ChatGPT Agent Mode: The Complete UK Business Guide (2026)

What ChatGPT Agent Mode actually is

ChatGPT Agent Mode is the feature that lets ChatGPT complete multi-step tasks autonomously using its own virtual computer. Instead of replying to a prompt with text alone, the agent can open a browser, navigate to websites, click buttons, extract data, run code, build spreadsheets, and produce presentations.

Think of it as the difference between asking someone how to book a restaurant and asking them to actually book it. Standard ChatGPT gives you the instructions. Agent Mode picks up the phone.

The feature launched on 17 July 2025, consolidating what were previously two separate OpenAI products: Operator (browser automation) and Deep Research (multi-source analysis and report writing). Agent Mode brings them together with tool-use, code execution, and file handling in a single interface. It is available to Plus, Pro, Business, and Enterprise users.

How it differs from other OpenAI products:

Custom GPTs and the Assistants API remain valuable, but neither takes action independently. A Custom GPT can answer questions about your pricing policy; it cannot log into your CRM and update a customer record. Agent Mode can. For more on how AI engines now retrieve and cite business content, see our guide on how to get cited by ChatGPT, Gemini, and Perplexity.

Why UK businesses should care

Three reasons stand out.

1. Time savings on repetitive research tasks. Agent Mode excels at jobs that involve gathering information from multiple sources and synthesising it. Comparing supplier prices across ten websites, compiling a competitor pricing matrix, or researching planning permission requirements across local council portals — these tasks routinely consume 2-4 hours when done manually. Agent Mode completes them in 15-30 minutes.

2. Consistency in process-driven work. When you delegate a task to Agent Mode, it follows your instructions precisely every time. No variation, no forgotten steps, no inconsistent formatting. For businesses that produce regular reports or follow standard operating procedures, that consistency matters more than raw speed.

3. Accessibility for non-technical teams. You do not need to write code to use Agent Mode. The interface is conversational — you describe what you want in plain English and the agent figures out the steps. That makes it usable by solicitors, estate agents, accountants, and other professionals who would never touch a scripting language.

The real value emerges when you identify tasks that are important enough to care about but tedious enough to avoid. Agent Mode handles those well.

[Get a free growth audit to see where Agent Mode could save your team the most time →](/audit)

8 concrete UK use cases

Solicitors: case intake triage

A small high-street practice receives 15-20 enquiry emails per day. Each one needs to be read, categorised (family law, conveyancing, probate), assessed for urgency, and filed into the case management system.

Before: A paralegal spends 90 minutes daily reading and sorting, often missing urgent matters buried in a batch of routine queries.

After: Agent Mode reads each new enquiry, extracts matter type and client details, categorises it, flags urgent items, and drafts a response for a solicitor to approve.

Time saved: ~75 minutes per day, freeing the paralegal for billable work.

Estate agents: property-matching automation

An independent estate agent has 40 active buyers, each with specific criteria — budget, bedrooms, location, garden preference. When a new instruction comes in, someone must cross-reference buyer requirements manually.

Before: Sales negotiator spends 20-30 minutes per new instruction. Suitable matches get missed under time pressure.

After: Agent Mode compares each new listing against buyer profiles stored in a spreadsheet or CRM and generates a match list with personalised email drafts.

Time saved: 25 minutes per instruction, plus improved match accuracy.

Accountants: invoice categorisation

A small practice processes 200-300 client invoices monthly. Each one needs categorising into the correct expense type for VAT returns.

Before: Junior staff manually review, categorise, and enter data into Xero or QuickBooks. A single client's monthly batch takes 45 minutes.

After: Agent Mode receives uploaded invoice PDFs, extracts supplier, amount, and line items, and prepares entries for review in the accounting software.

Time saved: ~30 minutes per client per month.

E-commerce: return request handling

An online retailer receives 50-60 return requests weekly. Each one requires checking the order, verifying return eligibility, processing the refund or exchange, and confirming with the customer.

Before: Customer service staff handle each request individually — 8-10 minutes per return on admin.

After: Agent Mode logs into the e-commerce platform, retrieves order details, checks against returns policy, processes eligible refunds with human approval for edge cases, and drafts customer communications.

Time saved: 6-7 minutes per return, freeing staff for complex customer issues.

Service businesses: quote generation

A landscaping company receives 10-15 quote requests weekly via its website form. Each requires reviewing the request, estimating materials and labour, and producing a formal quote.

Before: The owner spends 25-40 minutes per quote, usually in the evenings after site work.

After: Agent Mode reads incoming requests, applies pricing from a reference spreadsheet, generates a quote document, and queues it for approval.

Time saved: 20-30 minutes per quote.

SaaS: customer onboarding sequences

A UK-based SaaS company onboards 20-30 new customers per month. Each needs a welcome email, setup confirmation, a personalised getting-started guide, and a 7-day follow-up check-in.

Before: Customer success staff compose each email manually and track follow-ups in a spreadsheet, occasionally missing the 7-day window during busy weeks.

After: Agent Mode drafts personalised onboarding emails, creates account-specific guides by pulling from internal help documentation, and schedules follow-ups.

Time saved: 15-20 minutes per customer, plus consistent timing.

Consultants: meeting prep and follow-up

A freelance strategy consultant runs 8-10 client meetings per week. Before each, they review recent email threads, uploaded client documents, and notes from previous sessions. After each, they send a summary and action items.

Before: 20-30 minutes prep plus 15 minutes follow-up per meeting. Total: 35-45 minutes each.

After: Agent Mode compiles a briefing document from recent email threads and uploaded client files before the call, then drafts a follow-up email with action items afterwards.

Time saved: ~30 minutes per meeting.

Marketing agencies: campaign reporting

A small agency runs campaigns across Google Ads, Meta, and LinkedIn for 12 clients. Monthly reporting requires logging into each platform, exporting data, and compiling a presentation.

Before: An account manager spends 3-4 hours per client per month on reporting alone.

After: Agent Mode logs into each platform, extracts key metrics, builds charts in a spreadsheet, and generates a slide deck with commentary.

Time saved: 2.5-3 hours per client per month.

How to actually set it up

You do not need technical skills to use Agent Mode. Here is the step-by-step process.

Step 1: Check your plan

Agent Mode is not available on the Free or Go tiers. You need Plus or higher. Message limits (per OpenAI's Help Center):

• Plus: 40 agent messages per month (see pricing below)
• Pro: 400 agent messages per month
• Business: 40 messages per user per month
• Enterprise: 40 messages per user per month, with optional flexible pricing at 30 credits per message

Only initial user-initiated requests count against the limit — clarification prompts and authentication steps do not. Plus users can purchase additional credits once they hit the cap; check OpenAI's pricing page for current rates.

Step 2: Activate Agent Mode

In the ChatGPT interface, open the Tools menu by clicking the '+' icon in the message composer. Select Agent mode from the dropdown. Alternatively, type /agent directly into the message box.

Step 3: Describe your task clearly

Be specific about three things:

• The goal — e.g. "compare pricing for X across these five websites"
• The output format — e.g. "create a table showing provider, price, and key features"
• Any constraints — e.g. "only include options under £500"

Example prompt:

"Log into our Xero account, download the last 3 months of bank transactions, categorise each transaction as Income, Expenses, or Transfers, and create a summary spreadsheet showing monthly totals by category."

Step 4: Approve logins and confirmations

When Agent Mode encounters a site requiring login, it pauses and prompts you to take control of the virtual browser directly. You enter credentials yourself; the agent never sees them. This is a deliberate security boundary.

For sensitive actions (sending emails, submitting forms, making purchases), Agent Mode will ask for explicit confirmation before proceeding.

Step 5: Review and refine

Once the task is complete, review the output. If something is wrong or incomplete, you can correct it conversationally — the agent retains context, so you do not need to repeat the full request.

Step 6: Schedule recurring tasks (optional)

For tasks you want to repeat, click the clock icon after completion to set a daily, weekly, or monthly schedule.

GDPR, PECR, and privacy considerations for UK users

If you are processing personal data through Agent Mode, UK GDPR applies. In January 2026, the UK Information Commissioner's Office (ICO) published an early-thinking report on agentic AI as part of its Tech Futures series. The ICO was explicit that this report is not formal guidance or a statement of regulatory expectations — it signals where the regulator's focus is heading, not where the rules already sit. The existing UK data protection framework (UK GDPR, PECR, the Data Protection Act 2018) still governs everything you do with these tools.

Six things to check before running Agent Mode on live customer data:

Connector availability. Some ChatGPT Agent connectors (Gmail, Google Drive, Slack, Notion, Dropbox) had regional restrictions for UK and EEA users when they first rolled out in mid-2025, citing data protection concerns. Availability has been expanding since, but varies by subscription tier and region — and Google has, at times, independently blocked agentic access to Gmail. Before committing to a workflow that depends on a specific connector, verify it is available on your plan at chatgpt.com/settings/Connectors. Agent Mode's virtual browser login is the reliable fallback and works for any site you can log into manually.

Data residency. ChatGPT Enterprise and API customers can store data at rest in the UK — OpenAI added this option on 24 October 2025. Plus and Pro subscribers cannot. If you handle personal data of UK residents, Enterprise is the safer choice.

Training opt-out. Business and Enterprise plans do not use your data for model training by default. Plus and Pro users can opt out, but the default is inclusion. For client work, this distinction matters.

Purpose limitation. Define specific purposes for each processing task. Agent Mode should not have unrestricted access to all company data — grant access only to the systems and documents necessary for the specific job.

Automated decision-making. UK GDPR Article 22 gives individuals rights around decisions made solely by automated means where those decisions significantly affect them. If Agent Mode is approving applications, triaging complaints, or screening candidates, you need human review in the loop.

Data minimisation. Only give Agent Mode the data it needs. If it is researching market prices, it does not need access to your customer database. If it is processing supplier invoices, it does not need access to HR records.

Practical steps:

1. Use Business or Enterprise plans for client work involving personal data.
2. Enable UK data residency where available.
3. Record Agent Mode tasks in your Record of Processing Activities.
4. Run a DPIA (Data Protection Impact Assessment) for anything high-risk.
5. Build human review checkpoints into decisions that affect individuals.

For a wider view of where AI fits into UK SME operations, see our article on AI for business growth in the UK in 2026.

Pricing reality

Here is what you will actually pay. Plus is billed in GBP for UK users (£20 inc VAT). Pro, Business, and Enterprise are billed in USD — UK users will see currency conversion on their card statement, plus applicable VAT.

A few clarifications:

• Go is a new budget tier that launched in January 2026. It is a cheaper entry point than Plus but does not include Agent Mode.
• Plus is the entry tier for Agent Mode access. Billed in GBP for UK users (£20 inc VAT).
• Pro is billed in USD at $200/month. UK users pay the USD amount plus any currency conversion fee their card applies, plus applicable VAT.
• Business starts at $25/user/month when billed annually ($30 month-to-month).

For most UK SMEs, the real decision is Plus vs Business. Plus is cheaper and fine for personal exploration. Business adds UK data residency support, excludes your data from training by default, and offers admin controls — which matters once you are running client work through it. The price delta is small relative to the compliance benefit.

When NOT to use Agent Mode

Agent Mode is not a universal solution. Three scenarios where it is the wrong tool:

1. High-stakes decisions without human oversight. Do not use Agent Mode for decisions that significantly affect individuals without review — loan approvals, hiring, medical triage, legal advice. The ICO explicitly flags automated decision-making in sensitive domains as requiring human involvement.

2. Real-time or latency-critical tasks. Agent Mode is not fast. Tasks typically take 5-30 minutes depending on complexity. If you need a response in seconds, use standard ChatGPT or a purpose-built automation tool.

3. Sites with aggressive bot protection. Agent Mode often struggles with sites protected by Cloudflare challenges, CAPTCHAs, or other anti-bot measures. If your workflow requires navigating heavily-protected sites, expect failures.

A simple rule: if the task requires judgement, keep the human in the loop. Agent Mode is best for execution, not decision-making.

Alternatives worth knowing

Claude with Computer Use

Anthropic's Claude offers a similar capability called Computer Use, exposed through their API. It provides desktop control rather than browser-only access — it can operate native applications, not just websites. The trade-off: it requires developer setup. You need to run a Docker container, manage API calls, and build your own interface. For technical teams building custom automation pipelines, Claude's approach offers more control. For non-technical users, ChatGPT Agent Mode is far more accessible.

Google Gemini agents

Google's Gemini ecosystem includes agent capabilities through the Agent Development Kit (ADK). It integrates natively with Google Workspace (Gmail, Docs, Sheets) and Google Cloud. For organisations already committed to Google's infrastructure, that integration depth is valuable. Gemini's agent capabilities are less mature than OpenAI's and generally require more development work to deploy effectively.

OpenAI Agent Builder (AgentKit)

OpenAI's AgentKit, released in late 2025, is a developer toolkit for building custom AI agents with more control than Agent Mode offers. It includes visual builders, pre-built components, and deployment options. This is aimed at teams building AI agents as products, not at businesses using them internally. If you are considering working with an AI automation agency, ask whether they use AgentKit or a similar framework — the answer tells you how technical their capability really is.

Next steps for UK SMEs

Agent Mode represents a genuine shift in what AI can do for a business. It moves from advice to execution, from generating text to completing work. For UK SMEs with repetitive, research-heavy, or process-driven tasks, the time savings are measurable and real.

Start narrow. Pick one task that currently consumes 30+ minutes per week. Run it through Agent Mode. Measure the time saved and the output quality. Then decide whether to expand.

If you want a clearer picture of how AI fits into your specific operations, Ampliflow runs a free growth audit that identifies automation opportunities and prioritises them by impact.

[Get your free growth audit →](/audit)

For a broader view of the tools worth considering in 2026, see our AI tools stack for UK SMEs guide.