{"id":6837,"date":"2025-07-19T21:32:13","date_gmt":"2025-07-19T21:32:13","guid":{"rendered":"https:\/\/www.newsbeep.com\/au\/6837\/"},"modified":"2025-07-19T21:32:13","modified_gmt":"2025-07-19T21:32:13","slug":"what-openais-chatgpt-agent-actually-can-and-cant-do","status":"publish","type":"post","link":"https:\/\/www.newsbeep.com\/au\/6837\/","title":{"rendered":"What OpenAI’s ChatGPT Agent actually can (and can’t) do"},"content":{"rendered":"

OpenAI is finally getting in on the AI agent game. The generative AI startup has launched ChatGPT Agent, which allows its chatbot to perform multi-step tasks on behalf of users. This includes planning meetings, generating presentations, and using tools like a code terminal or web browser.<\/p>\n

\n\t\tRelated Article Block Placeholder<\/p>\n

\t\t\tArticle ID: 314747\n\t<\/p>\n

\t\t\t\t\t\t
\n\t\t\t\t\"AI\t\t\t<\/a><\/p>\n

The new tool is currently only available to paid users. Pro subscribers can use up to 400 agent prompts per month, while Plus and Team users are limited to 40. Access for Enterprise and Education users is expected in the coming weeks.<\/p>\n

It follows a slew of other major tech players such as Google<\/a>, Salesforce<\/a> and HubSpot<\/a> investing heavily in agentic AI over the past 12 months.<\/p>\n

It also comes right off the back of OpenAI announcing its own browser just last week<\/a>, hot on the heels of Perplexity AI.<\/p>\n

How OpenAI\u2019s ChatGPT Agent works<\/p>\n

According to OpenAI<\/a>, ChatGPT Agent \u201cbrings together three strengths of earlier breakthroughs: Operator\u2019s\u2060 ability to interact with websites, deep research\u2019s\u2060 skill in synthesising information, and ChatGPT\u2019s intelligence and conversational fluency\u201d.<\/p>\n

Users can prompt ChatGPT to complete instructions such as \u201clook at my calendar and brief me on upcoming client meetings based on recent news,\u201d or \u201canalyse three competitors and create a slide deck\u201d.\u00a0<\/p>\n

The agent operates using its own virtual computer and can switch between tools, including a text browser, a visual browser, a code terminal, and third-party app connectors such as Gmail and GitHub.<\/p>\n

OpenAI says this combination enables the model to \u201cfluidly shift between reasoning and action to handle complex workflows from start to finish\u201d.<\/p>\n

ChatGPT Agent is designed for collaboration, OpenAI says. Users can interrupt, take control of the browser, or redirect tasks partway through, and ChatGPT will adjust accordingly. <\/p>\n

\n\t\tRelated Article Block Placeholder<\/p>\n

\t\t\tArticle ID: 319698\n\t<\/p>\n

\t\t\t\t\t\t
\n\t\t\t\t\"Neural\t\t\t<\/a><\/p>\n

\u201cIt will pick up where it left off, now with the new information, but without losing previous progress,\u201d the company said in a blog post.<\/a><\/p>\n

What ChatGPT Agent can\u2019t do yet<\/p>\n

However, the tool is not built for real-time responsiveness. In a demonstration, ChatGPT Agent took around 25 minutes to generate a slide deck based on company earnings.\u00a0<\/p>\n

\n\t\tRelated Article Block Placeholder<\/p>\n

\t\t\tArticle ID: 304089\n\t<\/p>\n

\t\t\t\t\t\t
\n\t\t\t\t\"Neural\t\t\t<\/a><\/p>\n

Product lead Yash Kumar said that for many jobs, users can \u201ckick something off in the background and then come back to it\u201d. Research lead Isa Fulford added, \u201cEven if it takes 15 minutes, half an hour, it\u2019s quite a big speed-up compared to how long it would take you to do it\u201d.<\/p>\n

Some features, including memory, have been disabled at launch due to security concerns. \u201cIt\u2019s not that we don\u2019t think it\u2019s safe,\u201d Kumar told The Verge, \u201cwe\u2019re just taking an extra precaution\u201d.<\/p>\n

ChatGPT Agent sets new AI benchmark scores but also raises safety questions<\/p>\n

OpenAI says the model behind ChatGPT Agent significantly outperforms previous versions of its technology across a range of benchmark tests.<\/p>\n

But the rollout comes with heightened risk. OpenAI has classified the model as having \u201cHigh Biological and Chemical capabilities\u201d under its internal Preparedness Framework \u2014 a designation used when a model could potentially \u201camplify existing pathways to severe harm\u201d.<\/p>\n

\u201cWhile we don\u2019t have definitive evidence that the model could meaningfully help a novice create severe biological harm \u2014 our threshold for High capability \u2014 we are exercising caution and implementing the needed safeguards now,\u201d OpenAI said.<\/p>\n

The company says it has implemented new mitigations to reduce the risk of real-world misuse. That includes explicit user confirmation before taking irreversible actions, and \u201cWatch Mode,\u201d which requires users to supervise higher-risk tasks such as sending emails or interacting with financial websites.<\/p>\n

OpenAI has also introduced protections against prompt injection, which it describes as malicious instructions hidden in content the agent might encounter online.\u00a0<\/p>\n

\u201cBecause ChatGPT agent can take direct actions, successful attacks can have greater impact and pose higher risks,\u201d the company wrote.<\/p>\n

Still, according to OpenAI, today\u2019s release is \u201cjust the beginning\u201d. The company says it plans to improve the agent\u2019s capabilities and polish them over time. <\/p>\n

\u201cWe expect continued improvements to ChatGPT Agent\u2019s efficiency, depth, and versatility over time, including more seamless interactions,\u201d OpenAI said.<\/p>\n","protected":false},"excerpt":{"rendered":"OpenAI is finally getting in on the AI agent game. The generative AI startup has launched ChatGPT Agent,…\n","protected":false},"author":2,"featured_media":6838,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[20],"tags":[256,8667,254,255,64,63,5004,113,8668,5044,8669,8670,105],"class_list":{"0":"post-6837","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-artificial-intelligence","8":"tag-ai","9":"tag-ai-agent","10":"tag-artificial-intelligence","11":"tag-artificialintelligence","12":"tag-au","13":"tag-australia","14":"tag-chatgpt","15":"tag-google","16":"tag-hubspot","17":"tag-openai","18":"tag-perplexity-ai","19":"tag-salesfoce","20":"tag-technology"},"_links":{"self":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/6837","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/comments?post=6837"}],"version-history":[{"count":0,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/posts\/6837\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media\/6838"}],"wp:attachment":[{"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/media?parent=6837"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/categories?post=6837"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.newsbeep.com\/au\/wp-json\/wp\/v2\/tags?post=6837"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}