Shifting Agentic Focus
Over the past 24 months the focus has been on different aspects of AI Agents.
Two things I find fascinating is how the focus is shifting collectively. It is also combined with the AI hype cycle I guess.
The second thing that stands out for me is how certain aspects of Agentic AI receives immense attention and hype, while other aspects are barely noticed by the masses.
Case in point, Agentic Commerce did not receive nearly as much coverage as MCP did. Somehow MCP captured the imagination of the masses and something like Agent to Agent protocol went largely unnoticed.
So there is this aspect of random focus on aspects of Agentic AI. These might be driven by influencer preferences or commercial incentives. Companies wanting to get their name on the proverbial map.
Another, and real aspect is that if implementation struggles. As AI Agents are being implemented, a number of studies are highlighting what impediments developers are facing.
And as challenges are faced, that becomes the problem which needs to be solved. This is why elements like intents are coming to the fore again. Intents was deemed dead and buried with traditional chatbots.
AI Evals also featured for quite a while, and now it is memory.
AI Agent Memory
AI Agent Memory Memory is the Cornerstone of True Intelligence AI Agents and Agentic Systems are rapidly evolving and…cobusgreyling.medium.com
So here is my perception of the shifting AI Agent focus over the last 24 months (roughly late 2023 to December 2025).
I tried to capture the major waves in the community, research, and industry hype. Here’s a refined timeline based on key developments, showing how the emphasis evolved…
Late 2023–early 2024
ReAct (Reason + Act) and basic tool-using single AI Agents
The ReAct paradigm (from the 2022 paper, but popularised in 2023 implementations) became the foundational pattern for agents…
A framework leveraging an LLM to reasons, calls tools, observes results and iterates.
Early frameworks like LangChain emphasised single agents with tool access (for example web search, calculators, APIs).
This shifted from pure chatbots to “agents that can act.”
https://cobusgreyling.medium.com/react-synergy-between-reasoning-acting-in-llms-36fc050ae8c
Mid-2024
Agentic workflows & Multi-Agent Systems
Focus moved to orchestrating AI Agents reliably.
Frameworks like LangGraph (stateful graphs for controlled workflows), OpenAI and others enabled multi-agent collaboration — agents delegating tasks, debating, or specialising.
For example, one agent for research, another for coding etc.
This addressed ReAct’s limitations in complex, long-running tasks.
AI Agents & Agentic AI
Understanding the Next Frontier in Artificial Intelligencecobusgreyling.medium.com
Throughout 2024–2025
Tools & Integration
Tool-calling matured, for example function calling in OpenAI/Anthropic models), but… the pain point was integrating with real-world systems.
What’s The Difference Between Tools & Functions?
What’s The Difference Between Tools & Functions? They’re related, but not exactly the same thing... In the context of…cobusgreyling.medium.com
This led to…
Early–mid 2025
MCP (Model Context Protocol)
Introduced by Anthropic in 2024 and gaining traction in 2025, MCP became a standard for connecting agents to external tools, data sources and apps (like a USB-C for AI).
It simplified secure, consistent integrations, enabling agents to access calendars, databases, or services without custom hacks.
Using LangChain With Model Context Protocol (MCP)
The Model Context Protocol (MCP) is an open-source protocol developed by Anthropic, focusing on safe and interpretable…cobusgreyling.medium.com
Mid–late 2025
Agentic Commerce
A big application-driven hype: AI Agents handling shopping autonomously (research, comparison, negotiation, purchase).
Players like Stripe, Shopify, PayPal, Visa, and McKinsey highlighted agentic commerce as a transformative shift, with protocols for payments and catalogs.
This made agents feel “real-world impactful” beyond internal tools.
The Four Protocols of Agentic Commerce
The Four Protocols of Agentic Commerce AI-Driven Transactions Across Ecosystems I generally don’t like writing content…cobusgreyling.medium.com
Late 2025 (current focus)
Memory and long-term autonomy
Yes, memory is a hot topic now — short-term (conversation history), long-term (persistent knowledge/skills across sessions), and reflection (learning from past actions).
Enhanced memory enables more reliable, personalised, and adaptive agents, especially in multi-session workflows.
AI Agent Memory
AI Agent Memory Memory is the Cornerstone of True Intelligence AI Agents and Agentic Systems are rapidly evolving and…cobusgreyling.medium.com
Chief Evangelist @ Kore.ai | I’m passionate about exploring the intersection of AI and language. Language Models, AI Agents, Agentic Apps, Dev Frameworks & Data-Driven Tools shaping tomorrow.
COBUS GREYLING
Where AI Meets Language | Language Models, AI Agents, Agentic Applications, Development Frameworks & Data-Centric…www.cobusgreyling.com
Super insightful timeline. The observation about MCP getting way more hype than agentic commerce is spot on. I've been building agents for a fintech client and the memroy layer is seriously the hardest part now, especially when agents need to remember past failures across sessions. The shift from just "can it call a tool" to "does it actually learn" feels like the real test.