Agentic audit is here. What it means for your firm
You opened your inbox last week and saw the word “agentic” again. This time it came with real names. On June 10, Caseware and MNP announced a partnership to build firm-specific AI agents that live right inside the engagement file.
Here’s the thing. You’re probably not MNP. You don’t have a co-development deal or an innovation team. So it’s fair to ask whether this matters to a 40-person practice in Ontario or a regional firm out west.
It does. Not because you need to act this week, but because the tools you already use are heading this way. The first pass of your file is about to get faster. And that raises a harder question. What still needs you?
In this post you’ll learn what has actually changed, what an agent does inside a file, and which review and documentation duties stay yours no matter what.
What Caseware and MNP actually announced
Let’s start with the facts, not the marketing.
The two parts of the deal
There are two pieces here, and they’re easy to blur together.
- A full platform rollout. MNP selected Caseware and will deploy its full Cloud Engagement Solutions suite across the practice. That’s the part most firms recognize. It’s the engagement software, scaled up.
- A co-development program. MNP also joined the Caseware Agentic Pioneer Program as a founding member, a partnership to build and deploy firm-specific AI agents on Caseware’s governed AI platform.
The second part is the news. MNP isn’t just buying agents off a shelf. Under the partnership it gets priority access to Caseware’s full Agent Builder suite, including the Agent Workbench, governed interfaces for firm-side development, and Agent Studio for building agents without heavy technical work.
In plain terms, MNP is helping build the agents it will use. Most firms will get the finished product later.
Why “governed” is the word that matters
Notice how often the word “governed” shows up. That’s not an accident.
An ungoverned agent is a chatbot you paste your file into. You can’t see how it reached its answer, and you can’t show a reviewer either. A governed agent runs inside the file, under the firm’s methodology, with a trail you can follow.
Caseware frames this as proprietary agents running on a secure platform built for the profession. The pitch is that the output stays reviewable and explainable. Keep that promise in mind. We’ll test it against Canadian standards in a minute.
What an agent inside the engagement file really does
So what’s the actual change to your day?
Think about what eats a junior’s week. Tying out balances. Pulling support. Drafting test procedures. Writing the first version of a memo. Summarizing a stack of contracts.
An agent aims at exactly that work. It links those single tasks into a chain and runs the chain.
Picture a simple example. You assign the “AR confirmations” step. The agent pulls the listing, drafts the confirmation requests, logs which ones came back, flags the gaps, and writes a draft conclusion. What used to be a half-day of staff time comes back in minutes as a draft.
That’s the first pass. It’s real, and it’s useful. But it’s a draft, not a conclusion.
What it inherits: methodology and audit trail
The selling point is that the agent inherits your file’s context. It works from the engagement methodology and leaves an audit trail behind it. It can request the right documents, run the right test procedures, document evidence, and flag issues for review.
That’s better than a generic chatbot. The work sits where your reviewer expects it. But “inherits the methodology” is not the same as “applied judgment.” It applied your rules. It didn’t decide whether the rules fit this client this year.
You don’t need a co-development deal to feel this
You might read all this and think it’s a big-firm story. It isn’t, and here’s why.
CaseWare Cloud already sits in thousands of Canadian small and mid-size files. Caseware has already rolled out a digital assistant, AiDA, inside the Cloud platform. The Pioneer Program is where the next layer gets built. When it’s stable, it ships down the product line to regular subscribers.
So MNP is the test lab. Your firm is the eventual customer. The gap between them is months and quarters, not years.
The vendor’s roadmap becomes your roadmap
This is the real takeaway for a mid-size practice. You don’t control when agents arrive in your file. Your software vendor does.
That cuts two ways.
- The good. You get enterprise-style tooling without an enterprise build. These platforms can give smaller firms enterprise-grade capability without the enterprise cost.
- The catch. A feature can land in your file before your firm has a policy for it. That’s the risk to manage now, before the toggle turns on.
The review questions that don’t go away
Here’s the core of it. An agent can do the first pass. It cannot take on your professional responsibility. CPAB is clear that as these tools spread, auditors still need a human-led approach. These four questions stay yours.
Who owns the conclusion
The agent drafts a conclusion. You own it.
That sounds obvious until busy season hits and a tidy, confident draft is sitting in the file. The risk isn’t that the agent is wrong. The risk is that it’s plausible, and a tired reviewer signs off without testing it.
Treat an agent’s conclusion the way you’d treat a first-year’s. Useful, and unsigned until you’ve checked it.
Does the documentation pass the experienced auditor test
CAS 230 sets a clear bar. The file has to let an experienced auditor with no prior connection to the engagement understand the procedures performed, the results, and the conclusions reached.
An agent’s output has to clear that same bar. So ask:
- Can a reviewer follow what the agent did and why?
- Is the support attached, or just referenced in a way only the agent could retrace?
- If a CPAB inspector opened this file, would the trail stand on its own?
A fast draft that can’t pass this test isn’t documentation. It’s a starting point that still needs work.
Where professional skepticism still lives
Skepticism is the thing software can’t inherit.
An agent matches patterns. It’s good at “this looks like prior year.” It’s weak at “this looks fine, and that’s exactly what’s bothering me.” Use the time the agent gives back on the odd contract, the related-party question, the estimate that moved for no clear reason. That’s where the audit gets earned.
Certification and your CSQM 1 system
This is where a lot of firms will get caught flat-footed.
You can’t just switch an agent on mid-engagement and sort out the rules later. CPAB expects firms to set policies around the responsible use of AI-enabled tools, including certification testing before they’re used in audits.
Under CSQM 1, your system of quality management is risk-based. An agent that touches audit evidence is a resource that introduces risk. So it belongs in your quality framework, with a response built for it. Decide who approves the tool, how you test that it does what it claims, and how you monitor it once it’s live.
The bottom line
Agentic audit stopped being a someday story on June 10th. The direction is set, and your software is heading there whether or not you have an innovation budget.
Here’s what to hold onto.
- The deal is a preview. MNP is the test lab. Your firm is the future customer, and the gap is short.
- The first pass is real. Agents can genuinely take routine work off your staff’s plate.
- The hard part stays yours. You still own the conclusion, the documentation, the skepticism, and the quality system.
- CPAB already set the expectation. Human-led approach, firm policies, and tool certification before use.
- Act before the toggle flips. A one-page policy and a named owner now beats scrambling later.
The firms that win the next few years won’t be the ones with the fanciest agents. They’ll be the ones who kept their review standards high while the first pass got faster.
We break down a development like this every week, in plain language a working CPA can use on Monday. Come back next week for the next one.