A Guide to Which AI to Use in the Agentic Era

A new guide to selecting AI tools argues that "using AI" has shifted from back-and-forth chatbot conversations to assigning tasks to agents that use tools. Because of that change, the guide says three things now matter when deciding what to use: Models, Apps, and Harnesses. It cites a test in which the same model, Claude Opus 4.6, was asked the same question across three apps and harnesses, producing out-of-date information with no harness, updated information and verifiable sources on Claude.ai, and a formatted head-to-head comparison in Claude Cowork.

The guide defines models as the underlying AI brains, naming GPT-5.2/5.3, Claude Opus 4.6, and Gemini 3 Pro as the big three, and notes version numbers may change in coming weeks. Apps are the products used to talk to a model, such as chatgpt.com, claude.ai, and gemini.google.com, plus coding tools like OpenAI Codex and Claude Code and desktop tools like Claude Cowork. Harnesses are systems that let a model use tools, take actions, and complete multi-step tasks.

On models, the guide says top models are close in overall capability, free models are optimized for chat rather than accuracy, and serious use requires paying at least $20 a month, which provides model choice and access to frontier models. It identifies Claude Opus 4.6 from Anthropic, Google's Gemini 3.0 Pro, and OpenAI's ChatGPT 5.2 Thinking as the big three frontier models, and says the other companies have fallen behind for now. It describes GPT-5.2 as multiple models, from mini to Thinking to Pro, with selection determining auto mode or manual choice. For Gemini, it lists Gemini 3 Flash, Gemini 3 Thinking, and 3 Pro, with Deep Think on the Ultra plan. For Claude, it advises Opus 4.6 with extended thinking, noting Sonnet 4.6 is powerful but not quite as good.

The guide says chatbot apps now differ in bundled features: Gemini offers nano banana, Veo 3.1, Guided Learning, and Deep Research; ChatGPT offers image creation, Study and Learn, a Quizzes creator, Deep Research, and Shopping Research; Claude offers Deep Research, with a study mode via Projects. It says OpenAI and Anthropic have clear leads over Google in harness capabilities. It also describes other apps and harnesses, including Claude Code, OpenAI Codex, Google Antigravity, Claude for Excel and PowerPoint, Claude Cowork, Google's NotebookLM, and the open-source OpenClaw, which it says is a serious security risk.