Six shifts we are watching
Updated October 2026. These are the shifts we expect to shape how people actually use AI over the next year.
Each one comes with something you can do now.
Agents are getting shared plumbing
MCP is now governed by the Agentic AI Foundation under the Linux Foundation, and its July 2026 specification made the protocol stateless and ready for large deployments.
Agent-to-agent communication has its own open protocol, A2A, also hosted at the Linux Foundation. Shared plumbing means a tool you connect once works across many assistants.
Now: favor tools that offer official MCP support.
Know-how is becoming portable
Agent Skills started at Anthropic and became an open standard.
The same SKILL.md folder now works across a growing list of agents. Your process is turning into an asset you own, independent of any one app.
Now: write your most repeated process as a skill.
Apps are moving inside the conversation
MCP Apps let tools return real interfaces, like forms, previews, and dashboards, right inside the chat.
Less switching between tabs, more getting things done in one place.
Now: when you evaluate a tool, ask whether it works inside the assistant you already use.
Small active models make local practical
Many new open models use a mixture-of-experts design that activates only a few billion parameters per token.
They run on well-equipped laptops at useful speeds, which makes private, offline AI a practical option.
Now: test a local model on one private workflow.
Voice and video are going real-time
Major labs keep shipping speech models that stream in real time across many languages, and video models keep getting longer, more controllable, and more consistent.
Conversation is becoming an interface, and generated footage is becoming a production tool rather than a novelty.
Now: try one voice-first workflow, like dictating a brief while you drive to a shoot.
Security is the new bottleneck
The more an agent can touch, the more a mistake costs.
Tool poisoning and over-permissioned connectors became mainstream concerns in 2026. Expect approval steps, scoped access, and audit trails to become standard features.
Now: review everything your assistants can access and remove what you do not use.
What does not change
Models will keep leapfrogging each other.
Pick the one shift that touches your work most and run one small experiment this month. Write down what happened. That note is worth more than any prediction, including ours.
Your context, your workflows, and your taste are what compound. Invest there and every new model makes you better instead of sending you back to the start.
Resources
Every link this breakdown leans on, in one place.