Thomson Reuters Releases Latest CoCounsel for Tax and Accounting

Thomson Reuters released a new CoCounsel for tax and accounting on October 22, 2026, an AI platform that writes and runs its own code to execute multistep workflows.

The platform is available now to all customers in the United States, and Thomson Reuters said more features will follow. The release covers tax, audit and accounting work, and adds expanded integrations with UltraTax, Engagement Manager, Guided Assurance, Ready to Review and Ready to Advise.

For finance professionals, the change is in who does the calculating. Earlier versions of the assistant answered questions and handled simple tasks. The new release is built to produce the finished workpaper, and the accountant’s job becomes checking it.

That shift matters to firms weighing software budgets and staffing models. It also puts Thomson Reuters in direct competition with other vendors, including Wolters Kluwer, which added agentic tax preparation to TaxWise, and Blue J, which raised a Series D round for AI in tax advisory.

CoCounsel for tax and accounting now writes its own code

David Wong, chief product officer at Thomson Reuters, said in a briefing on Wednesday that previous versions coped with simple tasks. They struggled when work was more complex, when several tasks had to be strung together, or when a case deviated slightly from what the model had been taught.

The company wanted an AI capable of complicated problem-solving and independent tool use. Examples include interpreting complex tax forms, producing workpapers and handling hard audit questions.

“Building AI for professions is much harder than trying to build for general purpose productivity or helping you to summarize emails or helping you book concert tickets. AI for tax and audit and accounting is hard.”

David Wong, Chief Product Officer, Thomson Reuters

The technical change is code execution. Lucas Adams, vice president of engineering, said earlier agents generally read and wrote text, and users then connected that text to code themselves. That was how “you can sort of simulate an agent doing these more complex tasks.”

Adams said it is fundamentally different when the agent writes code itself, whether snippets or whole files, and then runs it. The agent no longer works from a CSV export or text extracted from a document. It reads the information natively, as a computer program would.

He compared it to giving an AI agent its own computer. The approach helps most with “ad hoc deterministic things” such as running and verifying calculations.

What the platform builds and calculates

Thomson Reuters lists a set of tasks the platform performs. Calculations include depreciation, amortization, required minimum distributions, qualified business income, threshold checks and multi-scenario comparisons.

It builds multi-tab Excel workbooks with live formulas and a documented assumptions tab. It drafts memos and other deliverables in Word and PDF.

It also updates spreadsheets and documents uploaded by staff, following the structure of a firm’s own workpaper templates. A further function gathers client documents and communications in one place, so everyone working on an account sees the full context of the relationship.

The platform has native Model Context Protocol connectivity. That lets CoCounsel for tax and accounting operate within other applications and systems, and bring them into its own context.

Firms already using the CoCounsel Audit & Accounting product will recognise the workspace. The difference is the agent’s ability to carry a task through from documents to a finished file.

How code execution checks the numbers

Dave Ragones, head of product, demonstrated the platform building a full amortization table for a $5 million commercial real estate loan. He loaded in the loan agreement and a template.

In very little time it returned an Excel file containing a loan summary, a complete amortization table, a balloon payment calculation and notes for follow-up. Everything was sourced and cited.

Adams said the verification happened out of sight. “When we’re calculating the balloon payment as well as the interest payment in the spreadsheet, you don’t see this in the product, but behind the scenes we recalculated that in Python code,” he said.

The agent took the logic of the Excel formulas, rewrote it in Python, reran it and tied the two sets of numbers together to check they matched. Adams said the AI built the template, the workbook and the Excel files with code it wrote and executed, rather than producing something like a CSV.

The demonstration also showed the agent correcting itself. “At the end, you can see that there was actually some failure in the way the schedule was built. It fixed that failure, it rebuilt, and then it verified the values,” Adams said.

Ragones said the platform can create finished work products and explain how the work was done. That matters for firms that must document their reasoning in the file.

Human review stays in the process

Thomson Reuters is not presenting the platform as a replacement for the preparer or reviewer. Ragones said the aim is for CoCounsel to do much, if not all, of the work, with the professional acting as the human in the loop.

“The professional role in that process is, of course, to be the human in the loop to review it, to verify, to make sure that the work is correct,” he said.

Joel Hron, chief technology officer, said the code execution is only part of the design. The system draws on Thomson Reuters’ library of content so that it reasons from the right facts and the right authority before it runs a calculation.

“The code execution piece is table stakes now; every company in this space has it. What’s hard is making sure the system is reasoning from the right facts and the right authority before it ever runs a calculation.”

Joel Hron, Chief Technology Officer, Thomson Reuters

Hron said the content is kept current and tested by people who do this work, so a professional can check it rather than take the company’s word for it.

The workflow also fits the oversight questions regulators and professional bodies are raising. The AICPA’s AI tax risk group has put the duties of CPAs who use AI on its agenda, and the sourcing and citation in each output is what a reviewer will need.

CoCounsel for tax and accounting in a crowded field

Hron’s remark that code execution is table stakes is a statement about competitors. Vendors across tax and accounting are shipping agents that work inside the tax engine, the ledger or the close process.

Thomson Reuters’ argument is that its edge lies in the content behind the agent, together with the tie-in to its own preparation and workflow products. The company sells UltraTax for return preparation, and the new integrations connect CoCounsel to that product as well as to Engagement Manager, Guided Assurance, Ready to Review and Ready to Advise. Ready to Review was first introduced in the company’s November 2025 agentic AI announcement.

Firms comparing options can start with Accountio’s vendor comparisons, and its guide to AI audit software covers CoCounsel alongside DataSnipper, Inflo, MindBridge and Caseware.

The wider context is a partnership between OpenAI and Wolters Kluwer, showing that the large information providers are all building agents on frontier models.

What firms should watch next

Three questions will decide how useful the release proves. The first is whether reviewers find the output accurate enough to check quickly, or whether they end up rebuilding it. The second is how the Model Context Protocol connections behave inside a firm’s existing stack. The third is how firms charge for work the agent now does.

Thomson Reuters did not give figures for time saved in the briefing. Firms should test CoCounsel for tax and accounting on their own workpapers before relying on any such claim.

The platform is available only to US customers for now, and Thomson Reuters said additional features and capabilities are coming. UK and EU customers have no date.