Audit busy season has eased. The reflective phase is starting. Across firms we’re hearing the same conversation: AI proved itself this season. The question now is what it does next.
For most teams, AI in 2026 has been an assistant. It pulls anomalies, drafts memos, summarizes documents, answers questions. The auditor stays in the driver’s seat and reviews everything that comes out.
The next shift is already underway. AI is moving from assistant to agent – systems that don’t just answer, they act. They run the test, query the data, draft the workpaper, and hand back results ready for review.
This edition of Tomorrow’s Audit looks at what that shift means for audit, why it’s arriving faster than most expected, and what firms need in place to be ready for it.
What “agentic” actually means in audit
The language is everywhere right now. The substance is narrower than the hype suggests.
An AI agent is a system that:
- Receives a goal, not just a prompt
- Decides on the steps needed to reach it
- Executes those steps autonomously, including connecting to data and tools
- Returns a result the user can review, accept, or adjust
In audit terms, that means the difference between asking AI to summarize the GL and asking it to complete the substantive testing on revenue, flag anomalies, and prepare a draft memo for review.
The first is a tool. The second is a team member.
Resource: AICPA & CIMA – Artificial Intelligence:
Why this is happening now
Three things have come together to make agentic audit workflows realistic in 2026:
1. Models are reliable enough. Frontier models can now sustain multi-step reasoning across audit tasks without losing the thread or hallucinating critical numbers, provided the inputs are clean.
2. The plumbing exists. Open standards like the Model Context Protocol (MCP) let AI systems connect directly to source data without bespoke integration work. What used to take a six-month engineering project takes a configuration change.
3. Firms have proof points. Earlier this year Validis CEO Michael Turner demonstrated Claude running live financial analysis on standardized accounting data, with no exports, reformatting, or manual cleanup. Nine minutes of footage that compressed years of theoretical conversation into a working example.
What was experimental in 2024 is operational in 2026. The firms that move first are not building from scratch – they’re connecting components that already exist.
What changes inside the audit workflow
Agentic AI doesn’t replace the auditor. It changes where the auditor’s time goes.
In a traditional workflow, an audit team spends meaningful time on:
- Requesting and chasing client data
- Reformatting and reconciling files
- Running standard analytics
- Drafting initial workpapers
- Reviewing and signing off
In an agentic workflow, the first four shrink. The agent connects to the source, runs the analytics, drafts the workpaper, and presents it. The auditor’s time concentrates on review, judgment, risk, and sign-off – the work that requires their qualification.
The implications are not just efficiency:
- Junior auditors learn faster. They review agent output rather than producing it manually. The exposure to judgment-level work happens earlier in their career.
- Engagement timelines compress. Less time spent in preparation means more capacity to absorb scope creep or unexpected issues.
- Documentation improves. Agents log every step they took, producing audit trails that are often more thorough than human-written notes.
None of this works if the data is unreliable.
The data readiness question (again)
Every wave of audit technology has run into the same wall. Tools improve. The data underneath does not.
When data arrives late, in inconsistent formats, with reconciliation gaps, the agent inherits the same problems the human did. It cleans rather than analyzes. It asks for clarification. It loops. The promised efficiency disappears, and worse – the result is harder to defend because the agent’s reasoning gets tangled in remediation steps that should never have been needed.
This is the structural argument firms keep returning to:
- Standardized data is the precondition for AI value, not a nice-to-have
- Agent reliability scales with data reliability
- Audit defensibility depends on traceable data lineage from source to conclusion
Regulators are watching this closely. The IAASB’s Technology focus and the PCAOB’s quality management standards both reflect the same expectation: technology adoption should strengthen evidence, not obscure it.
Resource: IAASB – Technology and Innovation in Audit and Assurance
What firms should be doing in Q2
The firms moving fastest are not the ones launching AI labs. They are the ones quietly fixing the pipeline that everything else sits on.
A practical Q2 2026 list:
- Map your data sources. Where does client accounting data actually come from? How standardized is it on arrival?
- Pilot one agentic workflow on a low-risk engagement. Substantive testing on revenue or AP is a good starting point. Constrained scope, observable output.
- Define your governance early. Who reviews agent output? How is it documented? What’s escalated for partner sign-off?
- Audit your audit trail. If a regulator asked tomorrow how an AI-assisted conclusion was reached, could you reconstruct every step?
Each of these is achievable now. None of them require waiting for the next wave of model improvements.
Looking ahead
The shift from AI assistant to AI agent is not a single moment. It’s a gradient that’s already underway. Different firms will move at different speeds, and that’s appropriate, audit is not a sector where moving too fast is rewarded.
But the trajectory is clear. By the time the 2026 busy season arrives, agentic workflows will have moved from pilot to standard practice in the firms that are paying attention. And the firms that haven’t built the data foundation for it will be having a very different conversation.
Tomorrow’s audit is not just AI-powered. It’s AI-driven. The work to get ready for that started the moment busy season ended.
Final thought
AI doesn’t change what makes a good audit. Evidence still has to be reliable. Judgment still has to be sound. Documentation still has to be defensible.
What AI changes is how much of an auditor’s day goes into the work that demonstrates those qualities, versus the work that gets in the way.
Validis sits at the foundation layer of this shift. Standardized financial data, delivered straight from source systems including QuickBooks, Sage, Xero, Microsoft Dynamics, and NetSuite, ready for use in Excel, Caseware, Power BI, and AI tools like Claude.
Agentic audit workflows don’t replace the auditor. They free the auditor to be one.
