Building your personal AI stack

0:00 Your personal AI stack

This is the lecture where everything in the course comes together. You have learned the landscape, how to evaluate tools, and the strengths of Gemini, Copilot, Perplexity, open weight models and creative tools. Now you will design your own AI stack, the fewest tools that cover your real work, with reusable assets and clear data rules.

0:24 Why a stack

Why design a stack at all? Because tools without a plan waste money and, worse, time. Every day you spend deciding which app to open, or rebuilding a prompt you wrote last month, is time AI was supposed to save. A stack makes the right choice automatic. And the reusable assets you build, prompts, Projects, Gems and skills, compound. Each one makes next month's work faster.

0:53 The analogy

Think of a professional chef's knife roll. Not fifty gadgets, but a handful of excellent knives, each with a clear job, kept sharp. Your AI stack should look the same. A few tools you know deeply, each with a job, and a regular routine to keep them sharp.

1:14 Typical layers

A typical stack in twenty twenty six has a primary assistant for drafting, analysis and agents, one of ChatGPT, Claude or Gemini, on a business plan for work. A research tool, Perplexity or your assistant's research mode. An embedded copilot in your suite, Gemini in Workspace or Microsoft three six five Copilot, if your organisation provides it. Gemini Notebook for source grounded study and briefing. Creative tools only as needed. And optionally, automation with Zapier, Make, n8n or Power Automate, and a local model for private tasks.

1:52 Step 1 and 2

Now the five steps. Step one, list your ten to fifteen recurring tasks with the hours each takes per week. Be specific, weekly client report, two hours, proposal first drafts, three hours. Step two, map each task to a category from the landscape map, and to the data class involved, public, internal, confidential or restricted.

2:16 Step 3

Step three, choose one tool per category you actually need, using your test set and your cost and privacy review from the last two lessons. Prefer tools your organisation already approves or bundles. If two tools do the same job, keep the better one and cancel the other.

2:37 Step 4 and 5

Step four, build assets for your top five tasks. For each, a tested prompt, a Project, Gem, GPT or skill, a checklist, and an automation where it genuinely helps. Keep the master copies in your own storage so they survive a tool change. Step five, review quarterly. Re run key tests, check costs and terms, retire unused tools, and update your assets.

3:04 The data rules card

Then write your data rules card, one page that makes safe choices automatic. Public data can go into any approved tool. Internal data into your business assistant and suite copilot. Confidential data only into your business workspace, minimised, or a local model for contract restricted material. Restricted data, passwords, card data, IDs, health data, never goes into AI tools. Voice and recordings only with consent. And before adopting any new tool, check training, retention, residency and the data processing agreement.

3:39 Simple example

A simple example. A university student's stack might be one assistant for explaining concepts and planning essays, Gemini Notebook with one notebook per module loaded with lecture slides and readings, and one research tool for finding sources. Her data rule is simple. Follow the university's AI policy, and cite only sources she has personally read. Three tools, clear jobs, and far less time lost.

4:07 Business example: freelance marketer in Karachi

A realistic example. A freelance performance marketer in Karachi, serving clients in the UK and UAE, lists twelve recurring tasks. She keeps a business plan assistant with a Project per client and a reporting skill, Perplexity for research, Gemini in the Workspace her main client provides, one image tool for concept art, and n8n to pull ad platform exports into a sheet. She cancels three overlapping subscriptions, writes her data rules card, and builds assets for her top five tasks. Her time log shows reporting and proposals getting much faster within a month. At her first quarterly review, she found she had not used the image tool once, because clients were supplying their own creative. She cancelled it and redirected the budget to a higher usage plan for her main assistant, which she used every day. The review took twenty five minutes and saved money every month after.

5:11 For teams

For a team, the same design applies, plus two things. Shared assets with named owners, team Projects, GPTs, Gems, skills and a shared prompt library. And admin controls that enforce the data rules, rather than relying on memory. Start with the business plans your organisation already has before adding anything new.

5:33 Common mistakes

Four common mistakes. Starting from exciting tools rather than your actual tasks. Keeping every tool you have ever tried. Building assets that exist only inside one product. And never setting a review date, so the stack slowly drifts back into chaos.

5:51 Stacks by role

What does a stack look like for different roles? A content creator might pair an assistant with one image tool, one voice tool and a research tool, with a brand voice skill and a disclosure checklist. A sales professional, an assistant, an embedded copilot and a research tool, with a pre call research prompt and deal Projects. An agency account manager adds automation for reporting and lead triage. An operations manager leans on the suite copilot and a notebook of procedures. And a developer adds a coding agent and a local model.

6:31 The 30-minute quarterly review

And here is the thirty minute quarterly review. Usage, which tools did you actually use, and cancel the rest. Cost, any price or plan changes. Quality, re run three tasks from your test set. Terms, any changes to data terms or features. And assets, update the prompts and skills that drifted, and add one new asset. Thirty minutes, four times a year, keeps your stack lean and sharp.

7:01 Watch me do it, part 1

Let me design the Karachi freelance marketer's stack. I list her twelve recurring tasks with hours per week. Client reporting, four hours. Proposals, three. Ad copy variations, two. Research on new clients, two, and so on. For each I assign a category, assistant, research, embedded copilot, creative or automation, and a data class. Client reporting involves confidential performance data. Research is mostly public. The five tasks with the most hours are highlighted. That is where the assets will go first. Now one tool per category. Her business assistant plan, Perplexity for research, Gemini in her main client's Workspace, one image tool, and n8n. Three overlapping subscriptions drop out. Then assets for the top five tasks. A reporting skill with her metric definitions. A proposal Project with past winning proposals. A tested ad copy prompt. A research Space with source rules. And an n8n flow that pulls ad exports into a sheet. She writes her data rules card, and books a thirty minute review for the first Monday of next quarter.

8:15 Recap and try this now

Recap. Start from your recurring tasks, map them to categories and data classes, choose one tool per needed category, build tested assets for your top five tasks, and review quarterly, with a data rules card that keeps you safe. Try this now. Design your personal AI stack with the five steps, write your data rules card, and put your first quarterly review in your calendar before you close this lesson.