ATL269 — Microsoft Agent 365

Program: Accounting Technology Lab
Hosts: Randy Johnston and Brian F. Tankersley, CPA.CITP, CGMA
Approximate runtime: 31 minutes
Primary topic: Governed deployment, monitoring, security, and economics of enterprise AI agents

200-Word Episode Summary

In ATL269, Randy Johnston and Brian Tankersley examine Microsoft Agent 365 as a control plane for deploying, monitoring, governing, and securing AI agents. They frame the shift as a move from per-seat software licensing toward an AI token economy, where tokens function like staff time, prompts replace checklists, and agents perform repeatable work at speed. Practical examples include invoice extraction, email drafting, financial-statement analysis, and budgeting—tasks that may cost pennies in model usage while still requiring review and judgment. The hosts argue that Agent 365 gives accounting firms an alternative to “Wild West” experimentation by extending Microsoft 365 security, auditability, and governance over agents. They discuss Copilot, Agent Builder, Copilot Studio, Microsoft Foundry, Azure AI, Power BI, Defender, Purview, and the E7 licensing bundle. Privacy, records retention, e-discovery, intellectual property, and workflow ownership receive attention because firms may expose sensitive client data or proprietary processes when using public AI platforms. Their practical recommendation is measured experimentation: convert checklists into prompts, move tasks into agents, retain humans in the loop, establish token budgets, and evaluate results firsthand. The message is urgent but cautious: firms need not operate at the bleeding edge, but they must start learning before competitors pull ahead.

Key Takeaways

· The economic unit of AI is shifting from a user license toward token consumption and task-level cost.

· Tokens can be managed like staff time, prompts like procedures, and agents like digital staff assignments.

· Agentic workflows depend on steps, loops, exception handling, context, permissions, and human review.

· Agent 365’s differentiator is governance: visibility, monitoring, security, auditability, and centralized control.

· Public AI tools create material concerns involving PII, PHI, client confidentiality, retention, e-discovery, and vendor training practices.

· Proprietary workflows may be valuable intellectual property and should not be surrendered casually to a model provider.

· Model selection should balance quality and cost with privacy, security, regulatory fit, and data ownership.

· Accounting firms should begin with controlled experiments and measurable use cases rather than enterprise-wide autonomous deployment.

· Human reviewers remain accountable for conclusions, professional judgment, client context, and exceptions.

· Firms need AI governance policies, token budgets, approved-tool lists, monitoring, and documented escalation procedures.

Catchy Quotes and Video Locations

· 01:48–01:50 — Brian Tankersley: “Yeah, so it’s an F-150 and not a G-Wagon.”

· 04:22–04:30 — Brian Tankersley: “This is like sending it to staff first, and the staff costs three cents.”

· 09:35–09:51 — Brian Tankersley: “I’m seeing tokens as staff time on the schedule… We used to have checklists, and now we have prompts.”

· 10:09–10:22 — Brian Tankersley: “They get work done so fast that the human is now the logjam in the process.”

· 15:32–15:48 — Randy Johnston: “Microsoft Agent 365 is the control plane for agents… Is it perfect yet? No. Is it pretty doggone good? Yes.”

· 17:25–17:41 — Brian Tankersley: “We have the grown-ups in charge now, and we’re going to systematize this in a way that we can regulate and do the right way.”

· 28:02–28:13 — Brian Tankersley: “Take your checklists and turn them into prompts, and take tasks that are on the schedule and push them into agents.”

· 29:31–29:43 — Brian Tankersley: “The train is leaving the station… If you don’t get started on this stuff, you’re going to be behind, and so it’s time to go.”

· 30:19–30:30 — Brian Tankersley: “It is critical that you get your hands dirty with some of these things, because it’s the only way that you will be able to evaluate whether something really works or not.”