Nearly every company wants to experiment with AI agents, and rightly so. Agentic AI comes with the potential to uplevel your operations and customer service in previously unheard-of ways.
The possibilities are endless – agents that can kickoff back-office processes independently, agents that can find the exact right document from massive knowledge bases instantly, agents that can scan PDFs and extract necessary data, eliminating the need for a human to endlessly cut and paste – but the question of how to manage the risks of an agentic workforce loom large. For enterprises, this is a key concern; you can’t take advantage of the benefits if you haven’t first addressed the potential for failure.
Agentic systems in customer service, for example, can plan and reason based on customer requests – an enormous benefit to productivity. You can’t, however, allow AI agents to plan and reason new ways to serve millions of customers, each in a slightly different way, following a different path, with different rules. The result would be chaos.
And that chaos has come to pass in real-world scenarios. A quick search for "enterprise AI failures” pulls up numerous examples of gaffes that have caused reputational and financial damage at major companies, including examples like these:
- At Air Canada, a customer service AI agent incorrectly told a customer he was entitled to a refund – except that refund did not exist. When the airline attempted to deny responsibility for the AI agent’s promise, the customer took legal action, resulting in regulatory scrutiny and reputational damage.
- McDonald’s invested millions in an AI-powered ordering system, only to have the system completely misinterpret customer requests – resulting in bizarre orders being fulfilled, including hundreds of nuggets and bacon-topped ice cream sundaes. Unsurprisingly, the initiative was scrapped.
- At Amazon, recruiters were asked to use an AI-powered hiring tool to surface the best resumes for open recs. The AI model, however, began discriminating against women, refusing to surface their resumes and prioritizing male resumes instead. The project was cancelled, but only after raising serious concerns about bias in automated decision-making.
For each of these examples, the AI agent or model in question lacked governance and orchestration, creating significant business risks and casting doubt on the potential for success with future initiatives.
Overcoming challenges with enterprise AI agents: What’s holding companies back?
For many operations leaders, the challenge with agentic initiatives is launching them in a way that aligns with the needs of complex enterprises.
There are some common enterprise challenges that hold AI initiatives back, including:
- Data locked in legacy systems. AI agents need real-time data; without it, AI-driven decisions can be inaccurate.
- Compliance requirements. Without auditability and governance, AI deployments can violate critical industry regulations, leading to fines or lawsuits.
- An “experimental” mindset. Many enterprises are jumping into the fun part – using AI – without a structured rollout strategy, leading to multi-million-dollar projects that inevitably fail.
- Need for oversight. Enterprise operations demand consistency above all else. Deploying AI agents in siloed pockets of organizations creates fragmentation and unpredictability – the exact opposite of what optimized operations require.
The alternative: AI agents that work with – not against – your existing enterprise operational strategy
Rather than investing in isolated agentic initiatives, operations leaders should consider the value of an orchestrated approach to AI agents.
In an orchestrated, governed environment, AI agents:
- Adhere to enterprise regulations from day one
- Execute the same set of actions consistently
- Work smoothly with existing people and systems
- Only act within a controlled framework, ensuring full visibility into all actions and decisions
5 common agentic AI deployment mistakes and how to fix them
As many enterprises seek to invest in AI agents, they’re finding they need to start with the right foundational strategy. Without it, a number of things can go wrong. Here are five common deployment mistakes, complete with the best way to avoid or fix them.
- Starting from scratch
The mistake: Enterprises are spending tens of millions on standalone AI initiatives that ultimately fail due to lack of integration with existing systems.
The fix: Rather than creating new agents from scratch, enterprises need platforms that enable them to turn their existing workflows into agents while benefiting from a proven, trusted set of action steps and boundaries for AI to adhere to.
- Assuming agents don’t need context
The mistake: Agents that lack the context of organizational rules can act in ways that violate policies and diverge from standard workflows.
The fix: Find a platform that can fuel AI agents with enterprise knowledge and process context – for example, organizational structure, business logic, standard operating procedures – providing a structure to operate within that offers structured guidance, governance, and standards.
- Over-focusing on agent skills
The mistake: When you treat AI agents as individual tools and over-focus on their individual features, you stray from focusing on what really matters: how they can drive business results.
The fix: Keep your focus on outcomes and the end-to-end customer journey. And remember that regardless of an AI agent’s skills, agentic projects won’t succeed without a structured approach to orchestrating people and technology.
- Feeding AI low-quality data
The mistake: Without enterprise-wide access to real-time data, AI agents can make flawed decisions.
The fix: Leverage an agentic platform which brings together end-to-end data from across the customer journey – transaction data, customer data, documents, history, and more – to make informed decisions and take relevant actions.
- Lacking a framework for governance
The mistake: When you ignore orchestration and governance at the outset, AI agents become siloed tools that create vulnerabilities for your organization.
The fix: Build agents on a centralized platform that can provide auditing, management, and orchestration across agents, ensuring they work together rather than in isolation.
Orchestrate AI agents with Pega AgentX
AI agents have the potential to change the way business is done, but success at the enterprise level depends on structured orchestration.
Pega's new AgentX API combines the best of AI and workflows, enabling transformative outcomes with enterprise-ready governance. AgentX turns any workflow into a conversational interface automatically, enabling customers with 24/7 self-service help and empowering employees with the workflows they need access to through simple, natural language.
With AI and workflows together, enterprises have a strategic approach that ensures agents are embedded seamlessly into existing business processes and agents’ actions are governed and transparent.
Try Pega AgentX for yourself today, in Pega GenAI Blueprint: www.pega.com/blueprint