Consulting firm replaces 25 specialized AI agents with single central system

AI First, an eight-person consulting firm, replaced its previous architecture of 25 specialized AI agents with a single central agent system called the 'AI Operating System.' This shift was triggered by the release of Opus 4.6 models, which demonstrated the ability to solve complex tasks autonomously. The new system comprises seven components: the agent, skills, context, tools, rules, automation, and a learning loop. The agent, currently Claude, is treated as a commodity and is interchangeable with other platforms. Skills define work processes by specifying expected outcomes rather than step-by-step instructions, while context is stored in GitHub as Markdown files managed exclusively by the AI through daily 'context harvests' from Slack and emails. A traffic-light governance system dictates agent autonomy, requiring human approval for external actions like sending emails or invoices. The firm argues that treating AI as a single, versatile agent reduces complexity and eliminates handoff errors. However, the company noted that its initial version failed when a major offering change in early June caused a divergence between actual work and the AI's knowledge base, leading to a return to manual oversight. This experience highlighted the critical need for robust mechanisms to update existing context rather than just adding new information. Many companies still treat AI as a simple tool for specific tasks, leading CFOs to question the return on investment of six- to seven-figure AI platform expenditures. AI First contends that the primary barrier to effective implementation is not the technology itself, but the failure to maintain up-to-date context and skills.