Clark

• Preparing employee benefit plan audit workpapers: extracting contribution registers, consolidating payroll files and testing contributions and eligibility against the plan's records
• Selecting a reproducible, documented sample from an extracted population in Excel and testing each item against invoices or confirmations
• Agreeing a month's bank statement to the trial balance and preparing the exceptions for an auditor's judgment
• Chasing missing provided-by-client (PBC) items and picking the testing back up when the client responds
• Running a peer-review-style check of a completed engagement file before partner sign-off
• Standardizing workpaper templates and review gates across an audit practice with Methodology Studio

Clark is one of the more concrete attempts at an audit agent: it is built around the things that make audit work defensible (firm methodology, source-linked evidence, review gates and partner sign-off) and puts a plan-approval step in front of autonomous execution. Its strongest documented ground is EBP testing, where its own case studies describe the work in detail and are candid that the figures are benchmarks or reported results rather than measured before-and-after reductions. The Peer Review Guarantee is unusual and has published terms, but it is narrow and capped by contract. Firms should treat the rest as diligence items: pricing, SSO, data residency, subprocessors and exit terms are all settled in a sales-led review, the AI models are undisclosed, and the Clerks branding from the November 2026 launch is newer than the site's own Ava-based documentation.

About Clark

Clark, which was called uiAgent until June 2026, began with employee benefit plan (EBP) audits and now positions itself as an operating platform for audit and accounting firms. The platform keeps client requests, source documents, workpapers, review notes and sign-offs in one engagement record, and the AI does the preparation work inside it. Clark's website introduces that AI as Ava, an AI teammate, while the November 2, 2026 launch release calls its autonomous AI employees Clerks; both run on what Clark calls Material Intelligence. The workflow is delegation rather than chat. A team defines a task, such as preparing a workpaper, reconciling an account or testing a sample against supporting documents, and sets the client, reporting period, source records, working environment and the firm's procedures. Before starting, Ava shows the planned actions, the systems involved, permissions and review requirements, and the team can approve, adjust or cancel that plan. The agent then works without being prompted at each step, links every figure and proposed conclusion to its source document, and flags exceptions for an auditor. According to the launch release, a Clerk is assigned to an engagement the way a staff auditor would be, works through audit areas from start to finish, requests missing items from the client, tracks the response and resumes when the information arrives, and carries context forward with a multi-layered memory system. Clark's FAQ adds that its agents can navigate websites and pull data where no API exists, using access the firm authorizes. Firms encode their own approach in Methodology Studio (templates of sections, assertions, procedures, workpapers, review gates and sign-off requirements), and the June 2026 rebrand release also introduced Agent Studio for building custom agents without code. An Excel add-in runs an Extract, Sample and Test sequence beside existing workbooks, with reproducible, seeded sample selection. Clark lists financial statement, EBP/401(k), single, not-for-profit, SOC and payroll audits, plus accounting work such as bank reconciliations, month-end close, rollforwards, variance analysis and financial statement tie-outs. Two things set it apart from general AI tools: it is built around audit documentation and review standards, and Clark backs Ava's work with a Peer Review Guarantee (covered extraction errors or misses that become peer review findings, with damages capped by the agreement). The tradeoffs are typical of a young, sales-led vertical vendor: no public pricing, demo-only access, availability stated for the United States, and SSO, data residency, subprocessors, retention and export terms that are confirmed in a deployment review rather than published. Clark does not name the AI models it uses. Licensed CPAs remain responsible for judgments, review and the opinion.