2025 in AI: The Year Models Became Agents

2025 will be remembered as the year AI went from just “answering questions” to systems that execute tasks. Let’s take a journey through the moments that defined the year.

The Models That Marked 2025

November-December: The Avalanche

In just 4 weeks, four giants launched their most powerful models:

Date	Model	Company	Main Innovation
Nov 17	Grok 4.1	xAI	Humor and reasoning
Nov 18	Gemini 3	Google	Deep Think
Nov 24	Claude Opus 4.5	Anthropic	Autonomous agents
Dec 11	GPT-5.2	OpenAI	Response to competition

OpenAI’s “Code Red”

Sam Altman issued an internal “code red” memo in December after:

• Gemini 3 led benchmarks
• Claude gained enterprise market share in code
• ChatGPT lost 6% of traffic

This moment revealed that OpenAI is no longer the undisputed leader.

The Code Agent War

2025 was the year AI went from suggesting code to writing it autonomously:

Tool	2025 Milestone
Claude Code	$1B+ in annual revenue
GitHub Copilot	2M+ subscribers
Cursor	Series B round at $1B+ valuation
Codex	Specialized GPT-5.2-Codex

The Birth of Checkpoints

Anthropic introduced checkpoints in Claude Code 2.0, allowing:

• Save state before each change
• Revert with Esc+Esc
• Experiment without fear

This changed the trust dynamic between developer and agent.

The Agentic AI Foundation

Perhaps the most significant moment: competitors joined to create open standards.

Founders

• OpenAI: Agents.md (specification for describing codebases)
• Anthropic: MCP Protocol (communication between agents)
• Block: Goose framework (open source agent)

The “USB-C of AI”

Anthropic’s MCP (Model Context Protocol) is being described as the potential universal standard for connecting agents with the digital world.

Before:  Each agent has its own integration
Now:     One standard protocol for all

Acquisitions That Shook the Market

Nvidia + Groq ($20B)

Nvidia’s largest acquisition consolidated its dominance in:

• Training: GPUs
• Inference: Groq’s LPUs

Anthropic + Bun

Anthropic acquired the JavaScript runtime that generates $1B/year, strengthening its development stack.

Disney + OpenAI Sora

Disney signed an agreement to allow video generation with 200+ characters from its franchises.

IPOs on the Horizon

Anthropic hired Wilson Sonsini to prepare for a 2026 IPO:

• Potential valuation: $300B+
• Completing internal checklists
• Possible new round before IPO

Reality vs. Hype

What They Promised

“2025 will see the first AI agents join the workforce” - Sam Altman

What Happened

Upwork study (November 2025):

• Agents from OpenAI, Google, and Anthropic
• Failed to complete simple work tasks
• Far from being “digital employees”

The Hype Correction

MIT Technology Review declared 2025 as the year of the “Great AI Hype Correction”:

• Expectations adjusted to reality
• Focus shifted from AGI to practical applications
• Companies prioritized ROI over innovation

The 14 Terms That Defined 2025

According to MIT Technology Review:

1. Agentic AI - AI that takes actions
2. MCP - Model Context Protocol
3. Checkpoints - Saved states to revert
4. Inference at the edge - AI on devices
5. Multimodal - Text, image, video, audio
6. Context window - How much the model remembers
7. RAG - Retrieval-Augmented Generation
8. Fine-tuning - Model customization
9. Prompt engineering - (Started to die)
10. AI agents - Autonomous systems
11. LPU - Language Processing Unit
12. Synthetic data - AI-generated data
13. AI safety - Model security
14. Model collapse - Degradation from training on AI

What We Learned

For Enterprises

1. There’s no single winner: Multi-model is the strategy
2. Agents aren’t ready: But they’ll arrive soon
3. Governance is critical: Before it’s too late
4. Talent remains scarce: Invest in your team

For Developers

1. Learn to work with agents: Not against them
2. Understand the limits: Don’t delegate what you can’t verify
3. Checkpoints are your friend: Experiment with a safety net
4. Code still matters: Agents need context

Predictions for 2026

Area	Prediction
Agents	40% of enterprise apps will have them
Models	GPT-6, Claude 5, Gemini 4
Regulation	EU will fully implement AI Act
Code	Agents will write 50%+ of new code
IPOs	Anthropic, possibly OpenAI

Conclusion

2025 was the year when:

• Competition intensified
• Agents started working
• Hype was (partially) corrected
• Standards began to emerge

2026 promises to be the year all of this goes to production.

---

Want to prepare your company for the AI agent era? Let’s talk about technology strategy and roadmap.