The great power – and responsibility – of agentic AI

Summary:

• Agentic AI differentiates itself from traditional generative AI by acting autonomously after initial setup, managing complex, high-volume workflows for productivity gains without constant human prompting.
• Despite its promise, skepticism persists due to AI agents’ lack of transparency, potential for cascading workflow errors, and difficulty integrating with complex systems. Experts emphasize rigorous real-world testing and modular design to ensure reliability and accountability.
• In sectors like compliance and financial crime prevention, AI agents are expected to transform efficiency by automating routine tasks, but human judgment and transparency remain vital. Leaders must embed governance, oversight, and ethical practices into every AI deployment.
• Get the lowdown on using AI for compliance and fraud prevention by signing up for Sumsub’s event.

If you haven’t heard of agentic AI, think of it as the futuristic cousin of platforms like ChatGPT and DeepSeek.

Maybe you need to automate the production and sending of multiple newsletters in a hurry. With generative AI, you have to prompt the platform and make tweaks as you go, but with an AI agent, you can watch the whole thing come together after the initial setup.

This ability to work autonomously is what sets AI agents apart from conventional AI products, explains Pavel Goldman-Kalaydin, head of artificial intelligence at global full-cycle verification platform Sumsub. By putting complex and high-volume processes into the hands of an AI agent, organizations can raise productivity levels without the added resources.

Pavel Goldman-Kalaydin, head of artificial intelligence at Sumsub / Photo credit: Sumsub

That said, a lot of people are cautious about trusting AI systems. In particular, AI agents are attracting some skepticism, with a Workday survey showing that over a quarter of executives believe the tech is “overhyped.”

Even Goldman-Kalaydin admits that there are plenty of kinks to work out before these technologies can enter the mainstream.

“There’s a big question as to whether AI agents can function reliably and accurately under stress,” he says. “Can they integrate smoothly within complex technical and regulatory environments?”

Another risk stems from the fact that AI models are “black boxes” whose inner workings are impenetrable to users, Goldman-Kalaydin points out. This makes it critical for companies to establish processes that ensure accountability and transparency before launching these autonomous models.

“A lack of explainability can lead to compliance challenges and erode trust,” he adds.

Rewiring the organization

Despite these sentiments, leaders in Asia Pacific believe that AI agents represent the future.

According to an IDC report, 70% of firms in the region expect the tech to disrupt their business models within the next 18 months as leaders look to make operations more efficient.

Case in point: AI agents are expected to become game changers in the world of compliance and financial crime investigations, a sector often stymied by burdensome documentation and tedious (but necessary) compliance checks.

Photo credit: Shutterstock

Consider how Sumsub leverages AI in anti-money laundering (AML) processes. During onboarding, the platform automatically checks documents against its database to flag any high-risk entities.

“By using AI to reduce false positives and streamline reporting, you end up with compliance teams spending less time on irrelevant matches so they can focus on combating real threats,” says Goldman-Kalaydin.

The company’s platform also integrates AI into its transaction monitoring services to enable organizations to preemptively identify suspicious patterns and respond to risks in real time. Its AI assistant – named Summy – gives investigators the full context of each case, covering user background, transaction history, behavioral insights, and other details while suggesting next steps to accelerate decision-making.

“Essentially, we’re turning down the alert noise so internal teams can streamline how they’re investigating each case,” adds Goldman-Kalaydin.

Performance at scale

But there are still hurdles to overcome. The real test is how the tech performs in dynamic, high-volume conditions, says Goldman-Kalaydin.

“Errors in any part of a multistep workflow can multiply quickly,” he points out. “If an agent has a 10% error rate per task, then over a 10-task workflow, the odds of something going wrong go up to 65%. That’s a dramatic drop in overall reliability.”

If those mistakes carry over into the testing or monitoring stages, even minor blunders can be costly. In one real-world case study, a bank developed and deployed a generative AI-powered data extraction capability to support its know-your-customer processes. Despite rigorous testing, the solution struggled to integrate with businesses with complex structures.

All you need to know about using AI for compliance starts here

To overcome this issue, Goldman-Kalaydin highlights the need to test these agents using real-world, high-volume scenarios – not just pilot environments – to unearth hidden errors.

“You’ve got to pair this with ongoing training and monitoring so your agents are up-to-date on new risks and edge cases,” he says. “It’s also critical to design decision flows in modular steps, with independent checkpoints and human oversight to quickly catch small errors before they cascade system-wide.”

‘Garbage in, garbage out’

A common issue faced by companies looking to adopt AI agents is whether these technologies can access the data and systems they need to be effective.

“Each enterprise platform often has its own data format, schema, and processing logic, making cross-platform integration difficult,” says Goldman-Kalaydin.

Additionally, discrepancies in data updates and disconnected operating systems can lead to companies ending up with conflicting or outdated information.

Photo credit: Shutterstock

“‘Garbage in, garbage out’ absolutely applies, because even the best agents depend on consistent, high-quality input data and training sets,” he explains. “Poor integration, fragmented systems, or low-quality data can result in missed alerts, data silos, and unreliable outputs, which undermines the value of automation and makes business operations less agile.”

By standardizing data formats and interfaces, businesses can reduce the risk of fragmentation by ensuring the whole organization is “speaking the same language,” he advises. Businesses should invest in a centralized system that can bridge the gap between different tools and enable “smooth data sharing.”

“That way, you’re able to maintain oversight, resolve conflicts, and ensure alerts and cases aren’t missed between disparate platforms,” he adds.

Embedding transparency

While the autonomous nature of agentic AI is the technology’s big selling point, it’s also a major source of risk. If an AI agent decides to go rogue, organizations must be prepared to shoulder the blame for its actions, says Goldman-Kalaydin, citing recent regulations in Singapore.

This is especially relevant for firms operating in highly regulated industries like banking and healthcare, where AI agents are accessing sensitive information and making high-stakes decisions.

These organizations need to maintain a high degree of accountability and transparency through clear, traceable, and audit-ready logs that teams can use to peel open AI’s black box.

Goldman-Kalaydin also suggests that firms opt for solutions that directly assign accountability for AI-driven actions to designated humans.

GET A HEADSTART AT SUMSUB’S WTF SUMMIT

 “You’ve got to embed these principles into how you practice governance daily so that, no matter how autonomous your AI agents become, your organization remains in control, transparent, and regulator-ready,” he explains.

Human-centric leaders

Over the next few years, Goldman-Kalaydin expects AI to become a core tool for fraud detection and financial crime prevention, especially as Asia Pacific remains one of the world’s most cyber-vulnerable regions.

“Network fraud is eight times more likely in Asia Pacific than elsewhere,” he points out. “Leaders need to invest in multilayered defenses against advanced fraud schemes.”

Photo credit: Shutterstock

Sumsub’s platform enables organizations to improve its fraud prevention defenses with an AI- and machine learning-powered solution that combats multi-accounting, detects fraud rings, and even preempts criminal activity before it happens.

The firm envisions AI agents as trusted co-pilots for risk and compliance teams, automating routine reviews and highlighting only the cases needing human attention to boost productivity and reduce risk. These agents should require minimal setup, map workflows to regulations, and deliver real-time insights while filtering out false AML alerts.

Ultimately, AI agents aim to let teams focus on high-value decisions with full transparency, efficiency, and regulatory confidence.

Still, as organizations “embrace AI,” Goldman-Kalaydin stresses the importance of maintaining “human judgement and oversight” at the core of their principles.

“You’ve got to make sure you’re prioritizing transparency and explainability in every deployment. That way, you make AI-driven decisions understood by all stakeholders – from internal teams to regulators – and ensure that AI is augmenting your values and expertise,” he says.

Sumsub is a global leader in verification technology that leverages the power of AI to provide organizations with scalable, automated compliance tools. In November 2025, Sumsub will be hosting its WTF Summit to share firsthand insights on combating AI-driven fraud with the latest innovations and technologies, and there will be a dedicated fireside chat session for AI agents titled “AI Agents – Assistant or Threat?”

To register for the event, click on this link.

GET A HEADSTART AT SUMSUB’S WTF SUMMIT

For a special 15% discount exclusive to TIA readers, click here.

This content was produced by Tech in Asia Studios, which connects brands with Asia’s tech community. Learn more about partnering with Tech in Asia Studios.