In early 2025, OpenAI CEO Sam Altman predicted that AI agents, that is, software systems that can plan, execute, and complete tasks without constant human prompts, would begin “joining the workforce” and materially transforming how companies operate. That idea set a hopeful tone across the technology industry and shaped expectations for the year.
This IEEE Spectrum article examines how that narrative played out in practice. The article makes clear that while there was undeniable progress with AI agents in 2025, opinions differ sharply about how far autonomous systems actually advanced and what they achieved. For some engineers and programmers, agents already deliver real productivity gains. Developers now embed agentic tools such as Cursor and Claude Code into their workflows, letting systems write, test, and debug code with minimal human intervention. This early adoption shows agents can outperform earlier tools on specific tasks.
At the same time, many organizations remain cautious or underwhelmed. Industry consultants and executives report that few companies have deployed AI agents in production environments, with most still in experimental or evaluation stages. The technology’s promise bumps up against practical concerns about reliability, accountability, and risk: errors, unclear responsibility for automated decisions, and incomplete integration into business processes all temper enthusiasm.
The article also highlights that these divergent experiences reflect broader tensions in the field. For programmers deeply familiar with integrated development environments and tooling, agents augment work more easily. In domains where human responsibility is paramount, such as medicine and design, hesitation remains high because the consequences of mistakes are serious.
Smith’s assessment suggests 2025 was a year of significant agentic experimentation and proof-of-concepts rather than widespread operational transformation. AI agents made progress, but their full integration into everyday work at scale still faces hurdles in reliability, governance, and trust.