# AI Office Hours: Managing AI Agents and Building a Business Operating System

# AI Office Hours: Managing AI Agents and Building a Business Operating System
In this episode of **AI Office Hours**, presented as a special simulcast of **The Neil Haley Show**, Neil “The Media Giant” Haley and Patrick Riley share another week of real-world experimentation with AI. Their conversation covers AI agents, API costs, meeting preparation, business knowledge systems, publishing, web development, and one major lesson: AI can dramatically increase productivity, but it still needs human management.
Patrick opens with a costly reminder. After changing the model behind one of his agents, the system used a more expensive Claude option than expected and burned through about twenty dollars in a day. His takeaway: AI budgets need spending limits, monitoring, and human oversight.
He then shares a story about an AI agent helping locate a missing cat. A pet owner traveling overseas programmed an agent to repeatedly check a local humane society’s lost-animal listings. The AI compared new postings with the cat’s description and eventually alerted the owner to a possible match. It turned out to be his pet. For Neil, the story shows where agents can be useful: repetitive monitoring and research humans would otherwise perform manually.
Neil then updates listeners on his own AI agents. His Marbleism agents are improving, generating sponsorship interest, booking appointments, helping distribute books, and creating social content.
But there are still mistakes. One agent promoted a client’s book using a fabricated cover. The post performed well and generated monetization revenue, but Neil still had to replace the image. He also describes telling another automated campaign to stop, believing it had stopped, and later discovering the agent was still working.
That reinforces one of the episode’s biggest lessons: **AI agents still require management.** Patrick agrees that without supervision, agents can repeat outdated information, overspend, continue unwanted tasks, or work from stale data. Both believe companies using multiple agents will still need a human in the loop.
Despite those problems, Neil remains impressed by the productivity gains. He estimates that some systems already save him several hours each week. He is also using an automated meeting-preparation loop that checks his calendar, Gmail, Google Drive, and historical documents before a call, then creates a briefing about the person or client he is meeting.
The conversation then turns to Neil’s most ambitious experiment: turning Claude into a combination of **fractional CFO, CEO, executive assistant, business coach, and accountability partner**.
Neil has assembled more than two dozen working documents covering clients, revenue, time tracking, podcasts, websites, referrals, strategy, priorities, and automation systems. His startup prompt tells Claude to read every attached file before doing any work and prove that it understands key business numbers before proceeding.
Patrick compares that to asking a human employee to read every document and spreadsheet before starting work. AI can process and reference that information much faster. Neil says the result is becoming a complete business operating system. When a conversation grows too long, he exports the files and starts again in a fresh session.
The hosts also discuss how different AI platforms have different strengths. Neil uses Claude for large knowledge sets and strategy, Google tools for calendars and documents, ChatGPT for images, and other platforms for filmmaking, coding, and web development. No single AI system is best at everything.
Patrick shares progress on app development and publishing, including using AI as a developmental editor to identify weaknesses in a manuscript before revisions go back to his ghostwriter.
The episode closes with a point both hosts return to repeatedly: AI may automate enormous amounts of work, but human skills become even...