Workiva AI: what it does and how to prepare

Until this year, Workiva AI opened as a separate panel in each part of the product. In 2026 Workiva combined those features into one companion that follows you across the platform, then added agents that carry out multi-step reporting work. Workiva’s product announcements this year have been about agents, grounding, and governance.

A finance, audit, or risk team needs to know which of these features it can use on a task it already does, and what needs to be in place first. This guide describes each part as of October 5, 2026, and the setup decisions that come before any of it. Workiva is releasing features quickly, so check the linked documentation for the current status of anything you plan to use.

The companion opens from the sparkles icon in the platform toolbar. You can also right-click a file or search result and choose Ask Workiva AI, or start a quick edit. It keeps the conversation as you move between documents, spreadsheets, presentations, Wdata, and files, so you can ask about a table in one file and continue in another.

It drafts and edits text, answers questions about the files you reference, and runs multi-step workflows that your workspace has access to. Workiva lists benchmarking, tie-outs, the sustainability disclosure agent, and roll forward among those workflows. You only see the ones your workspace is entitled to.

In Wdata, the companion can create a query from a plain-language request, improve or fix an existing query, and explain the data sources a query uses. It writes the SQL for you to insert. Test the result against totals you already know before it feeds a spreadsheet; the Wdata guide explains how queries, tables, and connections fit together.

Agents handle a defined reporting or GRC job, usually with a person reviewing each result. Their availability varies more than the companion’s. Workiva ties several of them, including the tie-out, benchmarking, and sustainability disclosure agents, to Advanced solution tiers, and several start in Workiva’s early access program, which requires signing Workiva’s feature preview terms.

Tie-outs. Workiva announced a Tie-Out Agent in July 2026 for customers in advanced solution tiers. Workiva’s support center describes the Tie-out Assistant as an early access feature for 10-⁠Ks and 10-⁠Qs. It checks a report for internal consistency, groups related figures, and flags possible mismatches. General availability has not been announced. Reviewers mark results verified, flag issues, and add notes. Starting a new run clears the previous notes and resolved items, which matters if those notes are part of your review record.

Benchmarking. Benchmarking Insights compares your disclosures with peer filings. It began as an early access program in July 2026, and Workiva says continued access after general availability requires the Advanced tier.

XBRL validation. A validation agent explains XBRL validation errors before an SEC filing goes to EDGAR. Workiva currently supports it for US-GAAP filings only.

Roll forward. The Roll Forward and Prep Agent, in early access, moves data, formatting, links, and attachments into the next period. If your team rolls packages forward by hand today, the roll-forward checklist still describes the checks a reviewer should make afterward.

Sustainability disclosures. In Sustainability Reporting Advanced workspaces, the Sustainability Disclosure Agent can draft disclosure narrative or review a draft against the requirements you name, and returns a compliance scorecard. ESRS support requires the CSRD add-on.

Internal audit and GRC. Automated testing agents are generally available to entitled customers. They check population and sample files before testing, generate test attributes and rules, and execute test procedures. Policy management, audit report, and remediation agents are in Workiva’s early access program. Teams running SOX testing in Workiva can see how these fit with Controls Management.

Workiva Knowledge is a persistent layer built from your organization’s data, instructions, and content: prior filings, internal policies, auditor checklists, and similar material. Workiva AI uses it to ground answers and cite sources. Workiva announced Knowledge in July 2026; confirm with your Workiva account team whether it is available in your workspace. Knowledge is only as current as what you put in it, so each item needs an owner and a refresh date.

Agent Studio, introduced at Workiva’s Amplify conference in September 2026, lets your team build agents without code. As of October 5, 2026, Workiva has not published its general availability or the tiers that include it. You describe a process in plain language or adapt a prebuilt template, ground the agent in Knowledge, control who can build agents and what data they can touch, and schedule recurring runs. Workiva’s example is a period-end flash report with metrics, variances, and commentary. Treat an agent like any recurring process: it needs an owner, a reviewer, and instructions for when it fails.

The Workiva MCP connects an outside AI assistant to your Workiva content within your Workiva permissions. As of October 5, 2026, Workiva’s documentation describes it as an early access feature until November 1, 2026. At launch it is read-only, it is added to Claude Desktop as a custom connector, and a security administrator grants access to each workspace from a Connected AI agents console. Decide which workspaces it should reach before anyone connects.

Workiva AI becomes available once your organization signs Workiva’s AI Terms of Use, and it is then enabled for every workspace and user. Workiva states that the terms cannot be customized. Organization administrators can restrict it from Organization Admin, under Configuration and the Workiva AI subtab:

• Workspaces. Turn Workiva AI on or off for each workspace, or for the whole organization.
• People. In enabled workspaces, the Workiva AI User role is assigned to all users by default. An Org User Admin can limit access to named people by choosing Only users with the Workiva AI User organization role and assigning the role to them.
• Model providers. Workiva offers models from Google, Microsoft (OpenAI), and AWS (Anthropic). Administrators can enable or disable each provider, and the options vary by region.

Workiva states that customer data is not used to train Workiva AI or third-party models, and that third-party providers do not retain it for training. Its FAQ also says AI output is not always accurate, including on recent laws, regulations, and accounting guidance. The FAQ also says Workiva may store prompts and referenced files to operate, secure, and improve the service, and it strongly advises against entering sensitive customer data into Workiva AI. These points belong in your internal AI policy for the platform.

Start with one recurring task that has a named owner, a reviewer, and a prior cycle whose results you trust. Confirm which feature and tier the task needs. Clean up the links, labels, and reference files the AI will read, because weak source material shows up in the output. Then agree what the reviewer checks and which records you keep, such as activity log entries, tie-out notes, and approvals.

The Workiva AI readiness checklist turns these steps into a list you can work through with your administrator and reviewers. For the review itself, read how to review Workiva AI output before the first live cycle.