Key takeaways

  • An AI agent is software that carries out tasks with some autonomy, such as coding transactions or chasing receipts, rather than only answering questions like a chatbot.
  • UK QuickBooks users have had five AI agents since 14 November 2025, including a VAT agent that compares P&L income with VAT return net sales.
  • Xero launched JAX, its agentic platform, at Xerocon London on 8 July 2026, with Smart Document Capture and Auto Bank Reconciliation available at launch.
  • The FRC published the first audit regulator guidance on generative and agentic AI on 30 March 2026, stating that the human auditor is always accountable.
  • The ICO, FCA, Ofcom and CMA say existing UK law still applies to agentic AI, and the organisation deploying an agent stays legally responsible.
  • Practices using agents should keep human approval checkpoints, audit logs and least-privilege access, and record AI use in engagement letters.

Agentic AI means software that carries out multi-step tasks for you, such as coding bank transactions, reconciling accounts or chasing a client for a missing receipt, rather than just answering questions. Xero, Intuit and others now ship agents inside UK accounting software. For a practice, the gain is time on routine work; the cost is a new set of controls, because the firm stays responsible for whatever an agent does.

What is an AI agent?

An AI agent is an AI system that can plan and take actions towards a goal, often across several steps and tools, with limited human input. A chatbot answers a question and stops. An agent might read a supplier invoice, match it to a bank line, post the entry and flag anything odd for review.

ICAEW Insights, writing on 28 April 2026, describes agents as systems that act autonomously and can automate both front-office and back-office tasks. The autonomy is the point, and also the risk.

Tool typeWhat it doesWho acts
Chatbot or assistantAnswers questions, drafts textYou
CopilotSuggests the next step inside an appYou, after a prompt
AgentPlans and completes tasksThe software, with approval checkpoints

Which AI agents can UK practices use now?

Several mainstream UK accounting platforms already include agents. These are the ones we could confirm from the providers’ own announcements, listed alphabetically.

Intuit QuickBooks. Intuit launched five agents for UK customers on 14 November 2025: Accounting, Customer, Finance, Project Management and a VAT agent in beta. The VAT agent compares profit and loss income with net sales on the VAT return and flags differences. Intuit says “we recommend you review all outputs carefully.” Its newer Agentic Books Close is a US pilot; see Intuit’s AI agents for QuickBooks.

Xero. Xero launched JAX, which it calls its agentic platform, at Xerocon London on 8 July 2026. Smart Document Capture and Auto Bank Reconciliation were live at launch. Document chasing, Payment Follow Ups, Bill Protection and Cash Flow Actions were announced, along with XeroForce, a natural-language builder for custom agents.

Connectors to general assistants. Xero’s UK site lists a free Claude connector that is read-only: Claude cannot create, edit, approve, reconcile or pay in Xero. A read-only connection is a lower-risk way to test AI on live client data. Xero’s Google Gemini app is not available in the UK; see Xero and Google Gemini.

What can agents do in a practice?

Agents are best suited to high-volume, rule-based work with a clear right answer. Typical tasks are:

  • reading receipts, bills and bank statements and creating draft transactions
  • suggesting bank reconciliation matches
  • flagging missing receipts and chasing clients for them
  • checking VAT return figures against the ledger before filing
  • drafting overdue invoice reminders for approval
  • summarising a client’s position before a meeting

Judgement calls still belong to people. Tax treatment of unusual items, accounting estimates, going concern and anything you sign should stay with a qualified reviewer.

What are the risks of agentic AI?

The main risk is that an agent acts on bad data, or with too much access, and nobody notices. The ICAEW’s summary of the Digital Regulation Cooperation Forum paper lists these:

RiskWhat it looks like in a practice
AccountabilityUnclear who is to blame when an agent posts wrong entries
Data protectionAgents pulling more client personal data than needed
Cyber securityBroad permissions widening the attack surface
Black box decisionsNo clear record of why an entry was made
Vendor lock-inWorkflows that only work on one platform

The four DRCF regulators (the ICO, FCA, Ofcom and CMA) say existing UK obligations still apply to agentic AI. The organisation that deploys an agent remains legally responsible, however autonomous it is.

What do the regulators say?

The FRC’s Generative and Agentic AI Guidance, published on 30 March 2026, is the first such guidance from any audit regulator. It is aimed at audit firms and is non-prescriptive. Its key line is that “the human auditor is always accountable”, and firms should fit AI use into their ISQM (UK) 1 quality management.

For personal data, the ICO’s guidance on AI and data protection still applies. The ICO says the guidance is under review following the Data (Use and Access) Act. It covers accountability, data protection impact assessments, transparency and data minimisation. Our guide to AI and data privacy for accountants sets out the checks.

Which applies to you?

  • If you are a sole practitioner: start with agents built into the software you already use, with every action approved by you.
  • If you run a firm with staff: set a written AI policy first; our AI policy template is a starting point.
  • If you carry out audits: read the FRC guidance and document how you rely on agent output under ISQM (UK) 1.
  • If you only want answers from client data: use a read-only connector rather than an agent that can post.

What advisers should do

  1. List every AI tool in use, including staff experiments, and which client data it can reach.
  2. Give agents the least access they need. Separate read-only from posting permissions.
  3. Keep a human approval step for anything that posts, pays or sends to a client.
  4. Turn on audit logs and sample agent work each month.
  5. Run a DPIA where an agent processes client personal data at scale.
  6. Update engagement letters to say how you use AI on client work.

For a wider plan, see our AI adoption guide for accountants.

Frequently asked questions

What is the difference between an AI agent and a chatbot?

A chatbot answers questions. An AI agent plans and carries out tasks, such as posting transactions or chasing documents, with limited human input.

Are AI agents available in UK accounting software?

Yes. UK QuickBooks has had five agents since 14 November 2025, and Xero launched its JAX agentic platform at Xerocon London in July 2026.

Who is responsible if an AI agent makes a mistake?

The firm that uses it. The FRC says the human auditor is always accountable, and UK regulators say the deploying organisation stays legally responsible.

Can AI agents file VAT or tax returns on their own?

Not in the tools we checked. Intuit’s UK VAT agent checks figures before filing, and the business remains responsible for the return it submits to HMRC.

Do I need to tell clients I use AI agents?

There is no specific AI disclosure rule, but UK GDPR transparency duties apply to client personal data. Many firms now state their AI use in engagement letters.

Sources

Rates, thresholds and rules in this guide were checked against HMRC and GOV.UK guidance on 6 November 2026. Tax rules change, so check the current position on GOV.UK or with a qualified adviser before acting. This guide is general information, not personal advice.