Is SaaS Dead? Salesforce Leaders Think That’s a ‘Lazy Narrative’

As artificial intelligence rapidly shifts enterprise software from passive databases to active decision-makers, Srini Tallapragada, President and Chief Engineering and Customer Success Officer at Salesforce, and Arundhati Bhattacharya, President and CEO, Salesforce South Asia, believe the conversation must move beyond hype.

This is particularly relevant as the ‘SaaSpocalypse’ panic continues to grip the bourses. On January 30, Anthropic’s newly launched AI plugins led to a $285-billion wipeout from software, legal tech, and financial services stocks, as investors remain concerned over whether the tools will erode traditional SaaS advantages.

However, Tallapragada, in a conversation with AIM on the sidelines of the India AI Impact Summit, dismisses it as “a lazy narrative.” He notes that enterprise software hinges on “core and context.” Most companies will not rebuild mission-critical systems; they will rely on trusted platforms to handle complexity efficiently.

What ultimately matters is customer value. “As long as we continue to deliver value, we’ll be good,” he reassures.

Bhattacharya concurs. “You’re not just looking at a model,” she says. “You’re looking at workflows, governance, observability, context.” As businesses evolve, platforms must evolve alongside them. “We will do whatever it takes to add value.”

CRM Will Not Disappear

The future of CRM, they argue, is not about copilots assisting humans—it is about autonomous agents executing real work, at scale, within guardrails.

“Agents have to act,” Tallapragada says. “And for an agent to act, the first thing it has to know is the job to be done.”

That deceptively simple framing captures Salesforce’s broader strategy: redesign CRM from a system of record into what it calls a “system of work.” In this new architecture, agents don’t merely answer queries but perform specialised roles—whether as sales representatives in pharmaceuticals, service associates in airlines, or compliance assistants in financial services. 

The shift marks a structural change in how enterprise software is conceived, built and monetised.

Building Stack for Autonomous Action

An agent, he explains, requires four foundational layers.

First, it must understand its role in granular, industry-specific terms. “It has to know what a sales rep in the pharmaceutical industry does versus in financial services. They do different things,” he remarks. Agents need contextual understanding of workflows, regulations, and performance metrics specific to each vertical.

Second, it needs knowledge—and that is where enterprises often falter. Corporate data is scattered across systems, duplicated in multiple formats and buried in silos. “Sometimes there are 16 versions of a string in the system,” he says. “You have to consolidate that.”

Context, Tallapragada argues, will define the next competitive battleground in AI. “The more context you give, the better the answer,” he says. “Context is where the next battle will be.”

Third, agents must act. That requires deep API connectivity across enterprise systems, from credit checks and KYC verification to inventory look-ups and order fulfilment. “You need an API gateway like Mulesoft,” he adds, referring to the integration backbone that allows agents to execute workflows seamlessly across environments.

Fourth, and perhaps most critically, agents require guardrails. “You need to have guardrails: what the agent cannot do, not only what it can do,” he emphasises. Salesforce calls this its “trust layer”—governance, auditability, privacy controls, PII protection, data residency, and compliance frameworks embedded directly into the platform.

“You need to know later what the agent did and how it decided,” he says. “Without auditability, governance and privacy, it won’t work.”

All of these layers converge in Salesforce’s agentic platform, Agentforce, which represents a pivot from assistive AI to autonomous digital labour. According to Tallapragada, more than 18,500 customers have adopted Agentforce within a year, making it one of the company’s fastest-growing offerings.

When Agents Become Customers

Increasingly, enterprise leaders are looking at AI agents that transact not just with humans but with other AI agents. Tallapragada sees this less as futurism and more as protocol design. 

“An agent can delegate to another agent,” he explains. A shopping agent might query an inventory agent before confirming delivery timelines. In that exchange, “this agent is the customer of that agent.”

Such intra-company and inter-company agent communication will require identity frameworks, governance standards and traceability. Who can access what? How are permissions granted? How are decisions logged? These are architectural questions Salesforce is actively addressing.

Yet Tallapragada remains pragmatic about adoption. “Ninety-five percent of customers are in phase one,” he admits. Most enterprises are still experimenting with knowledge-based use cases in customer support before progressing toward multi-agent orchestration. “It’s crawl, walk, run.”

From Pilots to Production

That gap between experimentation and scaled deployment remains one of the industry’s biggest bottlenecks. “There’s a saying: everybody wants to write code, nobody wants to operate,” Tallapragada observes. 

Building a demo is easy; running resilient systems that survive real-world complexity is not. Production environments require logging, observability, hallucination mitigation, bias controls and robust governance. “Without a platform, it’s impossible to get real production value,” he says.

Bhattacharya offers a concrete example. In one ASEAN airline’s low-cost arm, customer frustration around flight changes had become a chronic pain point, hinting at AirAsia. By deploying agentic workflows, the airline achieved “95% accuracy” in handling change-of-flight requests within two months.

“You start with the use case that gives maximum trouble and maximum value,” she explains. “Once you’ve proved it works, you add the next one.” The airline is now expanding to additional use cases such as lost luggage and trip planning.

That incremental approach also explains why roughly half of Agentforce revenue now comes from what Salesforce calls “refilling the tank”, customers expanding usage after seeing measurable results.

Garbage In, Garbage Out

Generative AI’s promise, however, amplifies risk. What happens when flawed enterprise data feeds powerful models?

“AI is a data project,” Tallapragada says bluntly. “If you don’t have a good data foundation—garbage in, garbage out.”

Many enterprises still operate with what he describes as “dark data”—unstructured, uncatalogued information hidden across legacy systems. Cleaning, governance, lifecycle management, and accessibility are prerequisites for meaningful AI outcomes.

“If companies have invested in data maturity, our platform can access that,” he says. “If they haven’t, we help them get there. Otherwise, it will reinforce a bad experience.”

Where Accountability Sits

As AI agents begin making customer-facing decisions, accountability becomes a pressing issue. Should responsibility lie with model providers, enterprises, or platform vendors?

Tallapragada is unequivocal. “Model providers have no accountability,” he says. “If it gives bad output, they say it’s your problem.”

In enterprise settings, liability ultimately rests with the company serving the end customer. “Customers want a platform that helps them manage liability,” he adds. Salesforce’s trust layer is positioned as that safety net.

Bhattacharya extends the argument. Even if AI agents increasingly represent individuals in transactions, CRM does not become obsolete. “If I don’t have a CRM, I won’t know your requirements,” she says. “Even if it’s your agent interacting with me, behind that agent is still you.”

Understanding behaviour, preferences, and service history remains central.

Digital Labour

As AI agents proliferate, workplace dynamics may evolve dramatically. Tallapragada envisions a future where employees manage digital colleagues.

“Everybody will be a manager,” he proclaims. “Sometimes you’ll manage only agents. Sometimes humans and agents.”

Salesforce has built what he describes as an “agentic command centre,” enabling oversight of both human and digital workers. Agents must be monitored for drift, evaluated against KPIs and aligned with standard operating procedures.

Human-in-the-loop frameworks remain essential. On Salesforce’s own support portals, millions of service queries are now answered by agents, but complex cases escalate to human teams. Hybrid reasoning, combining generative creativity with rule-based compliance, ensures balance.

Pricing in Transition

As AI becomes native to SaaS platforms, pricing models are evolving as well. The traditional per-seat subscription may not fully capture agent-driven productivity.

Tallapragada describes a transitional phase. Salesforce continues offering per-user licences while introducing digital credits tied to agent consumption. “Each company is at a different maturity level,” he says. Flexibility is critical during transformation.

Importantly, AI adoption has not halted hiring. “Even AI companies are hiring aggressively,” he notes, suggesting augmentation rather than wholesale workforce replacement.

India’s Governance Balancing Act

In India, balancing speed of adoption with regulatory compliance is particularly crucial in sectors such as banking and public services.

Bhattacharya, who served as the Chairperson and MD of SBI for four years, believes these sectors are prime candidates for agentic deployment. “Banking and public services are areas where agents are very badly required,” she says. High-volume processes—address changes, nominee updates, document verification—can be automated securely.

“Verification of documents can be done very well by systems,” she explains. Automated triaging across trusted databases can accelerate turnaround times while freeing staff for more personalised engagement.

Rather than a sweeping transformation, she advocates a line-of-business approach. “Take up service requests first,” she says. “These are areas that will give maximum value.” Private sector banks are already moving; public sector institutions are proceeding more cautiously.

Narrowing the Digital Divide

Concerns that AI may widen the gap between digitally mature enterprises and smaller firms are, in Bhattacharya’s view, overstated. The real barrier is not cost but mindset. For small businesses with limited reach, AI can expand access and elevate service quality.

“For a small unit, the reach can become much bigger,” she says. “The quality of service you render can become much higher.”

But superficial adoption will not suffice. AI layered over broken processes yields little benefit. Holistic transformation is key.