tech-lab adoption · networks · no-code · automation · generative AI

1. Technology adoption — the S-curve

New technologies spread by word of mouth: adoption $\frac{dA}{dt}=pq$-style contagion produces an S-shaped curve. Tune the innovation rate $p$ and imitation rate $q$ of the Bass diffusion model and watch where Rogers' adopter groups (innovators → laggards) fall.

peak adoption at
50% penetration
adopters by end

2. Network effects — Metcalfe & beyond

A platform's value grows with its users. Metcalfe's law says value $\propto n^2$; Sarnoff (broadcast) is linear $n$; Reed (group-forming) is exponential $2^n$. Drag the slider to add users and compare why "winner-take-all" dynamics emerge.

connections / value
value per user
marginal gain (+1 user)

Each new user adds value to everyone already on the network.

3. No-code app builder — visual logic

No-code tools (Bubble, Adalo, Glide) let you assemble apps from blocks instead of writing code. Click to drop UI & logic blocks onto the canvas; the platform compiles your stack into a working spec. Watch the "lines of code saved" estimate grow.

blocks placed0
≈ code saved0
build statusempty

Click a block on the canvas to remove it.

4. Automation — trigger → action pipelines

Tools like Zapier and Make connect apps: a trigger fires, data flows through filters and actions. Build a pipeline, set how many events arrive per day, and see the time saved versus doing each step by hand.

pipeline valid
hours saved / month

5. RPA & automation ROI

Robotic Process Automation replaces repetitive human tasks with software bots. A task is a good RPA candidate when it is high-volume, rule-based and stable. Score a task and see the payback period for a bot that costs a fixed amount to build.

monthly saving
payback
verdict

6. Tokenizer — how an LLM reads text

A language model never sees words — it sees tokens: sub-word chunks mapped to integer IDs. Type a sentence and watch a (simplified) byte-pair-style tokenizer split it. Note how common words are one token and rare ones fragment.

characters
tokens
chars / token
≈ cost @ $3/1M tok

Colours mark token boundaries. Real tokenizers learn merges from data.

7. Attention & next-token prediction

A transformer predicts the next token by deciding which earlier tokens to "attend" to. Pick a context and the query token; the bars show a softmax over attention scores, and the model picks the most likely continuation. This is a hand-built toy of the mechanism behind ChatGPT.

predicted next
confidence

Lower temperature = greedier; higher = more random.

8. Diffusion — generating an image from noise

Image models (Midjourney, Stable Diffusion) start from pure noise and denoise step by step toward a target guided by a prompt. Press play to watch random pixels resolve into a pattern; each step removes a slice of noise.

step0 / 30
noise remaining100%

9. Chatbot — intent routing

No-code chatbots (ManyChat, Tars) route a user message to an intent by matching keywords, then fire the matching reply. Type a customer message and watch it score against each intent and fall through to a fallback if nothing matches.

matched intent
confidence
bot reply

Intents: greeting · hours · pricing · support · goodbye. Else → fallback.

10. No-code database — filter & query

Airtable and Google Sheets store data as records in a table. Build a filter the way you would in a no-code view: choose a field, an operator and a value, and watch the matching rows highlight — the same logic a WHERE clause runs under the hood.

rows matched
sum MRR (matched)

11. Technology with impact — prioritisation matrix

The course asks: which technology should we apply to a social problem? Score candidate solutions on impact (how much good) and feasibility (how buildable with today's no-code / AI tools). The matrix plots each on the classic 2×2; the top-right quadrant is "do this first".

quadrant
priority score

Each solution remembers its own scores — switch to compare.