The AI Tool Stack: A Practical Cheat Sheet

I use six or seven AI tools regularly, and for a long time I didn't have a real system for which one to open first. So I built one. This is that system: what each tool is actually good at, when to reach for it, and where they overlap so you're not paying twice for the same job.

This isn't a "best AI tools of 2026" listicle. It's a working document from someone who uses all of these daily, updated as the tools change.

Start here: "I want to… → use this"

Don't read this post top to bottom. Find the row that matches what you're trying to do, then jump to that tool's section.

The rule of thumb: go where the data already lives.

If the job touches your email and calendar, use the tool built into that ecosystem. If it touches your notes or task manager, use the tool connected to those. If you're not sure, that's usually a sign you're using the wrong tool for the job, not that you need a "default."

A quick bias check

I had Claude help me build this cheat sheet, and I specifically asked it to audit its own first draft for favoring itself. It found a few things worth calling out — "default to Claude" language, examples where a different tool was clearly the better fit — and rewrote them. I'd recommend doing the same with whatever AI helps you build your own version of this: ask it directly where it's steering you toward itself.

Where tools overlap (and who wins)

Claude

Lane: Writing, building things (dashboards, tools, code), and anything tied to your personal knowledge base or task manager.

Don't use for: Live social sentiment, high-volume cited research, or documents so long they blow past its context window — other tools do those better.

Chat + model picker

Pick a faster/cheaper model for quick tasks and a deeper model for hard reasoning or strategy work. Start a new conversation per topic — it keeps things faster and cheaper to run.

Projects

Persistent workspaces with their own instructions, files, and memory. One project per area of your life or work — a job search, a side business, a specific client — so you're not re-explaining context every time.

Memory

It remembers facts and preferences across conversations. Say "remember that…" to save something, "forget…" to remove it. Correct it when it's wrong — treat it like training a new assistant.

Artifacts (hosted pages and small apps)

Ask it to build a live web page — a calculator, tracker, or dashboard — and you get a shareable link you can keep updating in the same conversation.

Docs / Slides / Design

It can produce a real editable document, slide deck, or visual design you can tweak and export. Great when the argument matters more than the visual polish — for polish, a dedicated design tool usually wins.

Skills

Saved playbooks it follows automatically when a task matches. If you catch yourself explaining the same process twice, that's your cue to turn it into a skill.

Connectors

Live links to your other apps (notes, email, task manager, calendar) so it can read from and write to them directly. Just mention the app by name in your request.

Research mode / web search

Good when research is a means to a finished deliverable and perfect freshness isn't critical. For fast-moving facts you need to defend, go to a dedicated research tool first and bring the sources back.

A desktop agent (Cowork-style features)

Some AI apps can work as an agent on your computer — reading and writing files in folders you connect, using your other apps, driving a browser. Give it a whole job, not a single question, and it'll show its work as it goes.

Phone-to-desktop task handoff

Send a task from your phone that your computer executes using local files and connections. Useful for anything you'd otherwise have to sit down and do yourself. (Never queue these up while driving — do it before you start the car or during a stop.)

Scheduled tasks

Runs a prompt on a schedule with nobody watching. Keep each one self-contained and focused on a single outcome rather than trying to cram multiple jobs into one task.

Browser automation

Useful for multi-step clicking where the result needs to land somewhere specific — a form, a comparison table, a status check. Always watch before it submits anything that can't be undone.

A coding agent

Point it at a code repository, describe the feature, let it build and test. For small background fixes you can queue and walk away from, a lighter-weight coding agent (many platforms now offer one) is often the better fit.

Voice mode

Good for brain dumps you want turned into notes and structured tasks afterward. For a real back-and-forth conversation, a dedicated voice AI usually feels more natural.

Gemini / Google AI

Lane: Very long documents, anything already in your Google account, always-on automation, voice, media generation, and audio learning. Usually the most generous usage limits of the paid tools in a typical stack — and often the most underused.

Don't use for: Final client-facing copy, or anything that needs to be defended with sources.

Main chat (large context window)

Can hold an enormous amount of text in one conversation — think full contracts, long research PDFs, or entire reports at once. Upload the largest files you have and ask questions across all of them together.

Custom assistants (Gems)

Saved assistants with their own instructions and files, similar to Projects in other tools. Good for a recurring job that lives entirely inside your existing document/email ecosystem.

Deep Research

A multi-step research agent that can also pull from your own files and email. Best when you want research combined with your personal context — for research you need to cite publicly, use a dedicated citation-first tool instead.

Live voice + camera

Real-time voice conversation, and it can see what you point your camera at. Great for commute time — quizzing yourself, practicing a conversation, or getting a quick opinion on something in front of you.

Audio/video learning tool (NotebookLM-style)

Ground a notebook in only the sources you add — PDFs, links, videos — then generate an audio "podcast" of two hosts discussing them, plus quizzes and flashcards. This is genuinely one of the best tools for turning reading into commute or workout time.

AI built into your document/email suite

A side panel inside your email, docs, and spreadsheets for quick summarizing, drafting, and analysis without leaving the file you're working in.

Browser-native AI + AI-powered search

Use the search mode for complex questions instead of ten separate searches. Use the in-browser AI for quick questions about the page you're currently on — no extra tool required.

Media generation

Image, video, and music generation tools, generally the strongest quality in a typical consumer AI stack. Best for concepts and mood boards rather than finished, client-ready output.

Background coding tools

Some platforms now offer a coding agent that works on your repository in the background and opens a pull request when it's done — useful for small fixes you don't want to sit and watch.

Always-on agents

The newest category: an agent that runs in the cloud (not on your device), so it keeps working even when your phone and computer are off. It typically works through three pieces — tasks (the goal), reusable instructions, and schedules (time- or condition-based triggers) — and connects natively to your email, calendar, and documents. This is the right tool for any recurring job that only touches those apps. For jobs tied to a separate notes app or task manager, keep those with your main AI assistant instead, since always-on agents usually don't connect to third-party productivity apps yet.

Perplexity

Lane: Facts you can cite. Competitive research, market numbers, anything that needs a source attached. Think of it as the "prove it" step before you build anything.

Don't use for: Building the final deliverable — take the sources somewhere else for that.

Search with citations

Your default search engine for anything with a factual answer. Every claim comes with a linked source.

Deep Research

Multi-step research across hundreds of sources, delivered as a report. Use it for anything headed into a deck or a decision — export the findings, then build the final piece somewhere else.

Source verification

Checks each claim against its underlying source and tells you how well the evidence actually supports it. Run this before any number goes in front of a client or a boss.

Research hubs (Projects/Spaces)

Long-running collections with files and memory scoped to a single ongoing topic. One hub per research thread you're actively tracking.

A research-first browser

Some research tools now ship their own browser with built-in summarizing and multi-step task completion. Worth trying if you do a lot of deep-dive reading sessions, but it overlaps with browser-native AI for everyday questions.

Finance research

Dedicated tools for earnings calls, filings, and market data — useful during earnings season or before making an investment decision.

Model picker

Some research tools let you switch between different AI models mid-conversation. Handy for getting a second opinion without opening a separate app.

Grok (or the AI built into your social platform)

Lane: Real-time pulse on what's happening right now, culturally or socially.

Don't use for: Anything that has to be accurate or defensible — real-time social AI tools generally have the weakest citation discipline of the bunch. Also: never paste confidential work material into a personal-plan AI tool, since some plans use your conversations for training.

Chat with live platform access

A first-pass read on "what's the vibe right now" on a given topic.

Multi-step live search

Produces a quick summary of trend velocity across the web and social platforms. Verify anything important with a citation-first tool before you act on it.

Reasoning modes

Some of these tools now show their "thinking" process. Fine for brainstorming, not for anything that needs to be right.

Scheduled tasks / autopilot

Can run recurring jobs, but keep the scope narrow — the social-and-culture pulse is what these tools are actually good at, not general daily-briefing work.

Image/video generation

Fast, rough generation — good for throwaway concepts and memes, not for anything polished or client-facing.

A general-purpose backup chat AI

Worth having as a second opinion and a sanity check, even if it doesn't own a dedicated job in your stack. Paste in another tool's answer and ask it to "poke holes in this" before you send something important. Most of the advanced features (deep research, browser agents, scheduled tasks) genuinely overlap with what a dedicated stack already covers — so unless there's one specific feature you can't get anywhere else, the free tier is usually enough.

Your company's AI tool (Microsoft 365 Copilot, Google Workspace AI, etc.)

Lane: Anything confidential to your employer. It runs inside your company's security boundary — your personal AI tools don't, and that boundary matters more than convenience.

• Chat across your work inbox and calendar — great for a Monday catch-up across scattered threads without leaving your inbox.
• Meeting recap — automatic summaries and action items from recorded calls.
• Document/deck builder pulling from your own work files — useful for internal materials that need to reflect real company data, like a handoff document before you leave a role.

Design tools (Canva or similar)

Lane: Fast, good-looking visuals — social posts, carousels, one-pagers, decks where design matters more than argument.

Most design AI platforms split their AI features across free and paid tiers pretty aggressively. On a free plan, you typically get basic AI writing and simple image editing; the more advanced layout and batch tools usually require a paid tier. Check what your specific plan includes before assuming a feature is available.

Automation tools (Zapier or similar)

Lane: The pipes between your other apps — when something happens in one tool, something else happens automatically, without you or an AI needing to be in the loop.

This is the backbone of any freelance or consulting automation work. A typical flow: someone books a call → a lead gets created in your system → they get a confirmation → you get a prep task. Build and test on a free tier, then upgrade once you're doing this for a paying client.

Most platforms in this category also now offer:

• AI-assisted automation building — describe the workflow in plain English and have it draft the automation for you.
• Simple databases and forms — enough to build a lightweight client intake system with no code.
• A way for your main AI assistant to trigger these automations directly, which is worth setting up once you have a few flows built.

The tools behind the tools (if you're building anything)

If part of your AI workflow involves actually building software or automations for clients, a few non-AI-specific tools round out the stack:

• A backend-in-a-box (Supabase, Firebase) for database, login, and file storage — most now include an AI assistant for writing queries.
• A communications API (Twilio) for SMS/voice automations — a classic starter offer for small-business automation work.
• Booking and scheduling links with routing forms to screen leads before they hit your calendar.

The ignore list

Things that sound useful but usually aren't worth paying for in a typical stack:

• A second general-purpose chat AI's paid tier, when it overlaps heavily with your main assistant.
• The most expensive tier of a research tool's business/enterprise memory features, unless you're doing that volume of work.
• Any "companion" or persona features — novelty, not utility.
• A second AI-native browser, once you've already got browser-native AI plus one agent-capable browser tool.

How I'd actually start

1. Pick one tool as your default for writing and building. Don't split that job across two tools.
2. Pick one tool for cited research, and use it before you build anything that needs to be factually solid.
3. Route anything tied to your email/calendar to whatever AI is native to that ecosystem — don't rebuild it in your main assistant.
4. Everything else is a "nice to have" until you hit a specific wall your main tools can't solve.

The overwhelm mostly comes from treating every tool as equally important. Most of them aren't. Pick your two or three anchors, and treat the rest as specialists you call in when the job specifically needs them.

Have a stack of your own? I'd love to hear how you're routing between tools — reach out through the contact page.