The AI Breakdown

AI 101

Use AI to learn, create, build, and work better.

Curate the context. Pick the environment. Verify the result.

Presented by Adam Gurski • adamgurski-creative.com • Verified 2026-08-27 • Model references checked against official sources
Scan to open the AI 101 Toolkit
Scan for resources Tools, prompts, and links from tonight’s session.
https://ai101-toolkit.netlify.app/

Use arrow keys or buttons to navigate

One prompt, mid-edit
02 / Why listen to me

Why I’m standing up here

Three years ago I’d never built a website. AI didn’t do the work for me — it made the next step obvious enough to take.
Visuals for a Grammy winnerMy video, on Fatboy Slim’s stage. That one, playing right there →
Real business websitesDesigned, shipped, live — and I started from zero.
Started my own companyResearch, branding, launch — done before I could talk myself out of it.
Apps & gamesSolo builds that used to take a full dev team.
No degree. No team. No permission slip. Just curiosity.
My visual — on Fatboy Slim’s stage
03 / Serious work

AI is doing serious work

The same tech behind all of this is sitting in your pocket. Worth learning how to drive it.

SCIENCE AT SCALE

AlphaFold helped solve protein structure; GNoME proposed 2.2M candidate crystals.

AI helps search spaces too large for people alone.
CREATE + BUILD

Image models, coding agents, and deploy tools turn plain-language intent into visuals, websites, documents, and workflows you can inspect.

Proof: Microsoft VASA-1 - one photo + one audio clip becomes a lifelike talking face, in real time.

The point is not chat. The point is output.
IMO GOLD 2025

Gemini Deep Think solved 5 of 6 IMO problems and scored 35/42, reaching the official gold threshold. (IMO 2025)

The frontier moved from calculators to proofs.
More proof - click any
04 / Model map

Pick the right room

Start with the environment that fits the job. The detailed hosted and open/local directory lives in the toolkit.

ChatGPT logo

ChatGPT

OpenAI · GPT-5.6 family

Luna for speed, Terra for everyday work, Sol for the hardest tasks.

The All-Rounder
Claude logo

Claude

Anthropic · Claude Fable 5

Anthropic’s most capable generally available model — built for long projects, careful writing, and Claude Code.

The Writer / Coder
Gemini logo

Gemini

Google · Gemini 3.7 Flash

Plugged into Google — Search, Workspace, AI Studio.

The Integrator
Hosted choices + open/local familiesModel names move fast. Verified 2026-08-27 · compare access and licenses in the toolkit.
05 / Source grounding
The first unlock

AI gets better when it has your material.

Before you ask, hand it your stuff — the notes, the files, the examples, the constraints. That’s the difference between a guess and useful work.

NotebookLM Your sources Cited answers
Customer notesGoals, constraints, questions, rough details.
Source filesPDFs, docs, links, policies, research.
ExamplesBrand voice, past work, preferred formats.
Open my live notebooknotebooklm.google.com — a real grounded workspace
NotebookLM logo Grounded workspace
"Based on your sources, here are the three strongest options, what evidence supports them, and what still needs checking."
Source 1 Source 2 Open question
06 / Context + control

Three levels of context and control

Move from a conversation, to a bounded project, to an agentic workspace as the job becomes more durable and complex.

Three levels of context and control

Give AI the right-sized room.

Memory is not magic. Better work comes from context you can see, shape, and remove.

01Chat

Fast questions and one-off drafts. You bring the context into the conversation.

02Project

Persistent files, instructions, and examples for a body of work.

03Codex / agentic harness

Inspectable, user-controlled context plus tools, files, checks, and an approval loop.

Watch: context and memory control →
Recorded walkthroughFrom a chat to a working canvasClick the clip to watch the full build.
Your source material
Your project rules
Your preferred examples
Codex logoInspectable workspace
1Reads only the context you provide.
2Works with tools inside clear boundaries.
3Shows what changed for your approval.
07 / Prompt lab

Turn a vague ask into a working brief

If you remember one formula tonight, make it this one: Role · Task · Context · Format. Click any block — watch it light up in a real prompt.

“Make me a snake game”AI guesses the look, the controls, the rules — you get a coin flip.
The RTCF briefSame ask, fully specified — you control what comes back.
R
Role
Who should the AI act as?
T
Task
What exact job should it do?
C
Context
What facts and limits should guide it?
F
Format
What should the answer look like?
Client Reply 1 / 1
SCORE  0042
What the brief buys you — built live, in the demoprompt 1 of 6 → canvas
08 / Live demo board

Watch it happen, live

The long jobs launch right now and cook while we talk. The quick hits fill the gaps between sections. Nothing on this board is pre-baked.

Launching now — collect later

L1

Deep research

A real question from this room goes into deep-research mode. We read the cited report before you leave.

gemini deep research · cooks ~20 min

L2

Agentic build

A real brief becomes a working artifact inside a tool-enabled project, with visible files, checks, and approval.

Codex / agentic harness · deploys while we talk

precedent: one creator, a few hours, a live browser MMO flight sim →

Quick hits — between sections

Q1

Five-level explanation

One topic from the room, explained from toddler to PhD.

Q2

Restoration

A damaged black-and-white print, repaired and colorized in front of you.

Q3

On-brand creation

A logo, a truck, and a visual reference become a matching campaign asset.

Launch the long jobs, teach the method, then collect the results on stage.
09 / Human judgment

Keep human judgment

Use four checkpoints on consequential work: define the outcome, delegate the right work, discern what is true and useful, and document what you approved.

01 / Define What outcome and boundary matter?

Name the outcome, audience, evidence, constraints, and privacy boundary.

The letterA dense insurance letter lands. AI’s job: translate it, draft the reply. Your job: the decision.
02 / Delegate What should AI actually do?

Assign drafts, comparisons, practice, summaries, and first versions.

The letter“Plain-language insurance explainer” + the letter pasted in + reply under 150 words.
Human
in
control
03 / Discern Is it good and true?

Check facts, dates, citations, numbers, tone, and missing assumptions.

The letterCheck the deductible, the dates, and the claim number against the actual letter.
04 / Document What should another person be able to inspect?

Record sources, decisions, meaningful AI help, checks, and final approval.

The letterRedact the policy number going in. Sign what goes out.
Why the loop matters: In phase two, newer models, better tools, and a manager agent largely stopped money-losing weeks — but humans still did plenty.THE STORY →
THE AI
BREAKDOWN
“It made the next step obvious enough to take.”
10 / Your next move

Run the loop on one real task

1Pick one annoying task
2Gather the context
3Choose the environment
4Prompt with RTCF
5Check with the 4Ds
Your dare: replace one 10-minute chore with AI tonight. Grab a real note, a messy photo, a draft email, or a question you’ve been sitting on. Start with your own stuff — then let me know how it goes.
Start with sources
Scan to open the AI 101 Toolkit https://ai101-toolkit.netlify.app/
adam@adamgurski-creative.com · @Gingerbot5000