Introduction

Emma Farquharson, Ph.D., co-founder of SharedIntel AI, welcomes attendees to the AI at Work session focused on Chat GPT’s new agent mode. She describes it as one of ChatGPT and OpenAI’s most game-changing upgrades, recently made available to Plus users. While traditional ChatGPT use focuses on brainstorming, editing, and answering questions, agent mode takes functionality to a different level with hands-off execution capability.

Setting the Stage

Emma explains that one of the biggest shifts in 2025 is the collapse of traditional marketing and sales boundaries. Due to AI-powered search platforms like Perplexity, Gemini, and Search GPT, buyer discovery happens before anyone fills out forms or takes sales calls. This means marketing has effectively become sales—content is the pitch. Organizations that can automate both brand building and lead generation gain competitive advantage over those operating in separate silos.

Overview of Agent Mode

Agent mode transforms ChatGPT from a chatbot into a “do-it-for-you” digital assistant with initiative. It operates on OpenAI’s unified agentic system, which includes:

• A visual browser that can click, scroll, and interact with websites

• A text browser for data scraping and page analysis

• A command-line terminal for script execution and file management

• Access to tools like Gmail, Google Drive, Box, GitHub, Canva, and others

• A file system to create, edit, and save documents in real time

When given a task, agent mode isn’t just generating text—it’s opening websites, navigating buttons, saving files, and connecting workflows like a digital employee.

Availability and Limitations

Agent mode launched in July 2025 and is available to:

• Pro users: 400 agent mode messages per month

• Plus and Team users: 40 agent mode messages per month

Emma notes that 40 messages can be depleted quickly, with OpenAI reporting nearly 70% of Plus users running out by day 20. Enterprise and education accounts are coming soon. Access is still limited in some regions like the European Economic Area.

Users can access agent mode through the tools dropdown or by typing the slash command “/agent”. Once activated, there’s a visual indication of being in an agentic session.

Prompting Style Differences

The most important mindset shift: agent mode is about commanding, not chatting. Traditional ChatGPT works best with gradual exploration and iterative responses, but agent mode requires clarity and structure upfront.

For example, instead of a casual prompt, users should provide structured instructions like: “Research Apple, Samsung, and Google’s 2025 phone models, extract product names, pricing and key features, create a comparison table in Excel, and a slide deck summarizing position differences.”

Testing showed structured prompts had an 85% success rate compared to 40% for casual prompts. Users should batch tasks, include specifics about file formats and goals, and be precise when using tools like Canva.

Limitations of Agent Mode

Speed can be a bottleneck with some multi-step tasks taking longer than doing them manually. Agent mode can stall on cookie pop-ups, CAPTCHAs, or login screens requiring two-factor authentication. Most importantly, ChatGPT’s memory is completely disabled in agent mode—there’s no context carried over between sessions, so users must provide instructions every time.

For security, agent mode asks permission before taking major actions like submitting forms or sending emails. It’s not recommended for high-stakes use such as financial transactions, sensitive client data, or mission-critical workflows yet. It’s well-suited for content creation, research, lead list generation, and simple automations.

Future Developments

OpenAI is working to improve agent mode with better reasoning, transparency, more connector integrations, and eventually reintroducing memory with stronger privacy safeguards. Emma suggests now is an ideal time to start experimenting with the tool.

Practical Demonstrations

Emma demonstrates two real-world agent mode applications:

1. Creating a prospecting list for print industry professionals attending a trade show

2. Building and publishing a marketing campaign, including Canva graphics and LinkedIn scheduling

Demo 1: Lead Generation

Emma walks through using agent mode to generate a qualified list of 100 potential leads for her company, SharedIntel AI, focusing on decision-makers who will be at print industry conferences in Orlando. The process includes:

• Using perplexity to research high-impact use cases for agent mode in lead generation

• Developing a detailed prompt describing her ideal customer profile

• Setting up deliverables including a master sheet with prospect information and an outreach pack with connection requests and follow-up sequences

• Using agent mode to research and compile the information

The result is a comprehensive spreadsheet with 100 leads complete with company details, readiness scores, campaign triggers, and specific information to help with outreach. Agent mode accomplished this in 10 minutes, searching through 196 sources.

Demo 2: Marketing Campaign Creation

Emma demonstrates using agent mode to create a 30-day social media content plan across LinkedIn, Twitter, and Instagram. The demonstration includes:

• Creating a detailed prompt with brand information, target audience, content themes, and posting schedule

• Using agent mode to research trending hashtags and create content aligned with her brand voice

• Generating visual assets for each platform

• Attempting to use Canva integration to create and schedule posts (with mixed success)

• Successfully scheduling content to LinkedIn

Key Takeaways

Emma emphasizes that agent mode requires detailed instructions to work effectively. Users can either put effort into creating comprehensive prompts upfront or spend time guiding the agent step-by-step when it encounters difficulties.

She recommends experimenting with agent mode now to build workflows, so when additional features are released, users will be positioned to take full advantage of the technology.

Emma concludes by noting that with GPT-5 on the horizon, the capabilities of agent mode will continue to improve, giving early adopters a competitive advantage.