November 10, 2026

In-Person

Colorado Society of CPAs - Galleria Office Towers
720 S. Colorado Blvd, Suite 500N, Glendale, Arapahoe County, Colorado 80246

2 CPE Credits

From Chatting to Delegating: A Hands-On AI Agent Architecture

November 10, 2026
In-Person
2 CPE Credits

From Chatting to Delegating: A Hands-On AI Agent Architecture

  • View Bio

    Instructor

    Wesley
    Hartman

    View Bio

    Wesley Hartman

    Bio

    Wesley Hartman is the Founder of Automata Practice Development. He started in Information Technology in 2005 and specifically in IT for accounting in 2009 at Kirsch Kohn and Bridge, an accounting firm with 80 years of history.

    At Kirsch Kohn and Bridge, he pioneered an RPA platform in 2014 with an automation to input K-1 data from a spreadsheet into tax software. From there he grew the platform to include, data mining, custom emailing and invoicing, and decision making. He has spoken at conferences to share his knowledge and experience in robot engineering along with other topics relating to technology and efficiency.

    He is part of the AICPA Engage Tech+ committee and speaks at the AICPA Engage conference on how to implement emerging technology and methods for making practices more efficient using technology.

    His focus is on analyzing procedures with the mindset of engineering efficiency and process improvement. He will break down tasks into pieces to look at all the cogs in the machine in order to find ways that technology can evolve and transform accounting.

    Wesley holds a Bachelor's of Science in Computer Engineering and worked on an autonomous vehicle project, so automation is part of his identity. He also practices martial arts, enjoys building furniture in his garage, and puts together a haunted maze in his driveway for Halloween.

Most professionals can write a decent prompt and get a decent answer. But AI can now do far more than answer questions. It can research, browse the web, operate software, follow set standards, produce finished deliverables, and build working tools from a plain-English description — but only when it is treated like a team member rather than a search box. This hands-on workshop covers the first steps in that shift.

The session opens with a candid look at what separates a chat from agent delegation: what an agent actually is, why a single long chat degrades on a multi-step job, and why defining the scope, handoffs, and checkpoints produces better results than writing a better prompt.

Participants then build. Working in VS Code with Claude Code, they will construct a bookkeeping pipeline that turns raw bank statement PDFs into a categorized transaction ledger, an Excel P&L workbook, and an interactive dashboard. A supervisor agent plans the work and delegates it to three specialist subagents, each with its own instructions, narrow permissions, and definition of done. Along the way, participants will review and update the agent files, practice handing work to agents and back, and see how agents respond to ambiguous or bad information.

The session closes with a discussion of what is possible, what the risks are, and how agentic AI may change the way work gets done.

Learning Objectives

  1. Differentiate between a single AI chat and agent delegation, including why long chats degrade on multi-step tasks
  2. Apply a supervisor/subagent structure to break a multi-step bookkeeping workflow into scoped handoffs with clear success criteria
  3. Use Claude Code in VS Code to review and modify agent instruction files, permissions, and definitions of done
  4. Determine where to place human checkpoints and how to pass work between agents and tools without losing context
  5. Identify common agent failure modes, such as ambiguous or bad input data, and apply permissions, checkpoints, and review habits to reduce risk

Major Topics

  • Chat vs. agent delegation: what an agent is and why long chats degrade
  • Defining scope, handoffs, and checkpoints
  • Supervisor and subagent architecture
  • Hands-on build: bank statement PDFs to a transaction ledger, Excel P&L, and dashboard
  • Agent instruction files, permissions, and definitions of done
  • Passing context between agents, tools, and people
  • How agents handle ambiguity and bad information
  • Guardrails, risks, and the impact of agentic AI on professional work

CPE Credits Available

2 CPE Credits
2
Computer Software & Applications

Colorado Society of CPAs - Galleria Office Towers

720 S. Colorado Blvd, Suite 500N, Glendale, Arapahoe County, Colorado 80246

Get Directions

Things to Know About This Course

Course Level

  • Intermediate

Professional Area of Focus

  • Technology
  • Artificial Intelligence (AI)

Prerequisites

Regular experience using an AI chat tool (such as Claude, ChatGPT, or Copilot) to write prompts and review outputs. No programming experience required.

Advanced Preparation

Bring a laptop with VS Code, the Claude Code extension for VS Code, and Node.js installed, and an active Claude account [confirm required plan]. Please confirm installs with your IT department before the session. A personal device works as well, since the exercises are self-contained.

Intended Audience

CPAs and finance professionals who are comfortable with basic AI prompting and want to learn how to delegate multi-step work to AI agents.

Provider

Colorado

Register For This Event

  • COCPA Member

    $ 120 120
    Join

  • COCPA Non-Member

    $ 156 156
    Your Price

Please login to register.