Narada AI: Betting on a Future Beyond SaaS with Agentic AI

SAN FRANCISCO – In a bold declaration at TechCrunch Disrupt 2024, Narada AI, a startup born from UC Berkeley research, is challenging the very foundation of modern knowledge work. The company is pioneering a move away from the dominant Software-as-a-Service (SaaS) model towards a future powered by "agentic AI." Their core innovation, Large Action Models (LAMs), enables AI agents to autonomously reason through and execute complex, multi-step tasks across a suite of enterprise tools, promising to dismantle the digital silos and inefficiencies that define the current workplace. This new paradigm, according to CEO Dave Park, represents not just an improvement, but a fundamental revolution in how humans interact with software, potentially making the act of switching between dozens of applications a relic of the past.

The Tyranny of a Thousand Tabs

The modern knowledge worker's daily reality is one of digital fragmentation. A single business process, such as onboarding a new client, can require logging into Salesforce to update a record, creating a dedicated channel in Slack, setting up a project in Asana or Jira, generating an invoice through QuickBooks, and finally, drafting a welcome email in Gmail. Each step involves context-switching, navigating different user interfaces, and manually transferring information, leading to what many describe as "death by a thousand clicks" or "SaaS sprawl."

Research consistently shows that the average enterprise now uses over 100 different SaaS applications, with some larger corporations juggling several hundred. This proliferation, while offering specialized tools for every conceivable task, has created a new layer of friction. Workers spend a significant portion of their day not on deep, value-creating work, but on the meta-task of managing their tools. Data becomes siloed within these applications, making a holistic view of a project or customer journey difficult to achieve without cumbersome integrations or manual data entry.

"We've spent the last two decades building specialized applications for everything, which was a necessary step," explained Dave Park in a recent interview. "But the side effect is a massively fragmented user experience. The goal of software is to make us more productive, but we've reached a point of diminishing returns where managing the software itself has become a major part of the job. That's the core problem we are built to solve."

Enter the Agent: Large Action Models in Practice

Narada AI's answer to this fragmentation is the Large Action Model, or LAM. While built upon the same foundational principles as the Large Language Models (LLMs) that power technologies like ChatGPT, LAMs are specialized for a different purpose: taking action. Instead of just generating text or code, a LAM is designed to interact with graphical user interfaces (GUIs) in the same way a human would.

The process is built on three key capabilities:

• Reasoning: An agent powered by a LAM can understand a high-level, natural language command, such as, "Prepare the full onboarding package for our new client, Acme Corp."
• Planning: The agent then breaks this ambiguous goal into a concrete, logical sequence of actions. It understands that "onboarding" involves specific sub-tasks across different platforms, creating a step-by-step workflow on the fly.
• Execution: The agent then autonmously carries out the plan. It navigates to the required websites, logs in, clicks buttons, fills out forms, and copies and pastes information between applications to complete the end-to-end process.

One of Narada AI's most significant technical breakthroughs is its ability to operate across applications even when they lack official Application Programming Interfaces (APIs). Traditional automation tools like Zapier or Workato rely on developers to create structured API endpoints for services to communicate. This process can be brittle, limited, and requires constant maintenance. Narada's agents, by contrast, use a combination of computer vision and UI understanding to "see" a screen and interact with its elements, just as a person does. This allows them to work with almost any web-based application out of the box, including legacy systems or internal custom-built tools that would never have public APIs.

This approach makes the automation far more robust and adaptable. If a button on a website moves or changes color, a human can easily adapt. Narada AI aims for its agents to have a similar level of resilience, learning to navigate UI changes without needing to be explicitly reprogrammed.

A New Software Abstraction Layer

The vision Park and the Narada AI team are selling is a radical departure from the current application-centric model of computing. They envision a future where the AI agent becomes the primary user interface. Instead of opening ten different tabs, a user would simply open their Narada agent and issue commands. The agent becomes a universal orchestrator, a digital chief of staff that handles the tedious, cross-application busywork, freeing up human workers to focus on strategy, creativity, and high-level decision-making.

"The fundamental user interface is changing from a graphical user interface to a language user interface," Park stated. "You won't need to learn the intricacies of Salesforce's UI versus Asana's UI. You'll only need to know how to articulate your goal. The agent becomes the abstraction layer that sits on top of all your existing software."

While Narada AI's initial target is the enterprise market, where the pain of SaaS sprawl is most acute, the implications for smaller businesses and even solopreneurs are profound. A single founder or a small team could leverage an AI agent to automate sales, marketing, and administrative functions that would typically require a dedicated hire or expensive, complex automation setups.

The Agentic AI Gold Rush

Narada AI is not alone in identifying this opportunity. The concept of agentic AI is rapidly gaining momentum throughout the tech industry. The summary of Park's talk noted the recent emergence of over 70 startups specifically focused on building AI agents. This burgeoning ecosystem signals a broad consensus that this is the next major frontier in software.

Moreover, established tech giants are making significant moves in this direction. Grammarly, long known for its writing assistance tool, is expanding its ambitions toward a comprehensive AI work stack designed to understand context and assist across various work-related tasks. Companies like Microsoft with its Copilot and Google with its Duet AI are embedding agent-like capabilities directly into their productivity suites, seeking to become the central nervous system for workplace operations.

The rise of agents poses an existential question for traditional SaaS companies. If users begin interacting with software primarily through a third-party agent, the SaaS provider risks losing its direct relationship with the customer. Their product could be relegated to the status of a "headless" backend, a utility that runs in the background, with its brand and user interface becoming invisible. This could fundamentally alter the economics and competitive landscape of the entire software industry.

Challenges on the Horizon

Despite the immense promise, the path to a fully agentic future is fraught with challenges. The most significant hurdles are security and trust. For an AI agent to be effective, it needs access to sensitive company data and credentials for multiple systems. Enterprises will require ironclad guarantees that this access is secure, auditable, and protected from misuse or external threats.

Reliability is another critical concern. How an agent handles errors, unexpected UI changes, or network interruptions will determine its viability. An agent that fails 10% of the time could create more work than it saves. Building trust will be a slow process, likely starting with low-risk, repetitive tasks before companies are willing to hand over mission-critical business processes.

Nevertheless, the momentum is undeniable. The shift from manual software operation to autonomous agentic execution represents a potential leap in productivity on par with the invention of the spreadsheet or the advent of the internet itself. As showcased by pioneers like Narada AI, the future of work may involve less clicking and more commanding, as we learn to delegate the digital minutiae to our increasingly capable AI counterparts.

Source: https://techcrunch.com/podcast/saas-is-in-the-past-the-future-belongs-to-agents-says-narada-ais-ceo/