XeroForce review: Xero’s AI agent builder, and who should wait

Status, as at September 2026: XeroForce is in early access, with general availability due later in 2026, according to Xero’s Xerocon Denver post. Xero has not published pricing or a list of regions, and the pre-built month-end agent is listed as coming soon.

Xero introduced XeroForce in May 2026 as a no-code agent builder that lets accountants, bookkeepers and small business owners describe a workflow in plain English and have it run automatically across Xero and the apps connected to it. It started as an invite-only alpha, moved to early access by July, and is due for broader release later this year.

The announcement leans heavily on a phrase Xero is now using everywhere: a move “from a system of record to a system of action.” That language is doing a lot of work, and it is worth examining what it actually means for the firms being asked to adopt this.

XeroForce is worth watching and, for most firms, not yet worth building on. As at September 2026 it is Xero’s no-code AI agent builder, in early access through a waitlist, with general availability promised for later in 2026 and no published price. The idea is strong: agents that understand ledger work, run across a whole client base and log every action. What nobody outside early access can judge yet is how it behaves on messy client files, what it will cost, and how much of your automation you want inside one vendor. Join the waitlist, test on non-critical work if you get in, and decide once pricing and case studies are public.

A practitioner describes a task in natural language. XeroForce builds and runs the agent. No code. The example Xero offers: “on the 5th of every month, get clarification on uncoded transactions.” The agent runs on schedule, watches for triggers, drafts client messages, and waits for sign-off where the workflow requires one.

Xero is leading with five claims. The agents are always-on, running in the background over days or weeks. Every action is logged for audit. Workflows can run across an entire client base, not one client at a time. The logic is purpose-built for financial workflows. And the whole thing sits on Xero OS, the same AI-native layer that powers JAX.

Stripped of the marketing, that is the announcement. An agent builder, with logging, scoped to financial workflows, running on Xero’s platform.

The purpose-built argument has weight. Generic AI platforms treat the ledger as raw input. They pull data out, run it through a model, and push something back. They do not understand what a BAS is, what a payrun is, or what a sign-off should look like. Xero does, because Xero has spent two decades building that context.

For a workflow like uncoded transactions chasing, that context is the difference between an agent that works and an agent that needs constant correction. Firms that have tried building this kind of automation on top of generic platforms know how brittle the result tends to be.

The bulk-action point is also real. The bottleneck in most practices is not knowing what to do. It is doing the same thing consistently across two hundred client files every month. If XeroForce delivers on practice-wide execution, that is meaningful leverage.

Audit trails are being positioned as a differentiator. They are not. They are table stakes. Any AI product touching client data without a logged audit trail should not be in a regulated firm in the first place. Xero gets credit for shipping it. Xero does not get credit for inventing it.

Always-on agents that run in the background over days or weeks are a more complicated proposition than the announcement suggests. The honest version of that feature is that something will be running, touching client data, sending messages, and updating records while no one is watching. The audit trail will tell you what happened after the fact. It will not stop a wrong thing from happening in the first place. Firms need to think hard about which workflows they actually want running unsupervised, and which they do not.

The natural language interface is also less novel than it sounds. Every AI agent platform shipping in 2026 lets you describe workflows in English. That is the baseline. The question is not whether the prompt works in the demo. It is whether the agent does the right thing on the messy, edge-case-ridden 47th client file of the day. That is not something a launch post can answer.

The bigger story, and the one Xero is not framing this way, is platform control.

XeroForce sits on Xero OS. JAX sits on Xero OS. Whatever Xero ships next will sit on Xero OS. Xero is building an operating layer underneath its accounting product and asking firms to build their automations on top of it.

That has obvious upside. Tighter integration. Better context. Workflows that understand the ledger they are operating on.

It also has a cost worth naming. Every workflow a firm builds inside XeroForce is a workflow that runs on Xero’s terms, on Xero’s pricing, with Xero’s roadmap deciding what is possible. Firms that build deep automation inside XeroForce are also building deep dependency on Xero. That is a reasonable trade for some firms. It is a real risk for others, particularly those who have lived through previous platform shifts and watched pricing or capability change without notice.

The right question to ask is not “does XeroForce work.” It is “what is the cost of building my practice’s automation layer inside Xero specifically.”

These are two routes to the same job rather than two rival products. MCP (Model Context Protocol) is the open standard that lets a general AI assistant such as Claude or ChatGPT connect to software you already run. Xero maintains an official open-source MCP server that can read Xero data and create records such as contacts and invoices, and Xero’s own Claude integration went live in May 2026. We explain the standard, and the questions to ask before connecting it to client data, in MCP for accountants, explained.

The practical difference is who builds and who governs. With XeroForce, Xero hosts the agent, supplies the financial logic and keeps the log, and your automation lives inside Xero. With MCP, you choose the assistant, one conversation can span Xero plus your practice management and document tools, and you own the prompts, the permissions and the review step.

• Lean towards XeroForce if nearly all the work sits inside Xero, you want scheduled agents running across the client base, and you would rather not maintain prompts yourself.
• Lean towards MCP if you need to work across systems today, want to stay portable between AI assistants, or want read-only access while you build trust.

Plenty of firms will end up using both. Either way the rule is the same: scoped permissions, and a human confirming anything that leaves the building. If a client has already connected their Xero file to an AI tool on their own, our guide to what accountants should do next covers that conversation.

For most firms, the sensible move is to wait. Early access is for firms that have the capacity to test something carefully on non-critical workflows, with someone senior watching what the agent actually does. That is a small number of firms.

For everyone else, the right posture is sceptical interest. Read the case studies when they arrive. Ask other firms what broke. Pay attention to what Xero quietly changes in the first six months, which will tell you more about the maturity of the product than any launch post will.

XeroForce might be the most significant thing Xero has shipped in years. It might also be a competent first version of a product that needs another two cycles before it is ready for production use across a real client base. Both are possible. Both have happened before in this category.

The waitlist is open. Worth knowing about. Worth waiting to see. For where XeroForce sits in the wider AI stack, see our AI for accounting firms guide.

Update, August 2026: At Xerocon Denver, Xero previewed a pre-built month-end agent, listed as coming soon, that will work through document and reconciliation status, then the prepayment and amortisation journals, and hand back a full account of what it did for a human to review and accept. Our Xerocon Denver recap has the detail, and our Xerocon London rundown covers what Xero said about it in July.

What is XeroForce (or Xero Force)?

A no-code agent builder from Xero, sometimes searched as “Xero Force”. You describe a workflow in plain English, and it builds and runs an agent across Xero and the apps connected to it. Xero’s own example is chasing clarification on uncoded transactions on the fifth of every month. It runs on Xero OS, the same layer underneath JAX, and every action it takes is logged.

Is XeroForce worth adopting yet?

For most firms, not yet. Early access suits firms with the capacity to test carefully on non-critical workflows with someone senior watching what the agent actually does, which is a small number of firms. For everyone else the useful posture is sceptical interest: read the case studies when they arrive, ask other firms what broke, and watch what Xero quietly changes in the first six months. That will tell you more about the product’s maturity than any launch post.

How much does XeroForce cost, and where is it available?

As at September 2026, Xero has not published a price, and it has not said whether XeroForce will sit inside existing Xero plans or be sold separately. It also has not published a list of regions. Access is through the waitlist on Xero’s site, with general availability promised for later in 2026. Until the number is public, do not build a business case that assumes it will be included.

XeroForce vs MCP: how is it different from a general AI platform?

Context and control. Generic platforms treat the ledger as raw input, without understanding what a BAS, a payrun or a sign-off should look like, and Xero has two decades of that context built in. MCP lets you connect the assistant of your choice to Xero and other systems, which gives you portability and cross-system reach, but you carry more of the setup and governance yourself. The natural language interface is not a differentiator either way, since every agent platform shipping in 2026 has one.

What are the risks of building automation inside XeroForce?

Two. First, always-on agents mean something is touching client data, sending messages and updating records while nobody is watching, and an audit trail only tells you what happened afterwards, so decide deliberately which workflows run unsupervised. Second, dependency: every workflow you build runs on Xero’s terms, pricing and roadmap. That is a reasonable trade for some firms and a real risk for others, particularly those who have watched a platform change pricing or capability without notice.