From Apps to Agents: The Uncomfortable Truth About Enterprise Automation

The SaaS era is dying. Not slowly. Not gracefully. It's being replaced by something fundamentally different.

Most enterprises are still treating agentic AI like a new feature to bolt onto their existing tech stack. They're wrong. This isn't an upgrade—it's a paradigm extinction event.

Bhaskar Roy, Chief of AI Products and Solutions at Workato, recently laid bare the reality: companies today are predominantly stuck in what Gartner calls the "low agency section"—content creation, summarization, writing emails. The comfortable stuff. The stuff that doesn't really move the needle.

But here's the uncomfortable truth: The real value doesn't live in making your emails sound smarter.

The Agency Gap: Where Most Companies Are Actually Stuck

Gartner's research reveals a massive gap between current LLM-based assistants and full-fledged AI agents, and most organizations aren't even aware they're on the wrong side of it.

Think about your current "AI strategy." Are you:

• Using copilots to generate content?

• Implementing chatbots for customer queries?

• Automating simple, deterministic workflows?

Congratulations. You're building yesterday's solution for tomorrow's problems.

The enterprises that will dominate aren't asking "How can AI help my employees work faster?" They're asking: "How can agents execute our core business processes autonomously?"

Roy frames it bluntly: You need to shift from app-centric thinking to outcome-centric orchestration. Your finance team shouldn't be clicking through five applications to close the books. An agent should be orchestrating quote-to-cash across your entire stack—CRM, ERP, billing, provisioning—without human handholding.

Stop Automating Tasks. Start Automating Outcomes.

Here's where most companies get it catastrophically wrong: They're still thinking in workflows.

"I need this form filled out, then approved, then sent to finance."

Agents don't think in workflows. They think in outcomes.

Consider Roy's example: Instead of building a workflow for sales reps to generate quotes, the agent should:

1. Analyze opportunity status in your CRM

2. Review call recordings and deal documentation

3. Generate the quote based on actual context

4. Execute the entire order-to-cash process across sales, finance, and customer success

Zero clicks. Zero app-switching. Just outcomes.

The enterprise agentic AI market is projected to explode from $1.5 billion in 2025 to $41.8 billion by 2030. But most of that value will accrue to companies that fundamentally rethink their operations, not those that sprinkle AI features on their existing processes.

The Architecture Problem Nobody's Talking About

Your tech stack is a liability.

Every SaaS application you've purchased over the last decade was designed with a human operator in mind. Buttons to click. Forms to fill. Dashboards to monitor.

Agents don't need dashboards. They need APIs, event streams, and permissioned data access.

Bain & Company research shows that most organizations still lack the required ingestion pipelines for unstructured sources such as documents, emails, voice recordings, images, and videos—critical data sources for agentic reasoning.

Roy emphasizes three architectural imperatives:

Purpose-specific agents with defined capabilities. Not one super-agent that does everything. Multiple specialized agents that excel at specific domains—a finance agent, a supply chain agent, a compliance agent.

Enterprise skills that limit scope. Your marketing agent shouldn't have access to payroll data. Obvious? Apparently not, based on how most companies are approaching this.

Agent authentication that respects user permissions. If a sales rep can only access certain accounts in Salesforce, the agent acting on their behalf should have the same constraints.

Most enterprises will botch this. Gartner predicts over 40% of agentic AI projects will be canceled by the end of 2027 due to escalating costs, unclear business value, or inadequate risk controls.

Human-in-the-Loop Is a Bridge, Not a Destination

Here's the cognitive dissonance: Everyone says they want autonomous agents. Then they immediately build elaborate approval workflows.

Roy's perspective: Human-in-the-loop is necessary today because trust hasn't been established. But it's not the endgame.

The question isn't "Should agents operate autonomously?" It's "What needs to happen before we're comfortable removing the human?"

The answer: Governance that scales at agent speed.

Traditional governance—committee reviews, manual audits, quarterly compliance checks—moves at human speed. Agents operate at machine speed. The mismatch creates paralysis.

Enter what Roy calls "supervisor agents"—agents that monitor other agents. Compliance agents that audit finance agents. Security agents that observe how customer support agents handle PII.

Gartner predicts that guardian agent technologies will account for 10-15% of the agentic AI market by 2030, providing automated oversight and control for AI applications.

The Interface Is Dying (And That's Good)

Conversational UI isn't a feature. It's a burial rite for traditional software interfaces.

Roy's insight: Users are voting with their behavior. They don't want to learn your application's UI. They don't want to memorize where the export button lives or which dropdown contains the setting they need.

They want to say what they need and have it happen.

This doesn't mean chatbots. It means intent-based interfaces where users articulate outcomes and agents determine execution. The "how" becomes invisible.

For product teams, this is existential. You're not building screens anymore. You're building:

• Intent parsers that understand what users actually want

• Reasoning engines that determine optimal execution paths

• Orchestration layers that coordinate across systems

• Trust frameworks that let users verify without micromanaging

Your beautifully crafted UI? Your competitor's agent won't see it. It'll hit your API, extract value, and move on.

The Uncomfortable Playbook

If you're a business or tech leader, here's what Roy's decade at the forefront of integration and automation says you need to do:

1. Shift your mental model. Stop thinking in applications. Start thinking in outcomes that span multiple systems. What does "improve ROAS by 10%" actually require? Not one tool. Not one workflow. An orchestrated symphony of agents modifying ad copy, adjusting targeting, refining SDR sequences, analyzing results, and continuously optimizing.

2. Audit your tech stack for agent-readiness. How many of your critical systems have robust APIs? How much of your institutional knowledge lives in unstructured formats? How permissioned is your data access? Organizations will need to direct 5-10% of technology spending toward building foundational capabilities including agent platforms, communication protocols, and real-time data access.

3. Start with core processes, not peripheral conveniences. The companies winning with agents aren't generating better meeting summaries. They're automating procure-to-pay, hire-to-retire, and incident-to-resolution. Roy explicitly calls this out: Content creation and email writing won't move your business forward.

4. Get hands-on immediately. Don't study this. Build with it. Workato's internal "Workato on Workato" initiative—where employees across functions build agents for their own work—reveals how transformative this is when people actually use it.

The Reality Check

Gartner estimates only about 130 of the thousands of agentic AI vendors are real, with the majority engaging in "agent washing"—rebranding existing products without substantial agentic capabilities.

Most "AI solutions" you're being pitched? Glorified chatbots with a new PowerPoint deck.

The vendors that matter will show you:

• How their agents orchestrate across your actual tech stack

• How they handle permissioning and governance at scale

• How they enable purpose-specific agents with defined boundaries

• How they provide observability into agent decision-making

If they're leading with "conversational interface" and not "orchestration layer," walk away.

The Bottom Line

We're not optimizing the SaaS era. We're ending it.

The companies that thrive over the next five years won't have the most applications. They'll have the most sophisticated agent orchestration—with robust governance, purpose-specific capabilities, and outcome-driven execution.

Most enterprises aren't ready. Their architecture is wrong. Their mindset is wrong. Their vendor selection is wrong.

As IBM's research indicates, 2025 is being called the "year of the agent," with 99% of developers exploring or developing AI agents for enterprise. But adoption without transformation is just expensive theater.

The question isn't whether to build agents. It's whether you'll build them in a way that actually matters—targeting core business processes with proper architecture, governance, and orchestration—or whether you'll join the 40% of canceled projects that never made it past proof-of-concept.

The uncomfortable truth? Most of you are building the wrong thing, in the wrong way, for the wrong outcomes.

Time to fix that.