AI Agents: Progress, Hype, and Consumer Hurdles Ahead

This is The Stepback, a weekly newsletter that analyzes a key story in the tech world. For more on developments in AI, follow Hayden Field. The Stepback is delivered to our subscribers’ inboxes at 8AM ET. Sign up for The Stepback here.

The journey of AI agents can be traced back to pop culture icons like J.A.R.V.I.S. from Marvel’s Iron Man films.

Browse Posts

OpenAI to Add Safeguards After Teen’s Tragic Death

Microsoft’s Copilot AI Takes Center Stage on TVs!

Taco Bell Rethinks AI Drive-Thru After Customer Backlash

Although J.A.R.V.I.S. isn’t the sole origin, the character has inspired discussions around AI agents and their potential functions. Industry professionals often reference J.A.R.V.I.S. as a model for the perfect AI tool—one that anticipates user needs, processes extensive data, and provides strategic business insights. While definitions of AI agents vary, at their essence, these systems are designed to undertake complicated tasks autonomously without constant user interaction, creating a prioritized list of actions to achieve a desired outcome. Despite the enchanting prospect of such agents, practical limitations remain for the average user.

The term “AI agent” gained significant traction in the tech sphere in 2023, a year that marked the conceptual emergence of AI agents. Many were captivated by the idea, although concrete applications were scarce at that time. The real push towards practical implementation began in 2024, when companies started testing their code in live environments, though outcomes were often disappointing and riddled with technical issues.

A surge of excitement in the AI community can be traced back to a pivotal announcement in February 2024. Klarna, a fintech company, revealed that its AI assistant, backed by OpenAI technology, had effectively accommodated the workload of approximately 700 customer service representatives within its first month, handling two-thirds of incoming support requests. This statistic quickly became a staple in AI discussions.

The excitement surrounding AI agents persisted, with executives from major tech companies like Amazon, Meta, Google, and Microsoft championing the concept during their earnings calls. These leaders committed to developing effective AI tools, dedicating resources towards their realization.

The aspirational vision for AI agents suggests they could manage tasks ranging from travel booking to creating presentation visuals. Ideally, these agents would streamline social gatherings by cross-referencing various calendars, dietary needs, and preferences to secure dinner reservations and schedule events.

Historically, AI coding has bolstered the agentic AI landscape. When defining successful current applications of AI agents, practitioners typically reference AI coding as the primary example. Many software developers rely on AI-driven tools for coding tasks, yielding positive results. Reports indicate that at companies like Microsoft and Google, AI is responsible for generating up to 30% of their code, making AI coding tools a significant revenue source for startups like OpenAI and Anthropic.

Up until now, AI coding has dominated real-world applications of AI agents, which limits broader consumer engagement. The initial vision aimed for a more generalized AI agent accessible to all, a goal that remains partially unrealized. Nonetheless, progress has been made, especially by 2025.

In October 2024, Anthropic launched “Computer Use,” a feature enabling its AI Claude to navigate the internet similarly to a human, completing various tasks autonomously. Feedback indicated that while the launch represented a technological advancement, the tool’s performance left much to be desired. By January 2025, OpenAI introduced Operator, designed to handle tasks like form completion and ordering groceries, but it too faced critiques for sluggishness and inefficiency. Following this, OpenAI released Deep Research, an agent capable of generating extensive research reports, garnering mixed reviews on its utility. Later in July, OpenAI combined the functionalities of Deep Research and Operator to create the ChatGPT Agent—a notable upgrade, yet still prone to difficulties in real-world application.

While the path to achieving the most sophisticated AI agents is still lengthy, current developments signify a notable leap forward. This progress has prompted tech firms to amplify their investments in agentic AI through enhanced computational resources, research initiatives, and talent acquisition. Recent hires at Google—including executives from Windsurf—aim to advance its AI agent initiatives, as competitors like Anthropic and OpenAI rapidly innovate and release updates for consumer application.

Looking ahead, the field is poised for improvements in AI coding, which may overshadow entry-level software engineering roles. Enhancements in consumer-oriented AI products are anticipated, although the pace may be gradual. We can also expect increased usage of these agents for enterprise-level and governmental functions, particularly as dedicated platforms have emerged from Anthropic, OpenAI, and xAI targeting governmental needs.

As competition heats up in the AI agent landscape, the sector will likely encounter numerous challenges, including potential mergers and acquisitions. A central question for consumers and developers alike will be defining the role they envision for AI agents: Should they merely handle logistical tasks or also engage with more personal facets of life, such as drafting speeches for weddings or writing heartfelt notes? Assessments of their current effectiveness indicate limitations, particularly in more nuanced applications.