"What Counts as a Tool?" — Defining the Hex Boundary

The hex constraint says six tools. But what counts as a tool? Does Claude Desktop and Claude Code count as one tool or two? If Notion has AI features baked in, does Notion count as an AI tool? What about free tools — do those count toward the six? What about a browser extension that adds AI to every text field? What about the AI features your operating system now ships with — do Apple Intelligence or Windows Copilot consume a hex slot?

These are real questions, and they matter, because the hex constraint only works if the boundary is clear. A vague rule gets vague compliance, and vague compliance is the same as no constraint at all. So here's how I define the boundary, and why.

The Definition

A tool counts toward your hex if it meets all three of these criteria:

It's a distinct product with its own interface or workflow. Claude is a tool. Claude's API accessed through a custom script is still Claude — one tool. But Cursor, which uses Claude's models under the hood, is a different tool because it has its own interface, its own workflow, and its own capabilities that Claude alone doesn't provide. The test is whether you interact with it differently and whether it enables something the base product doesn't.

You make a deliberate choice to use it. Built-in OS features don't count because you're not choosing to use them — they're part of the environment. Apple Intelligence's text suggestions, Windows Copilot's system integration, the AI spell-check in your email client — these are ambient features, not tools you reach for. The distinction is intentional use versus passive presence. If you actively open it, direct it, and use its output, it's a tool. If it runs in the background or activates without your explicit engagement, it's a feature of another product.

It has a meaningful capability footprint. A browser extension that adds GPT-powered autocomplete to text fields is a feature, not a tool. It does one small thing within another product's context. A full LLM subscription that you use for writing, research, and analysis is a tool — it has a broad capability footprint that shapes your workflow. The gray area is real, and the judgment call is: does this product have enough capability to meaningfully occupy your attention and learning bandwidth? If yes, it's a tool. If it's so lightweight you forget it exists, it's a feature.

These three criteria — distinct product, deliberate use, meaningful footprint — are the boundary. Let me walk through the common gray areas.

The Gray Areas, Resolved

Claude Desktop and Claude Code. These are two interfaces to the same underlying model, but they're different tools. Claude Desktop is a conversation-first interface for general tasks. Claude Code is a terminal-native development tool with file system access, git integration, and agentic capabilities that Claude Desktop doesn't have. You use them differently, for different tasks, with different mental models. Two slots. If you only use one, one slot.

Cursor and the underlying model. Cursor uses Claude or GPT models, but Cursor is its own tool. The autocomplete, the agent mode, the IDE integration — these are capabilities the base models don't provide on their own. If you use Cursor and also use Claude Pro for non-coding tasks, that's two tools. If you only interact with Claude through Cursor, that's one tool.

Notion with AI features. If you use Notion as a project management and note-taking tool and occasionally use its built-in AI features for summarization or generation, Notion is one tool and its AI features are... features of Notion. They don't count separately. But here's the nuance: if you're using Notion AI as your primary text generation tool — writing first drafts, generating content, using it as a substitute for a dedicated LLM — then you're using Notion AI as a tool, and it should count. The test is whether the AI capability is incidental to your use of the product or central to it.

Free tools. Free tools count. The hex is about attention, cognitive load, and workflow complexity — not about money. A free tool that you use daily occupies the same mental bandwidth as a paid tool. Google's Gemini free tier, Perplexity's free version, free-tier access to any LLM — if you're using it deliberately and regularly, it's a hex slot. I'll address the hidden costs of "free" tools in a companion article, but for the boundary question: free tools count.

Mobile apps. If a mobile app is a companion to a desktop tool you already count — like Notion's mobile app or Claude's iOS app — it's the same tool. If it's a standalone AI tool that you use only on mobile, it's a separate tool. ChatGPT on your phone that you use for quick questions while your desktop tool is Claude Pro — that's two tools.

APIs and scripts. If you're calling an API directly — writing Python scripts that call Claude's API, for instance — the API counts as the same tool as the product it belongs to. You're using Claude. The fact that you're accessing it through code rather than a chat interface doesn't make it a different tool. But if you're using a different company's API for a specific capability — say, ElevenLabs' API for voice synthesis — that's a separate tool.

Automation platforms and their integrations. n8n or Make is one tool, regardless of how many services it connects to. The automation platform is the tool. The APIs it calls are the connections, not separate tools. You wouldn't count your email service as an AI tool just because n8n sends automated emails.

Why the Boundary Matters

The temptation, once you start defining the boundary, is to define it in whatever way produces the lowest count. "Cursor uses Claude, so that's really just one tool. And n8n calls Claude's API, so that's also just Claude. And Perplexity uses various LLMs under the hood, so..." This is exactly the kind of gaming that makes the constraint meaningless.

The hex constraint works because it forces prioritization. If you define the boundary so loosely that everything collapses into three tools, you haven't constrained anything — you've just relabeled your 10-tool stack as three meta-tools. The constraint should feel slightly uncomfortable. It should force you to make choices you'd rather avoid. That discomfort is the signal that the constraint is working.

Conversely, if you define the boundary so tightly that every feature and interface counts separately, you'll hit six tools immediately and the constraint becomes a ceiling rather than a discipline. Counting Claude Desktop, Claude iOS, Claude API, and Claude Code as four separate tools defeats the purpose. They're one product accessed through different interfaces.

The honest middle ground — distinct products, deliberate use, meaningful footprint — produces a count that's rigorous without being absurd. For most people, this puts the honest count somewhere between 5 and 15, which is exactly the range where the hex constraint does its useful work of forcing cuts.

The Meta-Question

Here's something I notice when people ask "what counts as a tool?" with great intensity: the question is sometimes a way to avoid the answer they already know. If you're spending 20 minutes trying to figure out whether a browser extension counts as a hex slot, the real issue isn't the definition. The real issue is that you have more tools than six and you're looking for a loophole.

There are no loopholes. The hex is a self-imposed constraint. Nobody is checking your count. Nobody is auditing your compliance. The only person the constraint serves is you, and the only way it serves you is if you apply it honestly. If your honest count is nine, the question isn't "how can I redefine the boundary so nine becomes six?" The question is "which three tools am I going to cut, and what happens when I do?"

The boundary definition matters because it makes the audit possible. Once you know what counts, you can count. Once you've counted, you can decide. And the deciding — not the counting — is where the value lives.

The Quick Reference

Counts as a tool:
- Any AI product you use deliberately and regularly with its own interface
- Free tools you use actively
- Mobile-only AI apps you use separately from desktop tools
- Each distinct AI product, even if they share underlying models

Does not count as a tool:
- Built-in OS AI features you don't actively direct
- Mobile companion apps of tools you already count
- AI features baked into non-AI products (unless those features are central to your use)
- API access to a product you already count
- Services called by your automation platform

Apply honestly. If something feels like it should count, it probably should.

This article is part of the Hex FAQ series at CustomClanker.

Related reading: The Hex Explained, "But I Need More Than Six Tools", What About Free Tools — The Hidden Cost