The AI Tools Landscape

Big idea

There is no single best AI tool — the major options differ in what they are built into and tuned for, so the smart move is matching the tool to the task instead of defaulting to one out of habit.

The idea

"Which AI should I use?" doesn't have one answer, because the major tools aren't identical products with different logos — they differ in real, practical ways: what they're trained on, what they're integrated into, and what they're actually optimized for.

Think of it like

A golfer's bag holds many clubs: a driver for long shots off the tee, irons for the middle of the course, a wedge for getting out of sand, and a putter for the green. You could technically putt with a driver, but it would be clumsy and slow. Club designs also improve every year, yet a putter still looks like a putter. AI tools work the same way: general chat assistants, assistants woven into a product suite, and coding helpers living inside your editor are all useful, but each shape is built for a different kind of shot.

Where the major tools differ

• ChatGPT (OpenAI) — broad general-purpose use, a large plugin/GPTs ecosystem, strong for everyday writing, brainstorming, and casual coding help.
• Claude (Anthropic) — strong at careful, structured reasoning, longer documents, and agentic coding tasks; the model family this project itself runs on.
• Gemini (Google) — deep integration with Google's own products (Search, Docs, Android) and strong multimodal (text + image + video) handling.
• Copilot (GitHub/Microsoft) — built specifically for the coding workflow, living inside your editor rather than a separate chat window, tuned for in-context code completion.

None of these differences are permanent or absolute — every vendor iterates constantly — but the shape of the difference (general chat vs. deep product integration vs. in-editor coding) tends to persist even as the underlying models improve.

Picking the right one for a task

The practical question is rarely "which is smartest" in the abstract — it's "which fits this task." Editing code inside your actual project favors an in-editor tool like Copilot or an agentic coding tool. A long, careful piece of writing or analysis favors a model known for structured reasoning. A quick question tightly coupled to your email or documents favors whichever tool is already integrated into that ecosystem. Cost, response speed, and how much context window you need also factor in — a huge document to analyze needs a large context window regardless of which vendor otherwise looks best.

Running models locally: what fits where

Local models are usually run quantized to 4 bits per parameter — about 0.5–0.6 GB of memory per billion parameters, plus some working space:

ollama pull llama3.2        # ~2 GB download, 3B parameters
ollama run llama3.2 "Explain a hash map in two sentences."

Bigger is smarter, but slower and hungrier. For summarizing, rewriting, and simple Q&A a 3–8B model is often enough; complex reasoning and coding still favour the largest hosted models.

Key terms

• Multimodal — Able to work with more than one kind of input, such as text, images, and video together.
• Context window — How much text a model can take in at once, which matters when analyzing a very large document.
• Ecosystem integration — How deeply a tool is built into other products you already use, like email, documents, or search.
• Agentic coding tool — A tool that works on code inside your actual project, taking multiple steps rather than just chatting.

Why it matters

Loyalty to a single tool for every task is usually a worse default than matching the tool to the job — the same way a good craftsperson doesn't use one tool for every task just because it's the one already in hand.