A promising tool, but still far from “workforce on demand”

Launched in mid-July 2025, Chatgpt Agent marks OpenAi’s desire to take a step towards autonomous assistants capable of navigating the web, performing scripts and performing complex end-to-end tasks. Moreover, Presented as a “workforce on demand”. Consequently, the tool arouses as much enthusiasm as frustration, if we believe the feedback from the first users.

Real capacities but contrasting performance – Promising tool, but still far

On paper. Therefore, Chatgpt Agent impresses by its ability to chain the steps of a mission: collection of online information, organization and synthesis, generation of documents (reports, presentations, calculation sheets), the tool combines Operator and Deepresearch features, already available on the platform for several months. Therefore, For structured. However, unambiguous tasks, several testers bring back a significant time saving, the agent managing to reduce by half, or even more, the time required for certain research and preparations of deliverables.
However, the limits appear quickly promising tool, but still far as soon as the tasks become more complex. Consequently, Many returns on Reddit. Meanwhile, LinkedIn underline notable slowness, repetitive action loops and a high failure rate on certain web interactions.

Sandbox: safety or shackles?

The functioning of chatgpt agent is based on a sandbox: a virtual environment hosted by Openai. However, incorporating a browser and a virtual office. This approach aims to limit security risks. maintain control over the agent’s actions, but it greatly slows down its capacities.

Many sites block access or detect the agent like a bot, preventing reservations, purchases or complex interactions.

The agent fails regularly in front of dynamic forms, very interactive sites, protected by Captcha or services such as Cloudflare…. making it unusable to fill most of the most attractive promises on paper.

This protective approach makes the agent usable for research and syntheses, but not reliable for advanced or critical web actions.

An attempt promising tool, but still far to catch up

Faced with Chatgpt Agent, manus, developed by Future AG, offers a more daring approach. Using multi-agent architecture (planning. execution, validation) and a more complete sandbox environment (advanced browser, terminal, multimodal generation), manus is perceived by some as more efficient on complex tasks and autonomous workflows.

HoweverManus remains inexpensive (limited access, high costs in credits, limited availability), which limits its adoption. User feedback reports frequent bugs and significant credits consumption in the event of failure, generating frustration. Its autonomy implies less control in real time. which can lead to drifts when the agent goes in an unexpected direction.

To agents integrated into the browser?

Another way could raise part of the limited: deporting the agent to the user’s browser. via an extension or a local module.

Such a model would offer:

• Faster execution by freeing itself from virtualization.
• Direct integration with local tools and data (CRM, messaging, promising tool, but still far documents).
• Better compatibility with modern sites, since it would behave like a real human browser.

But this choice would make two major challenges:

• Security: an agent with access to cookies. accounts and local data would become a privileged target for abuse.
• Marketing positioning: A local agent would look more like a personal automation tool than an “outsourced workforce”. changing the commercial promise of Openai.

A product still in running

As it stands. Chatgpt Agent remains an innovative but immature tool: useful for accelerating certain well -framed tasks, limited for more ambitious missions. The vision of a “Main – IA” remains more a marketing promise than a technical reality.

The evolution of the AI agent market could go through a hybrid compromise: agents capable of working in a secure environment while delegating certain interactions to the local browser. with increased control and keys.

In the meantime. users must promising tool, but still far consider Chatgpt Agent as an intelligent assistant in the test phase, not as a real replacement for human collaborators.

What future for the web in front of agents?

The Internet has always enabled software and computer systems to communicate with each other. Most websites implement at least one or more APIs, communication interfaces between software. Today, agents aim to transform the emerged part of the web, usable by humans, into software interfaces into natural language.

The question is not so much that of technical capacities as of consent. not of publishers to make their content accessible to software. The theoretical promises of agents are based on a world where agents have access to everything. One of the first obvious consequences would be the acceleration of the collapse, already started, of the advertising model.

This can also ask questions about changes. particularly in the context of marketplaces: how to consider that promising tool, but still far the good deals of advertising sites intended for individuals, such as Le Bon Coin or Vinted, are not completely pre -emptied by a few actors?

What will happen to competition and diversity of the offer if we all use the same price comparator? The publishers are taken in hold between a bypass of their use policy, and a progressive invisibility.