From Deal‑Making Bots to Superhuman Cybersecurity: The New AI Agent Reality
AI agents in business are moving from slide decks to real transactions. Anthropic’s internal “Project Deal” created a private marketplace where software agents acted as both buyers and sellers. After brief interviews with employees, Claude-based agents negotiated, counter‑offered and closed 186 transactions over more than 500 listed items with no human in the loop mid‑deal, demonstrating that autonomous systems can already handle complex, natural‑language bargaining end‑to‑end. In cybersecurity, Anthropic’s Mythos AI shows the same acceleration on the defensive side. Designed for research, Mythos uncovered more than 2,000 previously unknown software vulnerabilities in just seven weeks, roughly 30% of the world’s annual pre‑AI output of similar flaws. Experts warn that this mix of speed and autonomy could let even non‑experts identify and exploit weaknesses in minutes. Together, these examples show both the transformative promise and systemic risk of powerful AI agents when they are allowed to act largely on their own.
Synthetic Audiences: How Virtual Consumers Could Reshape Consulting and Marketing
Consulting and marketing are being quietly disrupted by synthetic audiences consulting tools. Instead of waiting months and spending heavily to survey thousands of people, companies can now ask AI to simulate how specific consumers or invented personas might think, shop and vote. By feeding models with demographic or behavioural clues, synthetic audiences generate instant, low‑cost responses that mimic real segments, turning traditional polling and user research into a two‑minute exercise. Startups such as Electric Twin, Artificial Societies and Aaru, along with incumbents like Dentsu and WPP, are already fielding products that let teams “interview” these digital stand‑ins at scale. The efficiency gain is undeniable, but accuracy remains imperfect. These systems approximate human behaviour rather than measure it directly, which means genuine customer insight still depends on human experts who can design sound questions, interpret noisy outputs and decide when live research is non‑negotiable.
Humans in the Loop: Why Full Automation Still Falls Short
Despite rapid advances, research on humans in the loop AI suggests that replacing people outright is rarely optimal. A review of 90 studies from the University of East London found a consistent pattern: AI systems excel at scanning huge volumes of data, ranking options and spotting patterns, but they struggle with judgment, context and accountability. Software can rapidly surface likely matches or anomalies, yet it cannot reliably decide whether a recommendation fits local constraints, social norms or strategic priorities. The study frames this as a “knowledge ecosystem,” where AI supplies signals and humans supply meaning. Leaving interpretation entirely to software often produces neat answers that fail in messy real‑world conditions. New HR‑focused platforms, such as Windmill’s “context graph” for people, start from the same premise: as AI takes over routine analysis, the uniquely human skills of sense‑making, coaching and role design become more important, not less, for workforce performance.
AI in Boardrooms, Hiring Pipelines and Classrooms
AI agents are quickly entering executive and customer‑facing roles. One U.S. bank CEO even let his AI clone conduct an earnings call with analysts, and the institution is now deepening its collaboration with OpenAI to automate more finance workflows, signalling that virtual executives and service agents will increasingly front external conversations. Hiring has also become a two‑sided AI game: employers use algorithms to screen CVs, while candidates rely on generative tools to tailor applications, extending a century‑long tradition of using tests and scoring systems to match people to jobs. In AI in higher education, usage patterns are uneven. A Virginia Tech study found engineering and computer science students integrate generative AI into coursework far more than humanities and social sciences peers, while many students remain unsure about when AI use is appropriate. Universities now face pressure to define basic AI proficiency across all majors, not just technical fields.
AI‑Proof Careers? What Malaysian Workers and Students Should Really Focus On
Fears about automation are driving students to hunt for AI proof careers. In one U.S. example, a business analytics student switched to marketing after realising that many entry‑level analytics tasks can be automated, aiming instead to strengthen critical thinking and interpersonal skills that AI struggles to replicate. Polls show most students view AI as a threat but also believe universities must teach them how to use it. For Malaysian workers and students, the practical takeaway is not to avoid AI, but to pair it with differentiating human capabilities. Valuable skills include domain expertise (finance, logistics, healthcare), communication and relationship‑building, ethical reasoning, problem framing and the ability to supervise AI agents in business workflows. Learn enough AI literacy to prompt, evaluate and combine tools, but invest deeply in judgment, collaboration and adaptability. Those who can manage, question and complement AI systems will be hardest to replace, whatever their industry.
