Key Insights:
Most firms already have the AI slide. It says the right things: a pilot program, a tools committee, a paragraph about the future of the profession. Then the deck closes and the engagement opens: a senior chasing two PBC items from a client who went quiet, a manager re-validating evidence that didn't tie out the first time, a partner reading a draft memo that should have been ready a week ago. The slide and the work never touch.
That distance, between the AI vision on the slide and the AI execution that reaches the engagement, is the real story for 2026. Firms already have plenty of assistive AI: tools that summarize a document or draft a memo when someone asks. The step beyond that is agentic AI, where agents execute the routine workflow steps and practitioner judgment goes where it counts: directing the work, deciding what's material, and standing behind the conclusion. The firms pulling ahead are building that operating model now. This article covers why most AI strategies stall at vision, which workflows actually move first, and what makes agent-executed work survive the review and regulatory bar.
Most firms know they need to do something with AI. Far fewer have actually done it. The 2025 National MAP Survey of 1,073 U.S. CPA firms put hard numbers on the gap: 65% of firms describe their mindset toward AI as proactive or open, yet most had not engaged in a formal program or begun to take real advantage of the technology. The biggest barrier firms named was lack of time, at 41%. Close behind: not knowing where to start.
"Where to start" is where most firms get stuck. The team has read the whitepapers, sat through the conference sessions, maybe stood up a pilot or two. The awareness is there. What's missing is a sequenced plan that touches the workflows the engagement team actually runs.
This isn't unique to accounting. Three years into the generative AI wave, enterprises across industries are still wrestling with the same fundamentals: use-case selection, business outcomes, a deployment runway, and scaling what works. Vision is cheap. Execution is hard.
For audit and advisory, there's an extra layer: trust. In a Deloitte survey of more than 3,300 finance and accounting professionals, trust came back as the number one barrier to agentic AI adoption. That's a reasonable instinct in a profession where every output carries professional liability. The firms moving fastest aren't ignoring the trust question; they're answering it, with purpose-built partners and governance that holds up under review.
Agentic AI in audit isn't a chatbot that learned accounting. It takes action: handling multistep work, pulling information from third-party systems, and working directly with files.
That changes what the strategy is for. A practical strategy moves specific engagement workflows from manual execution to agent execution, one tier at a time, rather than deploying a tool firm-wide and hoping. That's the work Fieldguide's Field Agents are purpose-built to execute. The firms making real progress share a pattern: they embed AI inside actual workflows, connect it to their core systems, and name the humans accountable for governance and judgment. The dividing line between firms that get production value and firms that get pilots is governance.
Before you can pick a path forward, you need an honest read on where you are. The AI Maturity Framework defines six levels of autonomy for audit and advisory firms, from Level 0, where every step of the engagement is manual, to Level 5, where the Agent Workforce runs the engagement and practitioners orchestrate at the firm level.
Most firms sit at Level 0 or 1 today. Work lives in email, spreadsheets, and disconnected tools, and capacity is fixed by headcount. Nearly everyone shares that starting line. Placing your firm on the framework identifies the next level, and the next level is always attainable: a firm at Level 1 can reach Level 2 by embedding purpose-built AI in the workflow steps practitioners already run, no five-year transformation plan required.
Start where the work is structured, the volume is high, the rules are clear, and the outputs can be verified. That sounds obvious, and it's where a lot of firms still go wrong. They point AI at the most complex problems first, when the durable gains start with the most routine ones. The practical entry point is intelligent process automation: retrieving client financial data from ERP systems, banking platforms, and spreadsheets. This is the first tier of work to move. Structured. Repeatable. Built around information the team can actually evaluate.
Industry-specific reconciliations sit in the same tier: grant tracking and expense allocation for nonprofits, rent roll and CAM reconciliations for real estate, receivables and revenue recognition for professional services.
The same logic covers the core engagement work itself: validating client evidence the moment it arrives, drafting precise PBC requests, and running control tests against defined procedures. The common thread is structure, repetition, and verifiability. If a workflow has all three, it's a good first move.
Risk assessment augmentation and report drafting reward agent execution too, but they sit higher on the judgment curve. They're the workflows where practitioner review carries the most weight, which is why they come after the review scaffolding is built, not before. Build the scaffolding first, then move up the tiers.
Governance isn't something you bolt on later. It runs in parallel with adoption, from day one.
Accountability gaps are a known problem across industries, and audit firms feel them more sharply because professional judgment still sits squarely with the practitioner. No AI changes that. The regulatory pressure isn't abstract either: the PCAOB's 2025 inspection priorities specifically call out auditor use of technology, including generative AI. No comprehensive AI standard for audit exists yet. Until one does, AI-related quality risks get evaluated through the quality management and auditing frameworks firms already follow.
So the operating principle for every workflow moved to agent execution is the same:
That principle is built into Fieldguide's operating model. Practitioners direct the work through Field Orchestrator, Field Orchestrator coordinates the Field Agents, and the Field Agents execute. Every output gets practitioner review and approval before it moves forward; no exceptions. When an agent runs, the platform captures its inputs, steps, and sources as a Trace, so the documentation is a byproduct of execution rather than a write-up the team produces after the fact. That's what makes the work reviewable in the first place.
Some firms look at agentic AI and think: we'll build our own. The instinct isn't wrong. The real decision isn't build or buy; it's how you do both. Firms with engineering talent get the most from it by partnering for the production-ready engagement foundation, then building what differentiates them on top of it: their methodology, their risk models, their client IP. A build team isn't the price of admission, either; the foundation is production-ready on its own, and firms without engineers adopt it as-is. Half of the Top 100 US CPA firms, including members of the Big Four, run on Fieldguide, and the firms getting the most from the platform aren't the ones that stopped building. They're the ones that stopped building plumbing. The firms that lead the next decade won't be the ones that wrote the most code; they'll be the ones that made the sharpest calls about where to build and where to partner.
McKinsey research finds that many organizations focus on enterprise-wide copilots and chatbots, and the gains spread so thinly across employees that they're hard to see in top- or bottom-line results. Agentic AI value comes from putting agents inside specific workflows: work the firm bills against, not a productivity wrapper around email.
That sequencing is built into Fieldguide's platform design. AI Assist, the platform's human-orchestrated category that includes AI Chat and AI Actions, covers practitioner-led task work. The Agent Workforce handles execution at specific workflow steps across the engagement lifecycle, with practitioner review and approval of outputs. In one BerryDunn case study, the firm reported doubled engagement capacity in its SOC engagement workflow.
Firms that build the operating model now, moving one workflow and one review point at a time, are better positioned than the ones still pointing at the 2025 pilot deck.
The firms pulling ahead have stopped treating AI as a separate step beside the engagement. Fieldguide puts the whole engagement on one end-to-end AI-native platform purpose-built for audit and advisory: workflow, agents, documentation, and sign-off in a single system. Because the work and its record are produced together, the audit trail regulators expect already exists when review begins. Field Agents execute the routine steps across the engagement lifecycle, and practitioners review every output and own the final professional judgment. Request a demo to see how a live engagement runs on the platform.