Where are we

ai

The pace is relentless. The breakthroughs are real. Here’s how to make sense of it all.

If you blinked in 2025, you missed something major. The AI landscape has shifted so dramatically this year that even those of us immersed in it daily struggle to keep pace. But here’s the thing: the speed isn’t just about new models dropping every week. It’s about a fundamental change in how AI works, who’s building it, and what it can actually do.

From Chatbots to Agents

2025 will be remembered as the year AI stopped just answering questions and started doing things. Agentic AI—systems that can plan, reason, execute multi-step tasks, and self-correct—has moved from research demos to production workflows. Microsoft and NVIDIA launched the “Agentic Launchpad” to fuel autonomous AI startups. Google added research agents to NotebookLM. Anthropic’s Claude now orchestrates complex tool chains with extended thinking. These aren’t incremental improvements—they’re a new paradigm.

The Reasoning Revolution

Reasoning defined this year. DeepSeek’s R1 proved China could hit frontier-level reasoning at a fraction of Silicon Valley’s cost, using a reinforcement-learning-first approach that challenged everything we assumed about training AI at scale. GPT-5.1 now runs 2-3x faster. Baidu’s ERNIE 5.0 claims to beat GPT-5 on visual understanding. World Labs released Marble—turning text prompts into explorable 3D worlds. Capabilities arriving each week would have seemed impossible two years ago.

Smaller Is the New Bigger

The industry pivoted hard toward efficient Small Language Models. Cost pressure, latency, and privacy demands forced a rethink of “bigger is better.” Models with 3B to 15B parameters became the workhorses—solid performance at under $0.0001 per request. The new question isn’t “how big?” but “what’s the smallest model that gets the job done?”

So How Do You Keep Up?

You don’t—not with everything. The key is selective depth over scattered breadth:

• Pick your lane. Follow 2-3 areas closely rather than skimming everything.
• Watch the builders, not just the announcements. What developers actually ship matters more than press releases.
• Embrace the plateau. 2025 showed us that making AI usable and dependable matters as much as making it smarter.

The real progress this year wasn’t just in the giants we built—it was in finally asking harder questions about what we’re building them for.