4 SET 2026 · AI Office Hours: Why AI Still Needs Humans, What Breaks, and How to Build Around the FailuresIn this episode of AI Office Hours, presented as a special simulcast of The Neil Haley Show, Neil “The Media Giant” Haley and Patrick Riley discuss the frustrating reality of working with AI every day: incredible productivity one moment, broken tools and lost work the next. They examine Perplexity, Manus, Claude, ChatGPT, Gemini, AI agents, job replacement, software reliability, and why businesses may need more human oversight—not less—as AI becomes more powerful.Patrick begins with an update on his ghostwriting project. After a Manus update required him to back up his tasks, he temporarily moved work into Perplexity. He connected it to Gmail and asked it to analyze months of correspondence with his ghostwriting company.The result surprised him.Perplexity quickly organized the communication and identified numerous upsell proposals the company had made while the primary book project was still unfinished. Patrick says AI helped him cut through a year of scattered emails and see the situation more clearly.He is also using his AI agent, “Rich,” as a developmental editor to review the manuscript, identify problems, and provide feedback before proofreading and copyediting.That leads into a broader discussion about autonomous agents and guardrails.Patrick references reports and experiments involving AI systems finding unintended ways around sandbox limitations while attempting to complete assigned tasks. His concern is that once businesses give agents access to networks, email, files, or applications, the system may find solutions the human operator did not anticipate.Neil connects that concern to his own AI experiences.He describes AI as a constant love-hate relationship. When it works, the productivity gains can be extraordinary. When tools break, however, hours can disappear.Neil has built a Claude-based business system containing client information, finances, projects, invoices, tasks, and other operating data. He frequently works through voice, treating the AI almost like an executive assistant or business partner.But reliability remains a major issue.Neil describes instances where Claude appeared to complete invoices but failed to preserve them, lost context shortly after a conversation began, failed to package information for a new session, and repeatedly encountered tool errors.For a solo entrepreneur experimenting with AI, losing 20 or 30 minutes can be frustrating.For a corporation depending on automation at scale, the consequences could be much larger.That brings Neil and Patrick to one of the episode’s biggest questions: Are AI companies overselling the idea that AI will replace human employees?Patrick believes AI itself is not a false promise. It is an extremely powerful tool.What may be oversold, he says, is the claim that companies can simply eliminate people and replace them with autonomous systems.The technology is not reliable enough for that in many environments.They argue that AI currently works best as an amplifier of human capability.Accountants, writers, developers, managers, marketers, and executives can use AI to work faster, analyze more information, automate repetitive tasks, and improve productivity.But someone still needs to supervise the output, catch mistakes, provide judgment, and take responsibility when things go wrong.Patrick says his own organization is experimenting with AI to improve productivity, but his goal is not to shrink his team. He wants to grow it.Neil goes even further, suggesting that as AI systems become more complex, businesses may actually need more people capable of managing AI, not fewer.Another lesson comes from Neil’s decision to make his own AI operating system platform-agnostic.When Claude became temporarily unavailable, he took the same bundled business files and successfully ran versions of the system through ChatGPT, Gemini, and Grok.That redundancy matters.If an organization builds its entire workflow around one AI provider and that platform fails, changes pricing, removes a feature, or encounters an outage, the business can become trapped.Neil believes the better approach is to own your information, prompts, workflows, and knowledge base so they can move between models.The central takeaway is clear: AI is powerful, but it is still unreliable enough that human judgment, backup systems, and multiple platforms matter.The future may not be AI replacing people.It may be people who understand how to manage AI replacing people who do not.For more from Patrick Riley, his books, and AI consulting, visit UnderwayBooks.com. Become a supporter of this podcast: https://www.spreaker.com/podcast/the-neil-haley-show--6503396/support.