• Must restart chat with smaller tasks after context overrun—no automatic recovery.
3. Kimi K2 (Chinese open-source model)
Use: PowerPoint generation; turns prompts into full slide decks.
Pros:
• Produces visually polished, clean presentations from minimal prompts.
• Outstanding for personal or public-domain material.
Cons/Pitfalls:
• Data-protection risk: servers located in China—unsuitable for corporate or confidential use (US/EU compliance issues).
• Not strong for reasoning or “thinking” tasks; benchmark results don’t match real-world performance.
• Style presets may constrain design flexibility.
4. Perplexity AI
Use: General web search, research summaries, and report drafting.
Pros:
• Fast, accurate research aggregation; “Labs” mode enables exploration and report generation.
• Good for broad-topic discovery.
Cons/Pitfalls:
• Weak for very recent or social-media-based information.
• Relies on publicly indexed data—can miss fresh or niche updates.
• Needs human interpretation; summaries can flatten nuance.
5. Grok (“Grok 4”)
Use: Real-time social-network monitoring and sentiment discovery (Reddit, X/Twitter).
Pros:
• Excellent at scraping and summarizing emerging discussions.
• Good for tracking reactions to AI launches or trending events.
Cons/Pitfalls:
• Not trustworthy for in-depth reasoning or structured research.
• Contextual accuracy can be low; summaries often need manual review.
6. Comet Browser
Use: Daily general-purpose browser integrating Perplexity and lightweight agent actions.
Pros:
• “Generative you” feature drafts LinkedIn or message replies for approval.
• Data-in/out connections reduce need to visit sites directly.
• Embeds Perplexity-powered search beside pages for analysis and off-page actions.
Cons/Pitfalls:
• Agent autonomy limited; still experimental.
• Requires careful review before sending generated communications.
7. Atlas Browser (ChatGPT-first design)
Use: AI-enhanced browsing and coding assistance using ChatGPT memory and reasoning.
Pros:
• Personalized: model “remembers” user preferences and prior interactions.
• ChatGPT-based search plus in-tab reasoning for contextual awareness.
• Integrates coding comprehension—can review GitHub repos or drive builds via Lovable.
Cons/Pitfalls:
• Replaces Google search with ChatGPT search—different results and filtering.
• Conservative “safety-first” agent design; some sensitive actions (banking, etc.) disabled.
• Users must accept OpenAI’s agentic-future approach; limited customization of autonomy.
8. Claude Code
Use: Task-running, file manipulation, and code execution in connected environments.
Pros:
• Friendly interaction style; connects to local files and MCP servers.
• Acts autonomously yet reports back (good balance of initiative and safety).
• Integrates smoothly with Anthropic’s wider ecosystem.
Cons/Pitfalls:
• “Bias for action” can make it execute prematurely—requires user oversight.
• Context and memory boundaries similar to Claude Sonnet.
9. Codeex
Use: Strategic reasoning and debugging at the command line.
Pros:
• Excellent “strategic thinker” for coding and shell analysis.
• Concise, logical feedback for problem solving and bug fixing.
• Complements Claude Code’s action bias with more deliberation.
Cons/Pitfalls:
• Less integrated environment (fewer cloud/file connections).
• Primarily analytical; not meant for autonomous workflows.
10. Overall Stack Insights
• Division of labor: ChatGPT for thought and logic; Claude Sonnet 4.5 for writing and spreadsheets; Kimi K2 for slide visuals; Perplexity + Grok for research; Comet + Atlas for browsing; Claude Code + Codeex for engineering.
• Common pitfalls: context limits, privacy compliance, over-automation, and model benchmarking hype vs. lived usefulness.
• Key theme: Jones stresses ownership of all output—AI as collaborator, not substitute.
TL;DR Summary
Nate Jones’s stack treats AI tools as specialized collaborators: ChatGPT (GPT-5) for reasoning; Claude Sonnet 4.5for expressive writing and spreadsheet work; Kimi K2 for slides (non-sensitive data only); Perplexity and Grok for research and social sentiment; Comet and Atlas for AI-augmented browsing; and Claude Code / Codeex for coding.
The recurring lessons:
– Match each AI to its strength.
– Expect context and privacy constraints.
– Benchmark scores don’t equal usefulness.
– Always remain accountable for the finished work.