Naganand Murty is the founder and CEO of Electric Sheep Robotics.
The work in AI is now starting to turn its attention to what will be its most significant breakthrough: autonomous agents. Positioned as the final frontier, these entities, ranging from run-of-the-mill software bots to physical robots, promise a future where prosaic everyday tasks such as bookkeeping, marketing and sales, or physical tasks such as gardening or doing the dishes, are managed efficiently, autonomously, and with human levels of accuracy. However, this future requires a radical shift in both technical approach and business mindset.
Building Agents Is Fundamentally Different From Building Software
The key to realizing advanced autonomous agents is the unlocking of their ability to function and interact within the real world’s chaos and unpredictability. They need to navigate the “long tail” of corner cases—scenarios that deviate from the norm but hold critical learning opportunities. Such interaction isn’t about passive observation; it requires engagement at a granular level, where the agent perceives, acts and, perhaps most critically, comprehends the rationale behind these actions.
For instance, customer service bots need to experience thousands of real customer conversations to learn proper dialogue, while warehouse robots require extensive testing to master grasping irregularly shaped objects. Both of these will need to have developed an innate “world model” that allows them to function competently and fulfill the given task.
What’s equally important is that these models can only be developed and refined by allowing agents the ability to learn on the job.
This stage of development is where AI-product businesses often find themselves at an impasse. This is both due to the scarcity of such rich, interaction-based data and the lack of business models that allow for imperfect agents to improve over time.
Unlike data sitting on the internet, these invaluable experiential learnings are locked within industries’ daily processes and the human workforce’s tacit knowledge. Janet from accounting and Bob from janitorial services may have a lot more to teach these robots—more than any amount of Wikipedia cramming or YouTube binge-watching can provide.
Also, our current paradigm for evaluating products leads to unrealistic expectations of perfection from what are fundamentally learning machines.
Businesses Need To Become RL Schools
Historically, investments in AI have gravitated toward amassing computational power or acquiring datasets. However, the evolution of autonomous agents dictates a different kind of financial strategy. Businesses must now pivot toward investing in environments where these agents can interact, learn and incrementally improve—essentially, they need to become more like reinforcement learning (RL) schools.
However, a significant bottleneck arises from the conventional product-service binary in business models. Product companies strive to sell complete, polished AI solutions, whereas service companies (their customers) seek out-of-the-box efficiency, avoiding the teething issues of new technology.
This dichotomy creates an “activation barrier,” preventing the iterative improvement of these agents due to the lack of real-world interaction and feedback within companies wary of immediate impacts on their bottom line. The resulting stalemate benefits no one and preserves the status quo, which continues to clutter up the world with a ton of useless me-too solutions all fighting for the same share of enterprise or customer wallet.
Agents Are The Ultimate Product, And Businesses Will Need To Build Them, Not Buy Them
The true jump in productivity isn’t going to come from yet another SaaS tool—it will instead be unlocked by the development of these agents in the petri dish of real-world businesses. To realize this, a radical new approach is required: Companies must internally evolve, allowing these agents to learn on the job, much like apprentices.
This strategy involves acquiring or partnering with firms not just for assets or market share but for arenas to cultivate AI—spaces where autonomous agents work alongside human counterparts, learning from daily operations and human feedback.
Such an approach disrupts traditional operational processes and requires patient investment, likely affecting short-term profitability. However, it paves the way for a deeply integrated AI workforce capable of driving unprecedented long-term efficiency and innovation. This strategy represents not just a technological shift but a cultural one, where businesses evolve from being mere consumers of AI to active participants in its creation and maturing.
Conclusion
Rather than thinking of AI agents in terms of some omniscient AGI, we need to look past the FUD and marketing hype and see them for what they are: assistants who will learn alongside their human teachers.
It’s evident that transitioning to a world where autonomous agents are reliable, efficient and commonplace involves concerted efforts on multiple fronts. Businesses need to reimagine their strategies, moving beyond traditional investment and product utilization approaches to become crucibles where these advanced AIs are forged and refined.
Much like bringing up a toddler or training an intern or apprentice, the journey involves embracing technical enhancements, facilitating real-world interactions for these agents, rethinking data and computational infrastructure, and, most critically, adopting a long-term view that values iterative progress over immediate perfection.
This paradigm shift, although challenging, spells a future where businesses are more efficient, innovative and dynamically adaptable to the ever-evolving marketplace. The autonomous agents era is not just about technological advancement; it represents a fundamental evolution in how industries operate, innovate and, ultimately, thrive.
Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?
