A Guide to Which AI to Use in the Agentic Era
The Core Framework: Models, Apps, and Harnesses
AI usage has shifted from simple chatbots to agents that can use tools and complete multi-step tasks autonomously. To choose the right tool, you must now evaluate three distinct layers:
- Models — The underlying AI “brain” (e.g., GPT-5.2, Claude Opus 4.6, Gemini 3 Pro). Determines reasoning, writing, coding, and analysis capability.
- Apps — The actual interface you use (e.g., chatgpt.com, claude.ai, gemini.google.com). These bundle features like voice, image creation, and data connectors.
- Harnesses — The system that lets the AI take action—using tools, browsing, executing code, and completing tasks.
Critical insight: The same model behaves very differently depending on its harness. Claude Opus 4.6 in a chat window is a different experience from Claude Opus 4.6 inside Claude Code autonomously building software for hours.
The Current Frontier Models
The “big three” are remarkably close in capability, but you must pay at least ~$20/month to access the advanced frontier models. Free models are optimized for speed and chat, not accuracy.
| Platform | What to Pick for Serious Work | Notes |
|---|---|---|
| ChatGPT | GPT-5.2 Thinking Extended ($20 plan) or Thinking Heavy (higher tier). For very hard problems, use GPT-5.2 Pro (most expensive). | Default “GPT-5.2” is often “auto” mode that picks weaker models. Always manually select. |
| Claude | Opus 4.6 with “extended thinking” turned on. | Sonnet 4.6 is powerful but not quite as good as Opus. |
| Gemini | Gemini 3 Pro or 3 Thinking. | Gemini Deep Think (on Ultra plan) for very hard problems. |
“When people say ‘Claude is better at writing’ or ‘ChatGPT is better at math,’ they’re talking about models.”
Chatbot Interfaces: Feature Comparison
Bundled AI Features (the “plus button”):
- Gemini: nano banana (best image generation), Veo 3.1 (video), Guided Learning (tutor mode), Deep Research.
- ChatGPT: Create Images, Study and Learn, Quizzes, Deep Research, Shopping Research.
- Claude: Deep Research only (study mode available by creating a Project).
Harness & Work Capability:
- ChatGPT & Claude: Can write and execute code, generate files (spreadsheets, PowerPoints), provide citations, and do extensive research.
- Gemini: Website harness is currently much weaker. In Mollick’s testing, Gemini could not produce working spreadsheets or PowerPoints and did not provide citations for research questions.
- All three: Offer data connectors (email, calendar, files), though specific integrations vary.
“For most people doing real work, the most important additional feature is Deep Research and connecting AI to your content.”
Specialized Apps and Harnesses
Coding & General Autonomous Agents
- Claude Code, OpenAI Codex, Google Antigravity — Give AI access to your codebase, a terminal, and the ability to write/run/test code autonomously.
- Claude Cowork — “Claude Code for non-technical work.” A desktop agent that works with local files and your browser in a secure VM. You describe an outcome, and it plans, breaks into subtasks, and executes.
- OpenClaw — Open-source, viral local agent. Runs on your machine, connects to any model, browses, manages files, sends email. Warning: Serious security risk.
Productivity & Knowledge Work
- Claude for Excel / PowerPoint — Deep integration allowing AI to act as a “junior analyst” inside spreadsheets. Easy to verify results because they live in Excel.
- NotebookLM — Google’s tool for synthesizing large amounts of information. Upload papers, videos, websites, and files to build an interactive knowledge base. Can generate slides, mind maps, and AI podcasts. Best for students, researchers, or anyone processing documents.