Amazon urges staff to treat AI agents as teammates, not workplace tools

Amazon is asking employees and customers to think of artificial intelligence systems as “teammates” rather than tools, arguing that the next wave of productivity will come from AI agents that understand context and act autonomously across workflows.

At AWS re:Invent in Las Vegas, senior executives said companies would unlock far greater efficiencies once workers stopped switching between fragmented systems and instead relied on a single, intelligent agent capable of orchestrating data, applications and tasks. The Wall Street Journal reported that Pasquale DeMaio, vice-president and general manager of Amazon Connect, told attendees the aim was to make AI “human-centred” and integrated into daily work rather than treated as a bolt-on utility.

DeMaio acknowledged, however, that most companies are still struggling with implementation. He said Amazon was hearing that about 42% of AI projects fail, while suggesting other estimates run higher. But where integrations succeed, he said, organisations were observing dramatic improvements, citing a 70% boost in work completion in some deployments.

Speakers described the reality of “AI sprawl”, with individual teams deploying isolated solutions that create yet more systems to manage. Jose Kunnackal John, director of Amazon Quick Suite for AWS, said many enterprises now juggle dozens of unconnected assistants, each with its own login, governance structure and data silo. He argued that Amazon’s Quick Suite provides a unified workspace to resolve that fragmentation by allowing a single agent to pull, analyse and act on information drawn from sources such as Microsoft SharePoint, ticketing platforms and email systems.

Kunnackal John said such an agent could respond to a request like a “Q3 performance update” by assembling documents, extracting operational data, referencing calendars and meetings, and producing a complete slide deck before packaging it into an email for distribution. Crucially, he added, this kind of conversational request could be converted into a reusable workflow and automated on schedule or on demand.

Enterprise adopters highlighted similar gains. AstraZeneca’s senior director of R&D IT, Vaishali Goyal, said clinical researchers had previously spent substantial time gathering research signals. Quick Suite allowed the company to automate this work, delivering “concise and consistent” insights and freeing staff for higher-value tasks. The system has been piloted in haematology, with broader deployment planned for 2026.

Despite the enthusiasm, Amazon executives repeatedly stressed the importance of guardrails. DeMaio warned against letting AI “take over” work or generate low-quality output, saying the aim was to enhance rather than dilute performance. He said internal research showed that when humans and AI collaborate effectively, teams see a 25% improvement in task completion and a 40% increase in work quality.

Speakers framed the challenge as a balancing act: while generative AI promises faster decision cycles and significant productivity uplift, execution remains complex, with risk of failure high. “If you build AI as something that’s amplifying human capabilities,” DeMaio said, companies can remove “mundane, boring” tasks and refocus people on meaningful work.

As AWS pushes deeper into enterprise automation, the question for customers will be whether organisations can move from isolated experiments to scalable systems — and whether employees embrace the idea of AI as a colleague rather than a competing force.