Who’s who in AI — the ecosystem

There is no single “an AI.” There’s a whole cast.

What we lazily call “an AI” is really a troupe of specialists — text models, image models, agents, retrieval, memory — wired together into a system. Here is the full cast, what each one actually does, and how they play together.

How it learns Pre-trained — frozen after training Pre-trained + RLHF — tuned to human preference Adaptive — keeps learning Retriever + generator
The whole cast on one page

One orchestrator, calling a bench of specialists.

This is the shape of almost every “AI product.” An agent in the middle takes your goal and calls the parts it needs — the generators that make things, plus retrieval and memory to stay grounded and personal. The dot on each node is its learning mode from the key above. And once these agents start transacting on their own, there’s a layer forming underneath the whole thing — shown below.

AI agent
The orchestrator

Takes the goal, decides which parts to call, and pieces the answer together.

Language model
TEXT TEXT
Multimodal
MULTI MULTI
Retrieval (RAG)
QUERY SOURCES
Image generation
TEXT IMAGE
Video generation
TEXT VIDEO
Speech-to-text
AUDIO TEXT
Memory
HISTORY CONTINUITY
Text-to-speech
TEXT AUDIO
The orchestrator can call this part Generators — make things Connective tissue — coordinate
↓ The layer forming underneath
Our take — the trust & settlement layer

The cast can generate and coordinate. It can’t prove.

None of the parts above can, by themselves, show where their data came from, prove an output wasn’t tampered with, or let one agent pay another. As agents get more autonomous, we think that job lands on a shared, tamper-evident ledger — a blockchain — sitting underneath the whole system.

Why we believe this →
01
Data integrity
Provenance for every dataset, model, and output.
02
Trust
Cryptographic proof instead of “trust us.”
03
Agent economics
Identity, permissions, and payment between agents.
The part everyone skips — how AI learns

A model’s “learning mode” decides what you can trust it to do.

People assume AI is “always learning” from them in real time. Mostly, it isn’t. When and how a system learns is the single most useful thing to understand — it explains why one tool has a knowledge cutoff, why another cites its sources, and why a third slowly gets better the more you use it. These are the four modes, and the colored dot on every card above tells you which one it uses.

Mode 01

Pre-trained & frozen

Learns once, then stops. Image & video gen, speech models.

The model trains on a huge dataset once, and then that knowledge is locked in place. Using it does not teach it anything — every request starts from the same frozen brain. This is why a model has a “knowledge cutoff” and hasn’t heard of anything after its training date.

THINK OF IT AS A printed textbook — authoritative, but never updates itself.
Mode 02

Pre-trained + RLHF

Reinforcement Learning from Human Feedback
Trained, then coached by humans. ChatGPT, Claude, Gemini.

Same frozen knowledge as Mode 01, but with an extra step: humans rank the model’s answers, and it’s tuned to prefer the helpful, honest, harmless ones. RLHF — reinforcement learning from human feedback — is why chatbots feel polite and on-task rather than like raw autocomplete. Crucially, this coaching happens before release; your individual chats don’t retrain it.

RLHF, in plain English
1. It drafts a few answers
The model writes several possible replies to the same prompt.
2. People rank them
Human reviewers mark which reply is more helpful, honest, and safe.
3. It’s retrained to match
The model is tuned to give more answers like the ones people preferred.
THINK OF IT AS A brilliant new hire put through etiquette training before their first day.
Mode 03

Retriever + generator

Looks it up, then answers. Perplexity, RAG, Copilot.

The model itself is still frozen — but before answering, the system fetches fresh, relevant documents and hands them over as reference. The model doesn’t know more; it just gets to read the right page at the right moment. This is how a tool can cite today’s news or your internal wiki without ever being retrained.

THINK OF IT AS A sharp analyst who Googles and opens the file drawer before replying.
Mode 04

Adaptive

Actually keeps learning. Agents, memory & recommendations.

The only mode that genuinely changes over time. It stores what it sees — your preferences, past conversations, what you clicked — and adjusts its behavior accordingly. Powerful and personal, but it’s also where the real privacy questions live, because the system is keeping a record of you to do its job.

THINK OF IT AS A personal assistant who remembers you and gets better with every week.
The myth to drop

“The AI is learning from everything I type.” For the chatbot in front of you, that’s usually false — it’s frozen (Modes 01–02), and your chat is forgotten when the window closes unless a memory layer (Mode 04) is explicitly switched on. Knowing which mode you’re talking to tells you whether to expect a knowledge cutoff, a citation, or a system that quietly remembers you.

The generators

— they make things

Each takes one kind of input and produces one kind of output. On their own they’re narrow; that narrowness is exactly why they’re good.

Language model

TEXTTEXT

Predicts the next word to write, summarize, answer, and translate. The workhorse of the toolbox.

GPT-4 Claude Llama

Multimodal model

TEXT·IMG·AUDIOSAME

Built from the ground up to take in — and produce — several modalities inside one unified model. No stitched-together pipeline.

GPT-4o Claude 3.5 Sonnet Gemini

Image generation

TEXTIMAGE

Turns a prompt into pictures, mockups, and illustrations — trained once on millions of image–text pairs, then frozen.

DALL·E Midjourney Stable Diffusion

Video generation

TEXTVIDEO

Generates clips and animation from a prompt — promos, explainer motion, B-roll. The newest and least predictable member.

Sora Runway Veo

Speech-to-text

AUDIOTEXT

Transcribes spoken audio. The front door for voice interfaces, meeting notes, and accessibility.

Whisper Google STT AssemblyAI

Text-to-speech

TEXTAUDIO

Reads text aloud in a natural voice — assistants, narration, and localization. The other half of the voice loop.

ElevenLabs Google TTS Azure Neural

The connective tissue

— they make the parts work together

These don’t generate content so much as coordinate it: setting goals, grounding answers in real sources, and remembering what came before.

AI agent

GOALACTIONS

Given a goal, it plans, calls tools and APIs, holds context, acts, and adjusts from the results. An assistant, not just an answer.

AutoGen LangChain

Retrieval (RAG)

QUERYGROUNDED ANSWER

Searches your documents or the web first, then feeds what it finds to a model — so answers stay current and citable.

Glean Perplexity

Memory & personalization

HISTORYCONTINUITY

Remembers preferences and past conversations across sessions — the difference between a tool and something that feels like a relationship.

Vector stores Profile layers
The four distinctions that trip people up

Most confusion about AI is really one of these mix-ups.

Language model vs. multimodal
Language model

Text-first. Some now read images, but language is the core.

Multimodal

Text, image, and audio handled natively in one model.

Pre-trained model vs. agent
Model

Answers your question, then stops.

Agent

Pursues your goal — plans, calls tools, and iterates until done.

Standard LLM vs. RAG
Standard LLM

Knows only its training data — can go stale or make things up.

RAG

Pulls live sources first, then answers with receipts.

Speech-to-text vs. text-to-speech
Speech-to-text

The ears: audio in, text out.

Text-to-speech

The voice: text in, audio out. Together they close the loop.

How the parts combine

Almost every “AI product” is several of these in a row.

Once you can name the parts, the products stop being magic. Here are four common assemblies.

Voice support assistant
Speech-to-text Language model RAG Memory Text-to-speech

Hears the spoken question, interprets it, pulls the right answer from your account, remembers past tickets, and replies in a natural voice.

Content studio
Language model + Image generation + Video generation + Text-to-speech

One brief in; the copy, the visuals, a short video, and a voiceover all come out the other side.

Autonomous research assistant
Agent RAG Language model + Memory

Plans its own search, reads the sources, tracks what it has covered, and writes the literature review — checking in along the way.

Personal AI assistant
Language model + Memory + Agent + RAG

Knows you, remembers context, and actually takes action on your behalf — the assembly everyone is racing to build.

Why the modular design matters

The power doesn’t come from one giant brain. It comes from clever arrangements of ordinary, specialized parts.

Flexible

Swap a part without rebuilding the system.

Scalable

Strengthen one component at a time.

Specialized

Each piece is best-in-class at its job.

Powerful

Combined, they outdo any single model.

Adaptable

New parts slot in as the field moves.

KEEP READING
AI broke the internet’s honor system. Blockchain is one of the repair crews.