What AI Trends Will Dominate the Next Year? A Deep Dive into Physical AI, Agents, and RAG

Key Takeaways

Physical AI will become a major frontier, bringing AI into robots and edge devices.

Retrieval‑Augmented Generation (RAG) is moving from experimental to standardized deployments.

AI is shifting from a mere assistant to a collaborative creator across the entire software lifecycle.

AI‑driven DevOps practices are gaining strong attention.

Human‑Computer Interaction (HCI) must focus on real‑world user needs and privacy.

New protocols such as MCP and A2A will enhance interoperability between AI clients and backend systems.

AI/ML Trend Report Overview

The report provides a concise overview of emerging trends in artificial intelligence, machine learning, and data engineering, summarizing insights from several expert podcasts. It presents a twelve‑month outlook on technologies that are expected to move from the "innovator" stage to broader adoption.

1. AI and Machine Learning Trend Graph

The core of the report is an annual trend graph based on Geoffrey Moore's "Crossing the Chasm" framework, showing how technologies evolve from the "innovator" phase to "early adopters" and "early majority".

Since the release of ChatGPT in November 2022, generative AI and large language models (LLMs) have dominated the AI landscape, with major vendors continuously launching more powerful models.

2. Innovator Stage

2.1 AI Agents

AI agents are rapidly advancing. Notable releases include Anthropic's Claude Subagents and Amazon's Bedrock Agents, which enable complex workflow orchestration and contextual decision‑making.

OpenAI announced a general‑purpose ChatGPT Agent for spreadsheet and presentation automation.

Amazon open‑sourced the Strands Agents SDK for building custom agents.

NVIDIA introduced Visual AI Agents for video analysis.

Daniel Dominguez notes that AI assistants can now schedule meetings, update databases, and launch cloud resources without managing underlying infrastructure, dramatically shortening the path from experiment to production.

Anthony Alford warns that agents with file‑system access can execute destructive commands (e.g., rm -rf ), amplifying security risks when AI can access sensitive resources.

2.2 Multimodal LLMs

Modern language models now process text, images, audio, and video simultaneously, enabling deeper semantic understanding and cross‑modal reasoning.

2.3 Physical AI

Physical AI represents the embodiment of AI in robots and edge devices.

Google released Gemma 3n, a generative model optimized for mobile and laptop devices.

Microsoft introduced Mu, a lightweight small language model for Windows settings.

Google DeepMind unveiled Gemini Robotics On‑Device for on‑device multimodal reasoning.

NVIDIA launched a three‑layer Physical AI stack: DGX supercomputers (training), Omniverse & Cosmos (simulation), and Jetson AGX Thor (robotic inference).

Savannah Kunovsky emphasizes the need for trustworthy edge experiences, where data stays local and privacy is preserved.

2.4 MCP (Model‑Centric Protocol)

Anthropic introduced MCP in November 2024 as an open standard for unified data‑integration interfaces for LLMs, reducing fragmentation.

OpenAI, Microsoft, and Google have announced support for MCP, making it a key technology for AI‑agent interoperability.

Anthony Alford highlights MCP's role in enabling cross‑company model collaboration, while Daniel Dominguez points out its importance for scenarios such as Playwright MCP servers and Figma prototype integration.

2.5 Human‑Computer Interaction (HCI)

Agentic AI and Physical AI are reshaping HCI, pushing for more natural, context‑aware interactions.

Savannah Kunovsky (IDEO) stresses that effective AI‑driven interfaces should embed information directly into daily contexts—e.g., showing recipes while cooking—rather than requiring users to stop and consult a device.

3. Early Adopter Stage

3.1 Language Model Innovations

Key breakthroughs include visual language models (VLM), small language models (SLM), reasoning models, and state‑space models (SSM).

OpenAI's GPT‑5 introduces automatic model selection, simplifying user experience.

OpenAI Sora pioneers generative video modeling.

3.2 Retrieval‑Augmented Generation (RAG)

RAG has transitioned from experimental to standard configurations in enterprise development, enabling automatic synthesis of internal documents and knowledge bases.

Anthony Alford notes that virtually every organization with large documentation is now building its own RAG system.

Savannah Kunovsky adds that RAG streamlines design research by automatically aggregating context, allowing non‑technical teams to build RAG‑powered applications.

Automated Machine Learning (AutoML) is also entering early adoption, being incorporated into production pipelines.

4. Early Majority Stage

Technologies widely adopted by development teams include:

Vector databases

MLOps platforms

Synthetic data generation

5. Late Majority Stage

Mature technologies now form core enterprise architecture:

Lakehouse data architectures

Stream processing frameworks

Distributed computing systems such as Apache Storm

6. Conclusion

AI is evolving from a task executor to a trusted collaborative partner capable of tackling complex real‑world problems. The next year is expected to see continued growth in AI agents and AI‑driven development tools, the rise of truly useful AI applications that will underpin the next generation of the internet, and the emergence of video‑RAG, which will create new challenges in distinguishing real from AI‑generated content. Industry discussions may shift toward an "AI bubble" focused on structural adjustments rather than technological failure, and AI will become increasingly invisible and contextual, operating seamlessly in the background of everyday life.

Final Thoughts

Stay tuned for the next installment of this series for deeper dives into specific technologies and practical implementations.