2024: The Year AI Became an OS Feature

2024 was the year AI moved from apps to operating systems. Apple Intelligence, Microsoft Copilot PC, and local LLMs became default features. Here’s how the landscape shifted.

The Big Picture

2023: "Check out this AI app"
2024: "AI is built into my computer"

The transformation from software to infrastructure.

Apple Intelligence

What Shipped

• Writing Tools: Rewrite, proofread, summarize everywhere
• Image Generation: Image Playground, Genmoji
• Siri 2.0: ChatGPT integration, on-device processing
• Priority notifications: AI-ranked importance

The Technical Approach

On-device models:
├── Small models for quick tasks
├── >3B parameters running on Neural Engine
└── Privacy preserved by default

Cloud (Private Cloud Compute):
├── Larger tasks
├── End-to-end encryption
└── Attestable compute environment

Apple prioritized privacy over capability.

Developer Integration

// Writing Tools SDK
let writingToolsController = WritingToolsController()
textView.allowsWritingToolsEditing = true

// Intelligence actions
let action = SiriAction(name: "Summarize meeting")

System-wide AI with consistent UX.

Microsoft Copilot+ PC

Requirements

NPU: 40+ TOPS
RAM: 16GB minimum
Storage: 256GB SSD

Qualcomm Snapdragon X Elite or compatible

New hardware category for AI PCs.

Key Features

• Recall: Searchable memory of everything on screen
• Cocreator: Real-time image generation in Paint
• Live Captions: Real-time translation
• Windows Studio Effects: Background blur, eye contact

Recall (Controversial)

What it does:
├── Screenshots every few seconds
├── OCR and indexes all content
├── Semantic search across history
└── "Find that email from last week about the project"

Concerns:
├── Privacy implications
├── Security of stored data
├── Delayed launch after backlash

Power users loved it. Security researchers worried.

Local LLMs Everywhere

Ollama Adoption

Downloads (2024):
├── January: 500K
├── June: 2M
├── December: 5M+

Running on:
├── M1/M2/M3 Macs
├── Gaming PCs with GPUs
├── Even Intel laptops (slowly)

Local AI became normal.

Model Progress

January:  Mistral 7B best open model
April:    Llama 3 8B matches GPT-3.5
October:  Qwen 2.5 challenges GPT-4

Quality improved faster than expected.

Real Impacts

Developer Workflow

2023:
- GitHub Copilot in VS Code
- Occasional ChatGPT tab

2024:
- Cursor as primary editor
- Claude/ChatGPT always open
- Local models for privacy
- AI in every tool

AI became expected, not novel.

Consumer Experience

Email: Smart compose everywhere
Photos: AI editing default
Search: AI summaries standard
Phones: On-device everything

Enterprise Adoption

2023: "Should we allow ChatGPT?"
2024: "Which AI tools are mandatory?"

CIOs went from caution to requirement.

The Concerns Materialized

Deepfakes

2024 election cycle:
├── Fake calls with cloned voices
├── Generated "evidence" went viral
├── Platform scramble for detection

Copyright

NY Times lawsuit:   Ongoing
Getty lawsuit:      Ongoing
Writers Guild:      Secured AI contract terms
Artists:            Still fighting

Jobs

Some disruption:
├── Content farms: Significant layoffs
├── Customer service: Automation accelerated
├── Translation: Quality good enough for many

Less than feared:
├── Software engineering: More productive, not replaced
├── Creative work: AI as tool, not replacement
├── Knowledge work: Augmented, not automated

What Didn’t Happen

No AGI

Predictions: "AGI by end of 2024"
Reality:     Better tools, not general intelligence

LLMs got better at more things,
but still fundamentally the same technology.

No AI Winter

Investment:  Still flowing
Progress:    Continued
Adoption:    Accelerated
Bubble talk: Quieted

The hype cycle didn’t crash.

No Single Winner

GPT-4 remained strong
Claude 3.5 matched or exceeded in some areas
Open models closed the gap
Multiple viable options

Competition, not monopoly.

For Developers

What Worked

1. AI-assisted coding - Genuine productivity gains
2. Local models - Privacy-preserving options
3. RAG patterns - Standard for enterprise
4. Multimodal - Vision and voice maturing

What’s Still Hard

1. Reliability - Hallucinations persist
2. Evaluation - Hard to measure quality
3. Integration - Connecting to real systems
4. Security - Prompt injection unsolved

Looking to 2025

Predictions

1. Reasoning models mainstream - o1-like thinking becomes default
2. Agents start working - Limited but useful automation
3. Regulation arrives - EU AI Act enforcement begins
4. Local models parity - Open matches closed for many tasks

Questions

• Will the next big leap come from scaling or architecture?
• Will regulation slow or guide innovation?
• Will the job market adapt or resist?

Final Thoughts

2024 normalized AI. It’s no longer a startup pitch or a research paper—it’s in your operating system, your text editor, your email client.

The technology matured. The infrastructure emerged. The debates continued.

2025 will be about making it all actually work reliably.

2024: AI became infrastructure.