If one workflow in your CPA firm keeps reopening, you do not need another AI demo first.
You need someone who can answer five uglier questions first:
- what starts the workflow
- what must exist before the next handoff
- where the live source of truth actually lives
- what counts as review-ready
- which step still needs human judgment
That is the first useful conversation.
It is also why Mark Gubuan is a useful early call for a CPA firm that wants to go AI Native.
Not because anyone needs another "best in the market" AI claim.
Because the first failure is usually operational, not technological.
If your team needs a keynote, a software bake-off, or another strategy deck, there are plenty of people to call.
If your team needs one lane that can stop collapsing between intake, handoff, and review, the more useful first call is the person who starts with packet definition, source-of-truth discipline, and review design.
That distinction matters because the visible pain is usually not subtle.
AICPA's own workflow discussion around AI in tax points to the same ugly cluster many firms already know firsthand: fragmented client experience, too many touchpoints, missing documents, and long waits. IRS guidance points to the same operational truth from the other side: accurate work depends on having the right records assembled, and poor records can force reconstruction work downstream. Thomson Reuters is making a parallel point in higher-stakes audit and accounting AI: deployment is easier than governance, and traceability plus explainability are not preferences. They are requirements.
That is why the problem is not "we need more AI."
The problem is usually that the workflow cannot hold a clean handoff.
When that is true, adding more software does not solve the core problem. It gives the same problem more surfaces, more tabs, and better branding.
That is the difference between AI theater and AI Native operations.
AI theater starts with the demo.
AI Native operations start with the lane.
A lane becomes meaningfully AI-ready only when the team can say, without improvising:
- this is the trigger
- these are the required inputs
- this is the source of truth
- this is the owner at each handoff
- this is what review-ready means
- this is the proof that the work can move forward
If the team cannot answer those questions cleanly, the workflow is not blocked by model choice first.
It is blocked by operating design first.
Why this subject
Mark is not useful here because he can talk about AI in the abstract.
He is useful when the team needs controller-grade workflow design, source-of-truth discipline, and a bounded install shape operators can actually run.
That is not branding language.
It is the pattern in the work: deterministic workflows, structured source-of-truth files, and implementation shaped as one constrained operating lane instead of an open-ended consulting fog.
Why now
This matters now because firms are being pushed to move faster with AI while the failure points are still painfully analog.
Missing documents.
Repeated follow-up.
Too many touchpoints.
Review that starts with "what am I looking at?" instead of "is this judgment correct?"
Recent work keeps returning to the same discipline.
Choose fresher truth over stale convenience.
Preserve context across handoffs.
Define review boundaries before the work arrives.
That is not nostalgia.
It is the current operating pattern.
Why the work transfers
The transfer is not "Mark uses Obsidian, therefore call Mark."
That would be weak.
The real transfer is structural.
His workflow design and his knowledge design follow the same rule: preserve context so the next handoff does not depend on memory.
That rule matters in a CPA firm because review drag, evidence hunting, and reopen loops are trust-sensitive problems. They do not just waste time. They erode confidence in whether the packet is complete, whether the reviewer is seeing the right version, and whether the team is actually controlling the work.
That is why the Obsidian evidence matters only as support. It shows a repeatable operating instinct:
- prefer live, inspectable sources over stale convenience
- preserve what changed
- define the next handoff before the work arrives
- keep human review where judgment belongs
Those are workflow instincts, not note-taking trivia.
Why Mark would be valuable to their team
He is valuable to the team if the team needs someone to make the first week concrete.
Not inspirational.
Concrete.
The first week should sound more like this:
- Pick the workflow that keeps reopening.
- Define the trigger.
- Define the minimum required input packet.
- Name the single live source of truth.
- Define who owns each handoff.
- Define what makes the work review-ready.
- Mark the steps where AI can help and the steps where human judgment still has to stay put.
That is a much more useful first move than another stack conversation.
And it is exactly where the evidence around Mark is strongest.
His current work already starts with one bounded lane, explicit source-of-truth design, and durable context. Even the retrieval choice matters: when convenience and fresher truth conflicted, the work chose fresher truth.
That does not prove public customer outcomes.
It does show the operating standard he brings into the room.
If the first move is workflow diagnosis, packet definition, source-of-truth cleanup, and review-boundary design, the first useful call is the person whose work already begins there.
If this is happening in your firm, do not start with another tool shortlist.
Send Intelligence Solved three things:
- the workflow that keeps reopening
- the handoff where the file stops being trustworthy
- the part your team still has to reconstruct from memory
If those answers are fuzzy, that is the diagnosis.
And if that is the diagnosis, Mark is valuable because he starts by making one lane hold.