Hire Digital Staff, Not a Search Engine

Why you are probably using AI wrong

Let's be completely blunt: you are probably using AI wrong.

You open Gemini. Blank screen. You type a question. You get an answer. You close the tab. Tomorrow morning, you log back in and start entirely from zero. You paste the same paragraph about your business into every single chat window. You upload the same price list every single week. You wonder why it forgot what you said on Tuesday afternoon.

You would not hire a human assistant, sit them at an empty desk with no files, no phone, no job description, and no idea what your business does — and then wonder why they are useless. Yet that is exactly what most people do with AI. They treat a multi-billion-pound neural engine like a temp worker with amnesia.

For the previous 18 months, I have been working on the ground, automating systems and workflows for small business clients using Google's tools. Back in November 2025, I founded Marra.AI to explore what human-first AI actually looks like. Then, at the beginning of May 2026, I took a massive personal leap — I stepped away from my corporate career to chase my dreams and run Marra.AI full-time.

Right as I made that move, Google dropped a massive structural overhaul of their entire AI platform. I have spent the last several weeks buried in deep, obsessive research, testing and figuring out exactly how this new ecosystem works. I am writing this article because I don't want to hide anything. I want to share my research openly, show you how it is actually built, and give you a straightforward blueprint on how to use it properly to run your business.

Now, I know what some of you are thinking. Writing a massive business guide about Google feels a bit weird because the vast majority of small businesses run on Microsoft. Not many people are naturally set up to utilise Google for business. But here is the honest reality of why that disconnect matters, and why Google represents a completely different ball game.

To make AI actually useful, it all comes down to context.

Think of a standard Large Language Model as a word calculator. If you open a blank chat and ask it to write a business letter, it will do it, but the answer will be completely generic. It is like a whole number: 1. But context acts like the decimal places. If you give the AI the exact context of your business, your customer, and your setup, the answer becomes incredibly precise and tailored: 1.25. The more decimal places of context you provide, the more accurate and valuable the output becomes.

With other AI tools like ChatGPT, Claude, or Grok, getting those decimal places is exhausting. You have to act like a makeshift IT technician, constantly copying, pasting, and piping external data into a chat bubble to try and bring your context in.

Google is completely different. They aren't an AI company trying to get you to integrate your files into their tool. They are a platform company that already owns the infrastructure where your business information lives. They have natively piped Gemini directly under the roof where your Drive, Gmail, Calendar, Tasks, Keep, and Docs already sit. The context doesn't need to be imported; it is already in the building.

Once you see the platform this way, it stops being a confusing search engine and starts acting like a real workspace. To show you how this holds together, I am going to use a real-world example throughout this guide: a woman called Linda who runs the digital side of a local community choir called the Carnegie Singers. Every time I say "choir," just put your own business name in its place. Plumber, café, electrician, dog groomer — the shape is exactly the same.

Meet Linda

Linda is on the committee of the Carnegie Singers. She is the one who does the heavy digital lifting: Facebook posts, posters, tickets, emails to local businesses, tracking meeting actions, chasing people, and trying to remember what was agreed on Tuesday.

She is not a tech person. A few years ago she barely used a computer. Now she has Google AI Pro, a laptop, and a Google Pixel phone. She types when she is at her desk, and talks to her phone using voice dictation when she is out and about.

She is brilliant at it, but she is exhausted. She is doing the job of five people, and a massive chunk of her information doesn't live in Google at all. Committee decisions happen in a fast-moving WhatsApp group. Some choir email goes to a separate account. Facebook is where posts go, but Facebook doesn't talk to her files.

If any of that sounds like your typical week, keep reading.

The blueprint: Moving from unwritten chaos to a digital team

To understand how we fix Linda's exhaustion, we have to look at the basic economics of how any business functions. A business takes an input (messy data, raw requests, unorganised files), the people perform value-added tasks (operations) on that input, and they produce an output that is worth more than what they started with. The business simply skims its profit margin right out of the middle.

To run that machine, every business relies on a simple, three-part equation:

When you get a generic, terrible output from an AI, you already intuitively know that context is missing. You don't need a lecture on that; you can see the results are poor. But while most people recognise this at a high level, they struggle at the low level — the practical setup. They try to cram their entire business equation into a single, temporary chat window.

The biggest bottleneck for most small businesses is that their processes are completely undocumented. The magic of how things get done lives entirely inside the owner's head. If you haven't written down your human Standard Operating Procedures (SOPs), you have no foundation for AI. You cannot tell an AI worker how to behave if you haven't defined the job.

Before I moved into AI, my background was as a Business Improvement Specialist in the nuclear industry at Sellafield. I was handed a massive problem: take a £700,000,000 facility (the BEPPS DIF project) from construction and commissioning and safely transition it into live, active operations processing spent nuclear fuel. I started with nothing.

I began by identifying every task that needed to happen to reach the end state — a fully live, operational facility. I built a master to-do list that defined what "done" looked like, then documented a delivery profile for every single task: who owned it, how it would be completed, and how we would know it had landed. I sat down with every department lead, built a digital framework containing over 200,000 data points, and created the processes that held team leaders accountable — delegating work, tracking actions, and integrating the programme with the wider site through a standard calendar exercise so nothing sat in a silo.

I wrote the meeting charter in detail — quorum, update rules, how actions flowed — and defined every metric on the wall. Each measure came directly from the processes we had built, so when you walked in, you could see whether delivery was on track. That was the War Room method: a dedicated space where we reviewed progress every week, with the core data, templates, master schedules, and step-by-step instructions pinned physically to the walls. When a team walked in, everyone shared the same context, knew their role, and followed the same process.

That wasn't decoration. It was the system. The room was how we turned undocumented knowledge into something a team could actually execute against — and how we managed one of the most complex operational transfers on the site. Internal regulators celebrated it as a site-wide best practice.

When Google rebuilt their AI platform in May 2026, I recognised the same shape immediately. Google didn't copy my War Rooms — but when you've spent years making operations work, you recognise the same pattern when you see it. They hadn't just added a chatbot to Drive — they'd built persistent project spaces, linked source files, specialist agents, live connections to Gmail and Calendar, and a proper workbench for drafts. Different names. Different buttons. But the same underlying idea: give the work a room, pin the truth to the walls, and bring in the right people to execute.

The table below shows how those pieces line up — what every operation needs, what we call it in the War Room, what Google built, and what it looks like for Linda.

Let's look at exactly how these layers interact, why they matter, and how value flows through them step by step.

Layer 1: The filing cabinet (Google Drive)

What it is

Google Drive isn't just a place to dump files in the cloud. In this architecture, it is your source of truth — the heavy metal filing cabinet sitting in the corner of your War Room holding your pristine master copies.

How it works

To give the AI those missing "decimal places" of low-level context, we organise Drive on purpose. We create dedicated folders for different types of files:

• Core Identity (Soul Docs): These are the foundational documents that define the heart and soul of your business. It includes your history, the founder's vision, your core values, and your definitive brand voice guidelines. (You must define this "Soul" first so the room knows what business it's working for!)
• Official Records: Formal meeting minutes, signed agreements, and legal records.
• Live Operations: Dynamic files that change daily, like master task lists or active job schedules.
• Templates & Assets: Reusable structures (like standard email layouts or post formats) and brand graphics like logos.
• Generated Drafts: A dedicated holding area where AI output lands for human quality control before anything is ever published.

The rule that matters

Link the live folder from Drive into your workspace; never upload isolated file copies. When you link the live Drive folder to your Notebook, you are pulling the folders out of the cabinet and pinning the relevant sheets to the wall. If you change the master sheet in the cabinet, the print-out on the wall updates itself. If you manually upload a copy of a spreadsheet into a chat window every Friday, you create version-controlled garbage, and the AI will eventually cite outdated information.

The Catchment Doc

Some of your business data lives outside of Google, like Linda's committee decisions on WhatsApp. To bridge this gap without drowning the AI in chat logs, we build a Catchment Doc — a single Google Doc in Drive called Live Updates. When a decision happens on WhatsApp, Linda spends 30 seconds opening her phone and dropping one clear line at the top: [15 June] Committee confirmed: summer concert tickets on sale Monday, £10. One live document for the AI to read first.

Layer 2: The project room (Gemini Notebook)

What it is

A Gemini Notebook is a permanent, persistent project room. It is the digital equivalent of the nuclear project War Rooms I built at Sellafield. It is not a temporary chat bubble that disappears when you close the browser.

How it works

You set up one notebook per major domain or project area in your business (e.g., Carnegie Singers Operations). When you walk into this room, your linked Drive folders — containing your Soul Docs, live operations, templates, and your Live Updates catchment file — are permanently printed out and pinned to the corkboards on the wall. Anyone — human or AI — who steps into that room can look up at the wall and instantly share the exact same context.

Inside the notebook instructions, you establish the Notebook Instructions (The Project Charter). These are the high-level rules written directly on the door of the room:

Because this room is entirely persistent, it retains a massive memory across weeks of active work. It completely remembers previous decisions, meaning you never have to re-explain your business identity to a blank screen ever again.

Layer 3: Your digital staff (Gems)

What they are

Now that the project room is built and the data is pinned to the walls, you bring in the people to execute the processes. These are Gems — custom-configured digital staff members.

A Gem does not hold your business files in its prompt window; it simply holds a behaviour blueprint. This is an Agentic Operating Procedure (AOP), which is just a clear job description and execution checklist derived directly from your human processes.

How it works

Linda "hires" two specialised digital workers, named after real-world inspiration to keep them memorable. When you invoke them, it is the digital equivalent of tapping a staff member on the shoulder and asking them to step into the War Room:

• LINDA (Logistics and Administration Assistant): Named after my own mam, this agent is built with a highly specialised prompt that gives her deep emotional intelligence, a warm tone, and a high EQ. Her job description is strictly administrative: she takes messy, dictated notes, structures clean action tables, and drafts formal community outreach emails.
• MARG (Creative Marketing Assistant): Named after my aunt Margaret, her job description is entirely focused on customer-facing promotional layouts. Her AOP enforces a strict quality checklist: "Every social media post must explicitly state the Date, Time, Venue, Ticket Price, and a formatted booking link. Use brackets as placeholders if any detail is missing."

How they interact

The true magic happens when a specialist steps into the project room. Linda walks into her permanent Carnegie Singers Operations notebook, opens a new thread, and calls her marketing expert into the space by typing @MARG.

Because MARG is invoked inside that specific notebook, she instantly inherits the entire environment. She looks up at the walls and immediately sees the Soul Docs and the Live Updates file pinned there. Linda issues a simple command: "Draft the Facebook post for the summer concert launch." MARG provides the structured, professional marketing behaviour, while the notebook room provides the verified facts.

Layers 4 & 5: The telephone lines and the workbench

What they are

To keep your digital staff from working in a total vacuum, you need live operational lines to the outside world, and a clear desktop to review their output.

How it works

The Telephone Lines (Connected Apps): By turning on native Workspace extensions, Gemini gives you landline desk phones sitting right in the room. Without leaving the workspace, you can securely access your live Gmail, Google Calendar, and Google Tasks. When Linda is out and about, she can talk directly to her phone: "@Gmail — summarise emails about the choir venue from this morning, check my @Calendar for conflicts on Monday, and log a Google Task to publish the poster text." The system coordinates across her live apps in real time without her having to step outside the room.

The Workbench (Google Docs & Gemini Canvas): When MARG generates a draft, it doesn't get trapped in a messy chat log. The workflow advances straight to a big shared table in the middle of the room. Using Gemini Canvas, Linda reviews the draft side-by-side with her chat screen. She uses visual sliders to tweak the length and tone, highlights sentences for real-time rewrites, and natively exports the finished text straight into her Generated Drafts folder in Drive as a crisp Google Doc. The AI handles the labour, but Linda maintains absolute human control at the workbench before she copies it over to Facebook.

Layer 6: The research desk (NotebookLM)

While Linda runs her daily operations, chats, tasks, and emails out of her Gemini Notebook War Room, she uses NotebookLM as a quiet research desk in the back of the room. It uses the exact same sources but is built for a completely different job: deep synthesis. You go to the research desk when you need to study complex source material or generate structured briefing artefacts, like an Audio Overview for a new volunteer or a slide deck for a big planning meeting.

Layer 7: The assembly line (Workspace Studio)

Workspace Studio is Google's native, no-code background automation tool. It acts like an automated assembly line running quietly in the corridor outside the room. It executes background operations automatically when a trigger happens (e.g., a file lands in a folder, an email arrives, or a form is submitted).

This is strictly gated for paid Google Workspace Business and Enterprise tiers. Personal AI Pro accounts run these workflows manually within the notebook workspace. You are not failing if you do not have this yet; the manual system on a personal account is the exact intended starting path for a one-person operation.

The three levels of instructions (do not mix them up)

To keep this entire digital department from experiencing a logical breakdown, you must keep your instruction layers strictly separate. Never copy and paste the same general rules across all three levels:

• Notebook Instructions (The Project Charter): Defines the world. What business is this? Which files win? Who is the user? (This maps directly to your Data Pillar and what is pinned to the War Room walls.)
• Gem Instructions (The Job Description): Defines the worker's behaviour. How does this specific specialist execute their role? What constraints must they follow? (This maps directly to your Process Pillar and your worker's checklist.)
• Studio Step Prompt (The Assembly Line): Defines a single automated background step on the machine outside.

If your notebook, your Gems, and your files all simultaneously say "be polite and check the price list," the hierarchy fractures. The AI won't know which rule takes precedence, causing conflicting outputs or hallucinations.

How a week actually flows (Linda's Tuesday)

Theory is completely useless without an operational picture. Here is how value smoothly moves through Linda's digital department on a typical Tuesday evening:

• The Input (1 minute): A committee decision happens on WhatsApp: summer concert tickets go on sale Monday for £10. Using her phone, Linda opens her live Live Updates catchment doc and types that single line at the top.
• The Project Room (2 minutes): She opens her Carnegie Singers Operations notebook. She asks LINDA (Administrative Assistant) to update the master task list and schedule a tracking item.
• The Specialist (3 minutes): She opens a new chat thread in the room and calls in her marketing expert: @MARG — draft the ticket launch post, checking the Live Updates doc for the confirmed price. MARG pulls the verified £10 fact from the room's walls and wraps it in a perfect promotional layout.
• The Workbench (4 minutes): Linda pushes the text to Canvas, verifies the details, exports it to her drafts folder, and schedules a reminder task on her phone.

Ten minutes of total work. No re-explaining her brand voice, no searching through forgotten chat logs, and no made-up ticket prices.

Building your digital department

Stop logging into AI as if it were a sterile search box that completely forgets who you are the second you close your browser tab.

The era of duct-taped, standalone chatbots is officially over. By designing a persistent architecture built the way Google natively engineered the platform to be used, you step out of the role of an administrative technician and into the role of an organisational architect. Give your platform a room, a structured filing cabinet, clear behavioural procedures, and a defined division of labour.

Build it right, and your small business won't just have an AI tool — it will possess a highly scalable, self-sustaining digital department.

Marra.AI helps non-technical people and small businesses design digital departments on Google's stack — built the way Google intended, explained in plain English. marra.ai

— Kris Wilkinson, Founder, Marra.AI

Page views: —