I went to an AI conference and got a crash course in middle management

The landscape of artificial intelligence is evolving, and at the recent AI Engineer conference in London, industry leaders gathered to discuss a pivotal shift in how we interact with AI technology. Presentations from major players like Google, Anthropic, and OpenAI illuminated the growing role of humans as managers of AI agents, emphasizing the necessity of guidance and oversight in this emerging era. One of the key takeaways was the transformation in the role of software engineers. OpenAI's Ryan Lopopolo highlighted that by late 2025, the responsibilities of these professionals will focus more on directing AI agents rather than traditional coding. This shift raises essential questions: How much autonomy should we grant to these agents? What form should they take? And is human language sufficient for effectively communicating with them? As the conference unfolded, it became apparent that the discussions echoed a business school curriculum tailored for the future of AI. Concepts such as "guardrails" and "context engineering" were prevalent, reflecting the industry's recognition that 2026 could mark a turning point, moving AI from experimental stages to practical applications. Anthropic's David Soria Parra expressed optimism about the next year, suggesting that 2026 will be crucial for deploying agents across various sectors, including financial analysis and marketing. In this envisioned future, AI agents could handle routine tasks, but human oversight remains essential to ensure they operate within established boundaries. An intriguing irony surfaced during the conference: as tech giants like Meta, Google, and Amazon streamline management structures, they may inadvertently transform all employees into AI supervisors. Tech contributors, who once focused solely on coding, are now tasked with overseeing AI-generated work. The potential pitfalls of granting too much control to agents were also a hot topic. Recent incidents, such as a disruption at Amazon linked to an AI coding assistant, underscored the importance of caution. Mario Zechner, creator of the coding agent Pi, advised a careful approach, recommending that software engineers utilize AI tools judiciously and maintain decision-making authority. The design and visibility of AI agents were further explored, with Ido Salomon from Monday.com showcasing Agentcraft, a program that visually represents agents in a gaming-like environment. This tool allows users to manage multiple agents effectively, minimizing the risk of conflicts when they operate simultaneously. Attendees from diverse backgrounds, including Yann Mainier from Sky UK, expressed a desire to learn how to enhance their agents' performance. Mainier noted the challenges of evaluating AI outputs, emphasizing the need for alternative methods of assessment, given that agents may not produce uniform results. The conference also highlighted the necessity for a rethinking of web architecture to accommodate AI. Vercel's CTO, Malte Ubl, pointed out that agents have become significant users of web content, prompting new considerations for feature proposals and interfaces to ensure compatibility with AI operations. As the dialogue around AI management continues, it's clear that the future will demand a new set of skills and strategies to effectively integrate these powerful tools into the workforce.