How AI Systems Actually Work
Every AI system is built from five distinct layers. Understanding these layers transforms AI from magic into mechanics.
Layer 1
Communication Interface
How humans interact with AI
Input Methods
Typing / Text
Voice (Wispr Flow, Dictation)
Camera / Vision
File Upload
Screen Share
Device Types
Desktop App
Mobile App (iOS/Android)
Web Browser
CLI / Terminal
IDE Integration
Messaging Bot (Telegram, Slack)
Examples
Claude.ai web interface
ChatGPT mobile app
Claude Code CLI
Cursor IDE
Telegram bot
Layer 2
Harness / Orchestrator
The cockpit that directs the brain
What It Does
Sends prompts to the brain
Manages context window
Connects tools and APIs
Handles memory / history
Orchestrates multi-step tasks
Examples
Claude Code (wraps Claude brain)
ChatGPT Interface (wraps GPT brain)
Cursor (uses multiple brains)
Windsurf (uses multiple brains)
Manus (autonomous agent)
Cyndra (custom orchestrator)
Key Insight
Sometimes harness + brain = same brand (ChatGPT + GPT)
Sometimes different (Cursor + Claude)
Layer 3
Brain / Model (LLM)
The actual intelligence doing the thinking
Claude Family (Anthropic)
Haiku ... fast, cheap
Sonnet ... balanced, popular
Opus ... deepest reasoning
GPT Family (OpenAI)
GPT-4o ... flagship
GPT-4o-mini ... fast, cheap
o1 / o3 ... reasoning models
Gemini Family (Google)
Flash ... fast
Pro ... balanced
Ultra ... powerful
Others
Grok (xAI)
Llama (Meta, open-source)
Mistral
Perplexity (search-focused)
Key Insight
Brains are INTERCHANGEABLE in some harnesses
Cursor lets you pick Claude OR GPT
Layer 4
Tools / Actions
What the brain can actually do
Web Search / Research
Code Writing / Execution
Image Generation (DALL-E, Imagen, Midjourney)
Music Creation (Suno, Udio)
Video Generation (Sora, Runway, HeyGen)
Data Analysis / Visualization
File Management (read / write / edit)
API Integrations (CRM, email, calendar)
Browser Automation (Puppeteer, Playwright)
Voice Cloning / Text-to-Speech
Document Analysis
Presentation Generation
Layer 5
Output / Destination
Where the results end up
Chat Response (in-interface)
Artifacts / Canvas
File Exports (PDF, DOCX, images)
Deploy to Web
Post to Social Media
Send Email
Update CRM / Database
Push Notification
Calendar Event
Spreadsheet
Claude
by Anthropic
Best coding, longest context, most reliable instruction following
Claude is built by Anthropic with a focus on safety and helpfulness. It excels at complex reasoning, coding, long document analysis (200K token context), and following nuanced instructions. The model family ranges from Haiku (fast and affordable) to Opus (deepest reasoning).
Interface
Web, iOS/Android, CLI (Claude Code), API
Harness
Claude.ai, Claude Code, Cursor, Windsurf
Brain
Haiku, Sonnet, Opus
Tools
Artifacts, Projects, MCP, Computer Use, Vision
Output
Chat, Artifacts, Files, API responses, Deployed apps
Model Tiers
- Haiku ... Fastest, most affordable. Great for simple tasks, classification, customer support.
- Sonnet ... The balanced workhorse. Best value for most business tasks. Strong at coding and analysis.
- Opus ... Deepest reasoning and most capable. Best for complex strategy, research, and multi-step problem solving.
Key Tools
- Artifacts ... Interactive code, documents, and visualizations generated inline
- Projects ... Upload documents and set custom instructions for persistent context
- MCP (Model Context Protocol) ... Connect to external tools and data sources
- Computer Use ... Can see and interact with your screen
Pricing
Free $0
Pro $20/mo
Team $30/mo
Max $100-200/mo
API Pay per token
ChatGPT
by OpenAI
Most widely known, strong general purpose, integrated image and video
ChatGPT brought AI to the mainstream. It offers a polished experience across text, image generation (DALL-E), code interpretation, web browsing, and now video generation (Sora). GPT-4o is the flagship model, while o1/o3 handle advanced reasoning tasks.
Interface
Web, iOS/Android, Desktop App, API
Harness
ChatGPT native, API Playground, Custom GPTs
Brain
GPT-4o, GPT-4o-mini, o1, o3, GPT-4.5
Tools
DALL-E, Code Interpreter, Browse, Sora, Voice
Output
Chat, Canvas, Files, Images, Videos
Model Tiers
- GPT-4o ... Flagship model. Fast, multimodal (text + vision + audio), strong general purpose.
- GPT-4o-mini ... Lightweight and affordable. Good for simple tasks at scale.
- o1 / o3 ... Reasoning models. Think step-by-step for math, science, and complex logic.
Key Tools
- DALL-E ... Image generation directly in chat
- Canvas ... Side-by-side document editing and code writing
- Code Interpreter ... Run Python code, analyze data, create charts
- Sora ... Generate videos from text descriptions
- Advanced Voice ... Real-time spoken conversation
Pricing
Free $0
Plus $20/mo
Team $30/mo
API Pay per token
Gemini
by Google
Google ecosystem integration, 1M+ token context, multimodal
Gemini is Google's AI, deeply integrated with Google Workspace (Gmail, Docs, Drive, Calendar). It has the longest context window in the industry (1M+ tokens), making it powerful for analyzing large documents. Available across consumer and enterprise tiers.
Interface
Web, Google Workspace sidebar, API
Harness
Gemini.google.com, AI Studio, Vertex AI, Workspace
Brain
Flash, Pro, Ultra, Nano (on-device)
Tools
Search grounding, Code exec, Workspace actions, Deep Research
Output
Chat, Docs/Sheets/Slides, Files, Workspace integration
Model Tiers
- Flash ... Fastest response times. Ideal for high-volume, simple tasks.
- Pro ... Balanced performance. Good for business analysis and content.
- Ultra ... Most capable. Complex reasoning and multimodal tasks.
- Nano ... Runs on-device (Pixel phones). Private, offline-capable.
Pricing
Free $0
Advanced $20/mo (with Google One)
Enterprise per Workspace seat
API Pay per token
Grok
by xAI (Elon Musk)
Real-time X/Twitter data, unfiltered responses
Grok is built by xAI and deeply integrated with X (formerly Twitter). Its unique advantage is real-time access to X's firehose of data, making it exceptional for trending topics, breaking news, and social sentiment analysis. Known for more unfiltered, direct responses.
Interface
grok.com (web), X/Twitter integration
Harness
Grok native, X integration
Brain
Grok-2, Grok-3
Tools
Image generation (Aurora), Real-time search
Output
Chat, Images, X post drafts
Standout Features
- Live access to all X/Twitter posts and trends
- Aurora image generator
- Less content filtering than competitors
- "Fun mode" for more personality in responses
Pricing
Free with X Basic access
SuperGrok $30/mo subscription
Perplexity
by Perplexity AI
Search-first AI with citations and real-time info
Perplexity reimagines search by combining AI reasoning with live web data. Every answer comes with cited sources, making it the go-to for research where you need to verify claims. It uses multiple AI backends under the hood, including Claude and GPT models.
Interface
perplexity.ai (web), iOS/Android apps
Harness
Perplexity native (search-optimized)
Brain
Multiple backends (Claude, GPT, own models)
Tools
Web search w/ citations, File analysis, Image gen
Output
Cited answers, Research pages, Shared collections
Why It Stands Out
- Every claim links to its source (clickable citations)
- Real-time web data, not just training data
- "Focus" modes: Academic, YouTube, Reddit, Writing, etc.
- Shareable research "Pages" for team collaboration
Pricing
Free Limited daily searches
Pro $20/mo
Manus
by Manus AI
Fully autonomous agent for complex multi-step tasks
Manus operates differently from chat-based AI. Instead of answering questions, it completes entire projects autonomously. Give it a task and it will browse the web, write code, manage files, and deliver finished work. Think of it as hiring a virtual contractor rather than asking a question.
Interface
manus.im (web)
Harness
Autonomous agent orchestrator
Brain
Multiple models (proprietary orchestration)
Tools
Browser automation, File mgmt, Code execution
Output
Complete project deliverables, Reports, Websites
Specialty AI Tools
Purpose-built tools that excel at one specific type of output
Music Generation
- Suno.AI ... Full songs from text descriptions. Lyrics, melody, instruments, vocals.
- Udio ... High-quality music generation with genre control and stem separation.
Video Generation
- Sora (OpenAI) ... Text-to-video, integrated with ChatGPT Plus.
- Runway ... Professional video generation and editing.
- Kling ... High-quality video with motion control.
- HeyGen ... AI avatar videos. Upload your face, type a script, get a video.
Image Generation
- Midjourney ... Highest quality artistic images. Discord-based.
- DALL-E (OpenAI) ... Built into ChatGPT. Easy to use.
- Imagen (Google) ... Built into Gemini. Photorealistic.
- Stable Diffusion ... Open-source. Run locally or via API.
Voice and Audio
- ElevenLabs ... Voice cloning and text-to-speech. Clone your voice from a sample.
- Wispr Flow ... Voice-to-text dictation that works anywhere on your computer.
Code-Focused IDEs
- Cursor ... AI-powered code editor. Lets you choose Claude, GPT, or other brains.
- Windsurf ... Another AI code editor with multi-model support.
- Claude Code ... Terminal-based coding agent. Reads, writes, and deploys code.
Presentations and Design
- Gamma ... Generate full presentations from a prompt. Auto-designs slides.
- Canva AI ... AI-assisted design within Canva's platform.
What Do You Want to Do?
Start with the outcome you need, then find the best tool for the job.
Research
Find information, analyze documents, summarize complex topics, get cited answers from the web.
Perplexity
Claude
ChatGPT
Gemini
Grok
Best for research with citations: Perplexity. Best for deep document analysis: Claude (200K context).
Writing and Content
Emails, blog posts, social media, marketing copy, proposals, reports, books. Any written content.
Claude
ChatGPT
Gemini
Best for natural, nuanced writing: Claude. Best for short-form copy: ChatGPT. Best for SEO content: Gemini (search grounding).
Coding and Development
Write code, build applications, debug issues, create websites, automate workflows.
Claude / Claude Code
Cursor
ChatGPT
Windsurf
Best overall coding AI: Claude (via Claude Code or Cursor). Claude Code can build and deploy entire applications autonomously.
Image Creation
Generate photos, illustrations, logos, marketing graphics, product mockups from text descriptions.
Midjourney
DALL-E (ChatGPT)
Imagen (Gemini)
Stable Diffusion
Grok Aurora
Best artistic quality: Midjourney. Easiest to use: DALL-E (built into ChatGPT). Best for free: Stable Diffusion.
Music Creation
Generate full songs with vocals, instrumentals, and production from text prompts or lyrics.
Suno.AI
Udio
Best for quick full songs: Suno.AI. Best for production quality and stems: Udio.
Video Generation
Create videos from text, generate AI avatar presenters, edit footage, and produce short-form content.
Sora (OpenAI)
Runway
Kling
HeyGen
Best for AI avatars (talking head): HeyGen. Best for creative video: Runway. Best integrated: Sora (in ChatGPT).
Data Analysis
Upload spreadsheets, analyze trends, create charts, find patterns in your business data.
ChatGPT (Code Interpreter)
Claude
Gemini
Best for visual charts: ChatGPT Code Interpreter. Best for large datasets: Gemini (1M context). Best for nuanced analysis: Claude.
Automation
Automate repetitive workflows, connect apps, schedule tasks, build agents that work while you sleep.
Claude Code + MCP
Manus
Zapier AI
Make.com
Most powerful: Claude Code with MCP tools (Sterling is built on this). Most accessible: Zapier AI for no-code automations.
Voice and Audio
Voice cloning, text-to-speech, transcription, real-time voice conversations with AI.
ElevenLabs
ChatGPT Voice
Wispr Flow
Gemini Live
Best voice cloning: ElevenLabs. Best real-time voice chat: ChatGPT Advanced Voice. Best dictation: Wispr Flow.
Real-World Workflows
See how the five layers connect in practice. Each arrow shows data flowing from one layer to the next.
The Sterling AI Setup
How Brad's AI chief of staff operates. A fully autonomous agent that manages operations 24/7.
Telegram
Interface
Cyndra
Harness
Claude Opus
Brain
50+ MCP Tools
Tools
GChat, Email, Web, Social
Output
Creating a Social Media Post
From a voice memo idea to published content across multiple platforms.
Voice Memo
Interface
Claude Code
Harness
Claude Sonnet
Brain
Image Gen + Scheduler
Tools
FB, IG, LinkedIn
Output
Quick Research Task
Ask a question and get a sourced, cited answer in seconds.
Web Browser
Interface
Perplexity
Harness
Multiple Models
Brain
Live Web Search
Tools
Cited Answer
Output
Building a Website
Go from idea to deployed website using AI-powered coding tools.
Terminal / IDE
Interface
Claude Code / Cursor
Harness
Claude Sonnet
Brain
Code, Files, Deploy
Tools
Live Website
Output
Pricing Comparison
What each platform costs across free, consumer, and business tiers.
| Platform | Free Tier | Consumer Paid | Team / Business | API / Pay-per-use | Best For |
|---|---|---|---|---|---|
| Claude | Yes Limited usage | Pro: $20/mo, Max: $100-200/mo | Team: $30/seat/mo | Haiku ~$0.25, Sonnet ~$3, Opus ~$15 per M tokens | Coding, long docs, reliability |
| ChatGPT | Yes GPT-4o limited | Plus: $20/mo | Team: $30/seat/mo | GPT-4o ~$2.50/$10 per M tokens | General purpose, images, video |
| Gemini | Yes Pro model | Advanced: $20/mo (incl. Google One) | Enterprise: per Workspace seat | Flash: $0.075, Pro: $1.25 per M tokens | Google Workspace, long context |
| Grok | Yes with X account | SuperGrok: $30/mo | N/A | API available | Real-time X data, trending |
| Perplexity | Yes Limited searches | Pro: $20/mo | Enterprise plans available | Sonar API available | Cited research, fact-checking |
| Midjourney | No | Basic: $10/mo, Standard: $30/mo | Pro: $60/mo | N/A | Artistic image generation |
| ElevenLabs | Yes Limited | Starter: $5/mo, Creator: $22/mo | Scale: $99/mo | Per-character pricing | Voice cloning, TTS |
| Suno.AI | Yes 5 songs/day | Pro: $10/mo | Premier: $30/mo | N/A | Music generation |
| HeyGen | Yes 1 free video | Creator: $24/mo | Business: $72/mo | Per-video pricing | AI avatar videos |
| Cursor | Yes Limited | Pro: $20/mo | Business: $40/seat/mo | N/A | AI-powered code editor |
What Should You Use? Decision Guide
I just want to try AI for the first time, no commitment.
Start with Claude Free or ChatGPT Free. Both let you chat immediately with no credit card. If you want search with citations, try Perplexity Free.
I want one paid AI subscription for my business. Which one?
If you do lots of writing and coding: Claude Pro ($20/mo). If you want the broadest feature set (images, video, voice, code): ChatGPT Plus ($20/mo). If you live in Google Workspace: Gemini Advanced ($20/mo).
I need AI to automate my entire operation.
Claude Max ($100-200/mo) + Claude Code. This is the Sterling AI model. Autonomous agents running 24/7 with access to all your business tools via MCP connections. The most powerful setup available today.
I want to create marketing images for social media.
Quickest start: ChatGPT Plus ($20/mo) with DALL-E built in. Highest quality: Midjourney ($10-30/mo). Both are excellent. ChatGPT is more convenient, Midjourney produces more artistic results.
I want to create video content with AI.
For AI avatar "talking head" videos: HeyGen ($24/mo). For creative text-to-video: Sora (included in ChatGPT Plus) or Runway. For repurposing existing video: CapCut.
I need to do research and want verified, cited answers.
Perplexity Pro ($20/mo) is the clear winner. Every answer comes with clickable source citations. Think of it as Google Search that reads all the results for you and synthesizes an answer.
Glossary
Key terms explained in plain language.
LLM (Large Language Model)
The "brain" of AI systems. A massive neural network trained on text data that can understand and generate human language. Examples: Claude, GPT-4, Gemini.
Token
The unit AI models use to process text. Roughly 3/4 of a word. "Hello, how are you?" is about 6 tokens. Pricing and limits are measured in tokens.
Context Window
The amount of text an AI can "see" at once. Claude has 200K tokens (~150K words). Gemini has 1M+. Bigger context = can process longer documents.
Prompt
The instruction or question you give to an AI. Better prompts get better results. Think of it like giving clear directions to an employee.
Agent
An AI that can take actions autonomously, not just answer questions. It can use tools, browse the web, write code, and complete multi-step tasks on its own.
Harness / Orchestrator
The software wrapper around an AI brain that manages how prompts are sent, tools are connected, and context is maintained. Claude Code and Cursor are harnesses.
MCP (Model Context Protocol)
A standard created by Anthropic for connecting AI to external tools and data. Like USB for AI, it lets models plug into any compatible tool.
API (Application Programming Interface)
A way for software to talk to other software. APIs let you use AI models in your own applications, pay per use, and build custom workflows.
Fine-tuning
Training an existing AI model on your specific data to make it better at your particular use case. Like giving an employee specialized training.
RAG (Retrieval-Augmented Generation)
A technique where the AI searches your documents first, then uses what it finds to generate an answer. Reduces hallucinations by grounding answers in real data.
Hallucination
When an AI generates information that sounds plausible but is factually wrong. All AI models can hallucinate. Verification is always important.
Multimodal
An AI that can process multiple types of input: text, images, audio, video. Most modern AI models are multimodal.
Temperature
A setting that controls how creative/random AI responses are. Low temperature = more focused and predictable. High temperature = more creative and varied.
Embedding
Converting text into numbers so AI can understand similarity and meaning. Used for search, recommendation, and finding related content.
Inference
The process of an AI generating a response. Every time you ask an AI a question, it runs "inference" to produce the answer. This is what costs money on API pricing.
Reasoning Model
A model designed to think step-by-step before answering (like o1 and o3 from OpenAI). Better for math, logic, and complex problems. Takes longer but more accurate on hard questions.
Artifacts
Claude's feature for generating interactive content (code, documents, visualizations) in a side panel. You can edit, iterate, and export them.
Computer Use
An AI's ability to see and interact with a computer screen, just like a human would. Claude's computer use feature can click, type, and navigate software.