04/09/2026
𝐓𝐡𝐞 𝐧𝐞𝐱𝐭 𝐟𝐫𝐨𝐧𝐭𝐢𝐞𝐫 𝐨𝐟 𝐀𝐈 𝐢𝐬 𝐧𝐨𝐭 𝐚𝐧𝐬𝐰𝐞𝐫𝐢𝐧𝐠 𝐛𝐞𝐭𝐭𝐞𝐫. 𝐈𝐭 𝐢𝐬 𝐚𝐜𝐭𝐢𝐧𝐠 𝐛𝐞𝐭𝐭𝐞𝐫.
OpenAI has just introduced GPT-6 Astra, and beyond the extraordinary benchmark numbers, there is one aspect that I find particularly significant: the growing ability to delegate complex work to AI with greater confidence in its judgement and, crucially, in its ability to remain within the intended scope.
This is where the evolution becomes particularly interesting from an Agentic AI perspective.
During my studies in Artificial Intelligence at MIT and, more recently, through the Certificate Program in Agentic AI at Johns Hopkins University, I have increasingly focused on a fundamental distinction:
𝐚𝐧 𝐀𝐈 𝐦𝐨𝐝𝐞𝐥 𝐠𝐞𝐧𝐞𝐫𝐚𝐭𝐞𝐬 𝐚𝐧 𝐨𝐮𝐭𝐩𝐮𝐭; 𝐚𝐧 𝐀𝐈 𝐚𝐠𝐞𝐧𝐭 𝐩𝐮𝐫𝐬𝐮𝐞𝐬 𝐚𝐧 𝐨𝐛𝐣𝐞𝐜𝐭𝐢𝐯𝐞.
And pursuing an objective changes everything.
An agent must be able to understand context, reason, plan, select and use tools, interact with digital environments, evaluate intermediate results, recover from errors and decide when human intervention is required.
If models such as Astra become significantly better at computer use, browsing and complex multi-step professional workflows, the potential evolution of agentic systems is considerable.
Imagine an AI agent that does not simply suggest what a manager should do, but, within clearly defined permissions, can research information, interact with applications, analyse documents and data, execute workflows, verify outcomes and return to the manager when judgement or authorisation is required.
But greater autonomy creates an equally important question:
How much autonomy are we prepared to delegate?
This is where innovation meets governance.
Read the entire article on LinkedIn https://lnkd.in/p/dNZTmn5i
