Скачать или смотреть STOP calling this Agentic AI

𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐀𝐈 𝐢𝐬 𝐧𝐨𝐭 𝐧𝐞𝐜𝐞𝐬𝐬𝐚𝐫𝐢𝐥𝐲 𝐆𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐯𝐞 𝐀𝐈.
And if we keep treating it like it is, we’re going to design the wrong systems.

For some reason, “agentic” has been used as shorthand for 𝑳𝑳𝑴𝒔 𝒕𝒉𝒂𝒕 𝒕𝒂𝒍𝒌 𝒃𝒂𝒄𝒌.
But that’s not what makes a system agentic.

𝘎𝘦𝘯𝘦𝘳𝘢𝘵𝘪𝘷𝘦 𝘈𝘐 𝘤𝘳𝘦𝘢𝘵𝘦𝘴 𝘤𝘰𝘯𝘵𝘦𝘯𝘵.
𝘈𝘨𝘦𝘯𝘵𝘪𝘤 𝘈𝘐 𝘱𝘶𝘳𝘴𝘶𝘦𝘴 𝘨𝘰𝘢𝘭𝘴.

An agent can plan, decide, act, and adapt over time. And guess what, sometimes without generating ANYTHING.

The generative bit is just a piece of the puzzle inside a larger loop. The agentic part is the orchestration: when to act, what to call, who to involve, and when to stop.

Before all this generative AI boom, reinforcement learning was already built around this idea: an agent observes an environment, takes actions, receives feedback, and learns a policy over time. Who doesn't remember hide and seek from OpenAI? I don't know why it seems like the whole world has forgotten about it.

This is also where Physical AI makes the distinction impossible to ignore. A hybrid approach in where generative AI is not sufficient in most scenarios.

So when we design agentic systems today, we need to stop thinking about an LLM (or SLM) as the starting point. We need to take step back an ask ourselves:

• What is the goal?
• What environment does the agent operate in?
• What tools, data, and controls does it need?
• Where does autonomy make sense and where does a human stay in the loop?

𝐎𝐧𝐥𝐲 𝐭𝐡𝐞𝐧 𝐝𝐨𝐞𝐬 𝐢𝐭 𝐛𝐞𝐜𝐨𝐦𝐞 𝐜𝐥𝐞𝐚𝐫 𝐰𝐡𝐞𝐭𝐡𝐞𝐫 𝐠𝐞𝐧𝐞𝐫𝐚𝐭𝐢𝐯𝐞 𝐦𝐨𝐝𝐞𝐥𝐬 𝐚𝐝𝐝 𝐯𝐚𝐥𝐮𝐞 𝐚𝐭 𝐚𝐥𝐥.

My personal take on this, is that the next thing in AI isn’t generative AI nor agents. It’s hybrid systems that can think, decide, act, and adapt across digital and physical worlds.

Follow me:
Instagram:   / theaiframepodcast  
  / arojomaths  
LinkedIn:   / ana-rojo-echeburua  
Website: https://www.anarojoecheburua.com
Spotify: https://open.spotify.com/show/5amoMjT...