Mapping the AI tool landscape in 2026

0:00 The AI tool landscape

Every week there is a new AI tool that promises to change everything. If you chase them all, you will spend more time signing up than working. In this lecture you will learn a simple map of the AI landscape that has stayed stable for years, the three big shifts of twenty twenty five and twenty twenty six, and how to draw your own map in half an hour.

0:30 Why a map matters

Why bother drawing a map at all? Because without one, two things happen. Money leaks into overlapping subscriptions, three assistants that do the same job, and important jobs go undone because nobody noticed a whole category was missing, like a proper research tool or clear rules for voice cloning. Think of it like a kitchen. You do not need every gadget, but you do need a knife, a pan and an oven. The map shows you which essentials you have, which you lack, and which drawer is stuffed with duplicates.

1:09 Think in categories

The trick is to think in categories that map to jobs, not in brand names. Leaderboards reshuffle every month. The categories have stayed remarkably stable. So first ask which category of job you are doing, then pick the best current tool within that category for your needs.

1:29 Assistants and copilots

General assistants like ChatGPT, Claude, Gemini and consumer Copilot are chat first tools for drafting, analysis and everyday reasoning, now with files, research, memory and agents. Embedded work copilots, like Gemini in Google Workspace and Microsoft three six five Copilot, live inside the suite where your work already is. Their superpower is context, your emails, files and meetings, within the permissions you already have.

1:57 Research and notebooks

Answer and research engines, like Perplexity, Google's AI Mode and the deep research modes in the assistants, are search first and cite their sources. Source grounded notebooks, like Gemini Notebook, which was called NotebookLM until July twenty twenty six, answer only from the documents you add, which is perfect for studying, briefing packs and audio overviews.

2:21 Agents and coding agents

Agents carry out multi step tasks, browsing, using connected apps and creating files. Examples include ChatGPT Work, Claude's agentic tasks, Google's Gemini Spark, Microsoft's Copilot Cowork and Perplexity's Comet browser. Coding agents work inside codebases, such as Claude Code, OpenAI's Codex, GitHub Copilot and Google's Antigravity.

2:41 Open-weight, media, automation

Open weight models, like Llama, Mistral, Qwen, DeepSeek, Gemma and OpenAI's g p t oss, can be downloaded and run where you choose, for control and privacy. Image, video and voice tools create media. And automation platforms, like Zapier, Make, n8n and Power Automate, connect apps with AI steps for repeatable processes.

3:04 Three big shifts

Three shifts define the last two years. First, from chat to agents. Every major vendor now ships an agent that plans and acts, with confirmations for consequential steps. Second, open standards for connections. The Model Context Protocol, introduced by Anthropic in twenty twenty four, is now supported by OpenAI, Google, Microsoft and many tool vendors. Third, multi model products. Microsoft three six five Copilot offers models from both OpenAI and Anthropic, and Perplexity lets you choose among frontier models. Picking a product no longer locks you into one model family.

3:43 Bundled vs specialist

Suites bundle good enough AI into tools you already pay for. Specialists go deeper, Perplexity for research, ElevenLabs for voice, Midjourney for visual style. A sensible rule is to start with what is bundled and approved, then add a specialist only when a frequent, specific job clearly benefits.

4:04 Simple example

Here is a simple example. A freelance copywriter maps her tools. One general assistant for drafting and editing. One research engine for sourcing facts with citations. Gemini Notebook for each big client, loaded with their brand guides and past campaigns. And nothing else, yet. No agents, no automation, no image tools, because her jobs do not need them today. Her map is small, clear and cheap, and she knows exactly which tool to open for each task.

4:37 Worked example: a Karachi agency

A six person social agency in Karachi mapped every AI tool in use. They found three overlapping assistant subscriptions, no approved research tool, client data going into personal accounts, and no rule about voice cloning. The result was one business assistant plan, Perplexity for the research lead, Gemini in their existing Workspace, ElevenLabs with consent only voice rules, and n8n for reporting, each with a data rules card. Three months later they redrew the map. One new category had appeared in their work, short video for clients, so they tested two video tools before adding one. And one tool had quietly gone unused, so they cancelled it. The map is not a one off exercise. It is a living picture that makes each new decision faster.

5:32 Try this now

Try this now. Open a blank table with the categories from this lecture as rows. For each one, write the tool you use, what you use it for, whether it is a personal or business plan, what data it is allowed to see, and who owns the decision. Then circle one gap, a category you need but do not have, and one overlap, two paid tools doing the same job. Those two circles are your first two decisions.

6:06 Common mistakes

Let's name the common mistakes. Chasing leaderboard headlines instead of testing on your own tasks. Paying for three assistants that do the same job. Letting a free trial tool touch client data before anyone checked its terms. And ignoring the AI already bundled in the suite you pay for. If your map helps you avoid those four, it has already paid for the half hour it took.

6:35 Watch me do it, part 1

Let me fill in the map for the Karachi agency. I open a sheet with the nine categories as rows and five columns, tool, what for, plan, data allowed and owner. General assistants, three tools, two on personal plans. Embedded copilot, Gemini in the Workspace they already pay for. Research, nothing, people just use whatever assistant is open. Notebook, nothing. Image, voice and video, ElevenLabs for voice overs. Automation, n8n. In the data allowed column, one personal assistant plan has client data, and that cell turns red. Owners go in last, and three rows have nobody.

7:17 Watch me do it, part 2

Now the annotations. The three assistant rows are an overlap, so I bracket them and write, consolidate to one business plan. The research row is a gap, so I write, add Perplexity for the research lead. Workspace stays, because it is already paid for and approved. For ElevenLabs I add a rule, stock voices only, cloning only with written consent. Then a one line data rule beside each tool, like client data only in the business assistant plan. The map took thirty minutes, and it produced four concrete decisions.

7:56 Recap and next step

Recap. Think in categories that map to jobs. Remember the shifts to agents, open connection standards and multi model products. Start with bundled, approved tools and add specialists deliberately. Your next step is the hands on in the lesson. Build your personal AI map, with the tool you use in each category, what for, which plan, and what data it is allowed to see.