Workshop 01 · Tasks, Jobs, Processes & Workflows · Launch AI Training

Launch · AI Training · Powered by Gemini

Training Day,
in motion.

A working session, not a deck. Three hands-on tasks designed to take you from understanding AI to actually building with it in Gemini, using your own work, your own clients, your own team.

Format

Hands-on

Platform

Gemini

Tasks

Facilitator

Jeremy

Task 01 · 30 mins

Tasks, Jobs,
Processes & Workflows.

Duration30 min

FormatHands-on

OutputTop 3 tasks

Module01

§ 01 · The Primer

Know the difference.

Before you start dumping tasks, get clear on the four levels. Most current AI use sits at task level. The opportunity, and your USP, lives further down the chain.

Level 01

Task

A single, discrete action. Summarising a transcript, drafting a client email, writing ad copy. AI handles this today

Level 02

Job

A collection of related tasks that form a role. Analyst, planner, account manager. AI assists parts, not the whole. Augments, doesn't replace

Level 03

Process

A repeatable sequence of steps with a defined outcome. Reporting workflow, data pull to client-ready deck. Now automating via agents

Level 04

Workflow

How processes connect across people, teams & systems. Planning → activation → reporting. The frontier for AI

Task ▸ Job ▸ Process ▸ Workflow

§ 02 · The Exercise

Four steps.
Thirty minutes.

Work through the four stages below in order. Don't skip the reality check. That's where the bad ideas die.

Step	What you're doing	Time
01	Task dump: list everything you do	7 mins
02	Identify your top three opportunities	7 mins
03	Score each task on Scale / Process / Data	10 mins
04	Reality check: risk, tools, approval	6 mins

§ 03 · Step One

Task Dump

7 mins

Don't curate. Don't edit. Write down everything you do, daily, weekly, monthly. The boring stuff matters most.

Daily

Weekly

Monthly

§ 04 · Step Two

Prioritisation

7 mins

Look at your list. Ask yourself:

What makes these tasks stand out?
Which would free you to focus on better or different work?
What jobs do you not enjoy?
Who else benefits if this is automated?

§ 05 · Step Three

Score on scale, process, data.

Rate every shortlisted task on the criteria below. The highest combined scores are your best AI candidates.

Criterion 01

Scale

How frequently is this task done?/10
How much time does it take?/10
If time were no limit, how much further could you scale it, and how much better would it be?/10

Criterion 02

Process

How easy is it to map the process?/10
How much is pattern-following vs. original thinking?/10
Does the task have a predictable structure?/10
Could you teach it to a new joiner in a day?/10

Criterion 03

Data

Do you have good examples of the output to reference?/10
How much would adding 1st or 3rd-party data improve it?/10
Is there a data source that would add value?/10

§ 06 · Step Four

Finalise & Reality-Check

6 mins

Pick your winners and pressure-test them.

Which tasks scored the highest?
Why would automating this be greatly beneficial for your work?
What do you need to do to explore this further?

Task 01 · Worksheet

Capture your Task 01 answers in the live sheet.

Open the worksheet on your laptop or phone before we move on. We'll share back as a group before Task 02.1.

Open Task 01 →

docs.google.com/spreadsheets/…

Task 02.1 · 15 mins

Prompting
that works.

Duration15 min

FormatBuild & run

Output1 live prompt

FrameworkRICCE

§ 01 · The Diagnosis

Why most prompts fail.

Prompts are the testing ground for a consistent approach and reliable outputs. Get them wrong and Gemini guesses. Get them right and Gemini delivers, every time.

Too vague

Not giving enough detail leaves Gemini guessing at your intent.

No context

Gemini has no idea about your audience, industry, or purpose.

One-shot thinking

Expecting perfect results on the first attempt without iteration.

No framework

Without guidelines, every prompt is a coin flip.

§ 02 · The Framework

Introducing RICCE.

Five letters. Five levers. Pull each one deliberately and your output quality compounds. Skip one and you're guessing again.

Role

Tell Gemini who to be. Shapes tone, expertise level, perspective and approach.

Input

Raw material: data, documents, examples, briefs. Label clearly.

Context

Audience, purpose, environment. Who reads it? Where does it appear?

Constraints

Rules, boundaries, requirements. Your biggest lever for output quality.

Examples

Show what good looks like. Two or three examples beat pages of instructions.

§ 03 · The Deep-Dive

Each lever, decoded.

What each element actually means, and how to write it well. Read these before you build.

Role

Tell Gemini who to be. This shapes tone, expertise level, perspective, and approach.

Without a role, Gemini defaults to generic helpful assistant. With a role, it draws on specific patterns.

How to write it

Be specific about expertise ("senior", "10 years experience", "specialist in...")
Include industry or context ("paid media", "performance marketing", "agency environment")
Add personality traits if relevant ("pragmatic", "direct", "warm but professional")

Input

The raw material you give Gemini. Data, documents, examples, briefs, whatever it needs.

Gemini can only work with what you give it. Garbage in, garbage out. Clear, labelled inputs equal better outputs.

How to write it

Label each input clearly with headers
Tell Gemini what each input is and how to use it
Be explicit about what's included and what's not
Tip: attach files directly, or pull from Drive with the Workspace connection on

Context

The background Gemini needs: who's this for, where does it appear, what's the bigger picture?

The same task needs different outputs depending on context. Context shapes everything.

Ask yourself

Who will see, read, or use this?
What do they already know or believe?
Where does this appear or get used?
What happened before? What happens after?

Constraints

Rules, boundaries, and requirements. What Gemini must do, must not do, and how output should be shaped.

This is your biggest lever for output quality. The more specific, the less editing you'll do.

How to write it

Include what to avoid as well as what you want (DO + DON'T)
Be specific on length. Gemini defaults to verbose, so specify word count
Avoid "make it good". Define what good means explicitly

Examples

Show Gemini what good looks like. Past work, competitor examples, templates, references.

This is the most underused lever. Examples do more than instructions to calibrate quality.

How to use them

Keep a file. Build a library of your best-performing examples
Show two or three examples. Gemini spots patterns. Tell it what to take from each
Include bad examples. Show what to avoid, not just what to do

§ 04 · The Build

Pick a real task. Build it. Run it.

Pick a real task from this week. A client performance report, a campaign brief, a client email, a research summary. Build a RICCE prompt and run it live in Gemini. Ten minutes, then we compare.

Your RICCE prompt

10 mins

Fill each row. Keep it specific. Don't write "be professional" when you can write "warm, direct, no jargon, no exclamation marks." The more constraints, the less editing.

RRole

Tell Gemini who it is

e.g. "You are a senior paid media director at a performance marketing agency with 12 years of client-side and agency experience…"

IInput

Paste or attach real data

e.g. campaign export, Looker Studio screenshot, client brief, call transcript…

CContext

Audience and environment

e.g. "This goes to the client's Head of Marketing. They're time-poor, want the 'so what', and are sceptical of agency spin…"

CConstraints

Word count, tone, house style

e.g. "Under 200 words. No clichés. No 'industry-leading'. UK English. Active voice. Lead with the result, then the why…"

EExamples

Paste a recent piece you liked (and one you didn't)

e.g. "Here's a strong report we sent in March: [paste]. Here's one we wouldn't sign off again: [paste]…"

Go.

10 minutes on the clock. Build it. Run it in Gemini. We'll regroup and compare what worked.

§ 05 · The Debrief

What worked?

5 mins

Once the timer's up, look at your output and ask:

What worked well with the prompt?
Which RICCE element made the biggest difference?
Where did the output fall short, and which element would fix it?
What would you save into your prompt library?

Task 02.2 · 20 mins

Map the workflow.

Duration20 min

FormatMap & build

OutputPrompt chain

BridgeTask 01 → 03

§ 01 · The Brief

From task to chain.

Pick a mid-difficulty task from your Task 01 worksheet, ideally one you do for a real client or department. Now break it down into the three phases below. Every step becomes a prompt. Every prompt becomes a building block for the Gem you'll build in Task 03.1.

Worked example A month of campaign data lands. Goal: draft a client-ready performance brief with the headline result, the why, and three recommendations.

Phase 1

Inputs

What goes in.

›

Read the performance data

Reference document: the monthly campaign export. Read it carefully. Summarise the headline numbers in one line.

Load client context

Reference documents: client KPIs, positioning, tone guide, last quarter's report. Confirm the key tone rules in three bullets.

Phase 2

Analysis

What gets thought about.

›

Find the headline and the risk

Using Step 01, name the strongest result and the biggest underperformer against KPI. One sentence each, with the likely 'why'.

Generate three recommendations

Using Steps 01 and 02, list three actions for next month. Tie each to a KPI and flag any that need client sign-off.

Phase 3

Outputs

What gets delivered.

Draft the performance brief

Write a 250-word client-ready brief. Three sections: what happened, what we recommend, what we need from them. Append the three recommendations from Step 04, ranked by impact.

Self-audit the output

Read the brief back. Flag any line that drifts from the tone guide in Step 02 or states a number not in the data. Rewrite those lines only.

§ 02 · Your turn

Now map yours.

Same structure, your task. Pick something mid-difficulty from your Task 01 list. Write each step in plain English. Then draft the prompt that would deliver it.

Your task Write the task you've chosen, then fill each column with the steps and prompts that would deliver it.

Phase 1

Inputs

What does Gemini need to start?

›

Your input step…

Your prompt here…

Your input step…

Your prompt here…

Phase 2

Analysis

What does Gemini think about?

›

Your analysis step…

Your prompt here…

Your analysis step…

Your prompt here…

Phase 3

Outputs

What does Gemini deliver?

Your output step…

Your prompt here…

Your output step…

Your prompt here…

Task 02.2 · Worksheet

Capture your workflow in the live sheet.

Open the workflow mapping worksheet on your laptop. Map your chosen task across the three phases, then bring it into Task 03.1.

Open Task 02.2 →

docs.google.com/spreadsheets/…

Task 03.1 · 30 mins

Gemini Workshop
Basics.

Duration30 min

FormatSet up & build

OutputWorking Gem

ToolGemini Gems

§ 01 · The Set-Up

Two builds. One environment.

Work through both sides. The first gets Gemini reading your documents and replicating your output format using a Gem. The second turns a recurring task into an always-on Scheduled Action.

Build a Gem.

Set up · Feed it · Format it · Run it

Build a custom Gem that can read your documents and produce output in your format. Tick each step as you go.

Step 01 Create a new Gem Open gemini.google.com, go to the Gem manager from the left panel, click New Gem, and name it after the work you're focusing on today.
Step 02 Add your Knowledge files Under Knowledge, upload the documents Gemini needs as context: briefs, past work, style guides, reference material. Up to 10 files; for a bigger library, build a NotebookLM notebook and connect it.
Step 03 Write the Instructions & output format Pick one document you want Gemini to replicate. In the Gem's Instructions, describe the structure, headings, tone, and rules, and attach a sample as a Knowledge file so it repeats the pattern.
Step 04 Run it: input in, output out Save the Gem, then chat with it. Paste or attach the input, ask for an output in your format, and export the result to Docs or Drive.

Scheduled Action.

Design · Build · Test · Schedule

Turn a recurring task into a Gemini Scheduled Action that runs on a clock. Think market monitoring, client news updates, weekly performance digests.

Step 01 Design the recurring prompt with Gemini Pick one recurring job: market monitoring, client news updates, performance digests. Work with Gemini to shape the prompt (or a Gem) that does it.
Step 02 Build & review the output Run the prompt once in Gemini. Read what it produces. Tighten the wording, add constraints, swap examples until it fits.
Step 03 Test the outcomes Run it against two or three real inputs. Compare results. If quality drifts, refine and re-run.
Step 04 Set it as a Scheduled Action Move it from on-demand to always-on. In your prompt, tell Gemini when and how often to run it, or set it from Settings → Scheduled actions. Up to 10 active at once.

Task 03.2 · 60 mins

Advanced Custom
Task Automation.

Duration60 min

FormatPlan & build

OutputAutomation plan

Builds onTask 02.2

§ 01 · The Brief

From workflow to automation.

Take an earlier task and map it out. Pick one you scored highly in Task 1.4 and mapped in Task 2.2, then use Gemini to build the automated version as a Gem plus a Scheduled Action. The point isn't speed for its own sake. The point is to design a system that runs the parts that don't need you, and asks for you only where judgement is actually required.

Worked example Performance Reporting: raw data gathered, cleaned, KPIs calculated, trends analysed, report drafted, approved, distributed. Nine steps across three phases — the kind of workflow you'll turn into a Gem.

§ 02 · Choose your tool

Three ways to build it.

Same workflow, three levels of horsepower. Pick the one that matches your comfort level and what the task actually needs. You can always graduate from one to the next.

Start here · Beginner

Gemini Gems

No code. Save your instructions and knowledge files once, then run the workflow straight from chat. The fastest way to a working, reusable version of your task.

Level up · Automated

AI Studio

Lock a system prompt, tune the model settings, and generate an API key so the workflow can run automatically and repeatably outside the chat window.

Power user · Data pulls

Colab

Write Python in a notebook to pull from APIs, Sheets and files, push it through the Gemini API, and process large or live datasets at scale.

§ 03 · Six questions before you build

Plan it before you build it.

The build is easy. The thinking is the work. Answer these six before you open Gemini. Get them wrong and you'll spend the next hour patching gaps. Get them right and the build takes a fraction of the time.

What is the output that I want?

Be specific. Not "a summary." A 200-word client email with three bullet points, sent by Friday 9am, signed off by the AD. If you can't picture what lands in the inbox, Gemini can't either.

Format · length · audience · sign-off

Where would this require context?

Identify the steps where Gemini needs to know something it doesn't know by default. Client positioning, tone guides, sensitivity flags, past coverage, internal context. Context belongs in the Gem's Knowledge files.

Knowledge files · past examples · style guides

What would the inputs be?

List every input the workflow needs. A campaign export. A spreadsheet of contacts. A client brief. A search query. The cleaner the input, the cleaner the output. Bad inputs equal bad outputs, no matter how good the prompt.

Documents · data · search queries · triggers

What would the best data source be?

Where does the data live? A folder in Drive, a Google Sheet, a CSV, or live results via Gemini's Google Search grounding and Deep Research. If the source is messy or unreliable, fix that before you build the automation, not after.

Live search · Drive files · Apps · manual upload

Is there a way this should be scheduled?

Does this need to run on a clock? Daily at 8am for market monitoring. Weekly Monday morning for status updates. Gemini Scheduled Actions handle the clock. Schedule is what turns a Gem into a system.

On a clock · on demand · up to 10 active

What rules, guardrails or processes does it need?

This is the Gem's Instructions. Tone of voice rules. Things to never say. The exact structure of the output. The approval gate where it stops and waits for a human. Write it once, save it in the Gem, never re-explain again.

Tone · structure · guardrails · HITL gates

§ 04 · Now build it

Build it in Gemini.

These steps use Gems as the worked example, but the same flow — context, data, rules, review, schedule — applies whether you build in AI Studio or Colab. Work through them in order. Don't skip step two: the Drive structure behind it is the boring bit that makes the rest of the automation possible.

Before you start, have your answers to the six questions above written down. They feed every step below. Open gemini.google.com and make sure your Workspace connection is on.

Create the Gem

Open the Gem manager and create a new Gem. Name it after the workflow, not the client. "Weekly Client News Roundup" beats "Acme Project." If you build five clients into one Gem, it gets messy. One workflow, one Gem.

Naming convention: [Frequency] [Workflow]. Example: Daily Market Monitor, Weekly News Roundup, Monthly Coverage Report.

Set up the Drive structure

Gems don't use folders, so organise Drive instead. Create three folders: /inputs, /knowledge, /outputs. Inputs hold the raw material. Knowledge holds the context (which you'll also load into the Gem). Outputs hold what Gemini produces. With the Workspace connection on, Gemini can read these files and write results back to Docs.

If your workflow has a manual review step, add a /for-review folder too. Gemini drafts there, you approve, you move it to /outputs.

Load the Knowledge files

Answer Question 02 with files. Client positioning, tone guide, past examples of the output, key contacts, sensitivity flags, anything Gemini doesn't know by default. Upload them under the Gem's Knowledge (up to 10 files; use a NotebookLM notebook for a larger library). The more context here, the less you re-explain on every run.

Naming matters. Use clear filenames: tone-guide.md, sensitive-topics.md, past-reports.pdf. Clear names help Gemini pick the right source.

Connect the data sources

Answer Question 04 with connections. If your data lives in Drive or Sheets, turn on the relevant Apps so Gemini can reach them. If you need live results, lean on Google Search grounding or run a Deep Research pass. If you need email or calendar, connect the Gmail and Calendar Apps. Bad inputs equal bad outputs, so spend time on this step.

Test each connection independently. Ask Gemini "read the first row of the contacts sheet" before you ask it to do anything clever.

Write the Instructions

Answer Question 06 in the Gem's Instructions. Write the rules of the workflow in plain English. Tone of voice. Things to never say. The exact output structure. The phrases the client likes. The phrases the client doesn't. Save it so the Gem applies it on every run. For longer rule sets, also attach them as a markdown Knowledge file.

Write the Instructions for a junior on their first day, not for the model. If a new joiner could follow them, Gemini will too.

Map the HITL gates

Look at your Task 02.2 workflow. Mark which steps Gemini can finish alone, and which need a human to approve before moving on. Drafts go to /for-review. Final outputs go to /outputs. The art isn't automating everything. It's knowing which steps should stop and wait for you.

Default to a HITL gate before anything client-facing leaves the building. Emails, social posts, client reports. You can always remove the gate once you trust the output.

Run the first pass

Trigger the workflow manually in the Gem. Watch Gemini work through each step. Read every output. Where did it drift? Where did it nail it? Where did the HITL gate save you? Where did it miss something the gate should have caught? Refine the Instructions, refine the prompts, run again.

Run it three times with different inputs before you trust it. Consistency across three runs is the bar. One good run is luck.

Schedule it

Answer Question 05 with a Scheduled Action. Daily at 8am. Every Monday morning. Set the schedule in your prompt or from Settings → Scheduled actions, then walk away. The Gem now runs whether you're there or not. You're live.

Check the first few runs. You want to know it ran, not assume it did. Once it's stable for two weeks, leave it to it. Remember the 10 active actions limit.

Task 03.2 · Build it in Gemini

Build your mapped workflow.

Take the workflow you mapped in Task 2.2 and build it — start in Gems if you're new, or step up to AI Studio or Colab. Load your context, write the rules, run a first pass, then set it on a schedule.

Open Gemini →

gemini.google.com

Training Day, in motion.

Training Day Workshops.

Tasks, Jobs, Processes & Workflows.

Know the difference.

Task

Job

Process

Workflow

Four steps.Thirty minutes.

Task Dump

Prioritisation

Score on scale, process, data.

Scale

Process

Data

Finalise & Reality-Check

Capture your Task 01 answers in the live sheet.

Prompting that works.

Why most prompts fail.

Too vague

No context

One-shot thinking

No framework

Introducing RICCE.

Each lever, decoded.

Pick a real task. Build it. Run it.

Your RICCE prompt

What worked?

Map the workflow.

From task to chain.

Inputs

Analysis

Outputs

Now map yours.

Inputs

Analysis

Outputs

Capture your workflow in the live sheet.

Gemini Workshop Basics.

Two builds. One environment.

Build a Gem.

Scheduled Action.

Advanced Custom Task Automation.

From workflow to automation.

Three ways to build it.

Plan it before you build it.

Build it in Gemini.

Build your mapped workflow.

Training Day,
in motion.

Training Day
Workshops.

Tasks, Jobs,
Processes & Workflows.

Four steps.
Thirty minutes.

Prompting
that works.

Gemini Workshop
Basics.

Advanced Custom
Task Automation.