Over the past few years, artificial intelligence has mostly played a passive role in consumer life. It helped me to summarise the document or write an email. These features, while useful, kept AI in the realm of text prediction and never far from the origins of humble chatbots. It is changing rapidly.
The latest iterations of AI don't wait to be prompted. They are learning to follow up, complete tasks, and even work independently. Whether it's Meta's new AI studio, new classes of AI are emerging through the recently launched ChatGPT agent.
From persona to pursuit
When Meta launched AI Studios, they pitched the product as a way for creators and businesses to build custom chatbots. These AI characters can simulate chefs, stylists, influencers, and even fictional characters. But beyond the fanfare of personality,
Deeper Ambition: An Endless Conversation.
According to an internal document first reported by Business Insider, Meta is training these bots to send follow-up messages. Under the project codename “Omni”, these bots can replicate conversations with users if certain criteria are met.
Users must be sending messages to the bot at least five times in two weeks, with aggressive outreach being restricted within 14 days of initial engagement. If the user ignores the first message, the bot will retreat.
Meta makes this a feature rather than a bug, enhancing interaction. Once the conversation is started, Meta AI Studio's AIS can also follow up with you to share ideas and ask additional questions.
But the problem is that Meta is no longer happy with the reactive tool, but they want these agents to push more information. For creators, the appeal is clear. It's something that keeps fans involved without directly involved. But it also places the platform straight in the middle of one-sided dynamics. There, in an automated intimacy and become useful, or deeply creepy.
ChatGpt gets body
Meta trains bots to advance the conversation, but Openai has released something much more ambitious. Recently, the company introduced a new class of tools called “ChatGpt Agents.” Unlike its predecessors, these agents can perform tasks from start to finish using what Openai calls “virtual computers.” Agents can browse websites, fill out forms, update calendars, write emails, send plan trips, and execute transactions.
It combines “operators” who can control the website with previous experimental tools from OpenAI, such as “Deep Research,” which conducted advanced comparative analysis. As Openai puts it in the release memo, “ChatGpt is now thinking and acting.” Agents are designed to narrate the steps and seek permission before taking any important actions, and may be suspended or redirected at any time. High-risk behaviors such as financial transfers and health decisions remain limited or supervised.
Boundary autonomy
Both Meta and Openai emphasize user control. Meta bots cannot be followed up indefinitely. Openai agents pause for confirmation before performing the task. These safeguards aim to reassure users in times of uncertainty. Once the AI waits for input, it starts now. Once you've answered, this is all you need to do. This is a functional change with philosophical weight. AI is no longer a tool, it's a proxy at work.
This is not necessarily the cause of the alarm. For many users, especially freelancers, small businesses and overload experts, Agent AI can represent the actual release, which is the end of a digital drudger. For platforms, it means deeper engagement and potentially new monetization pathways. For example, Meta expects generative AI in 2025 to drive “billions” of revenue in 2025, according to internal forecasts.
As AI agents begin to rise, the issues become less about capabilities and more about governance. Who controls the agent's behavior? Who is responsible for any errors? What happens when an agent imitates you very well and others can't convey the difference?
OpenAI's model context protocol is a technical framework for enabling interoperability between agents and platforms, and is a step towards standardization.
But that's not the same as accountability. Meta's enthusiastic bots are now strictly sandboxed, but their existence suggests the platform's owned personality could somehow provide the platform's goals. The rise of AI agents is not just a footnote to the AI story, but now it's the next chapter. And, like the delegation of power, it brings a paradox. The more tasks you have to offload to these agents, the higher the risk you will forget what it means to do yourself.
