As we dive into the world of AI and machine learning, we’re thrilled to sit down with Chloe Maraina, a Business Intelligence expert with a deep passion for crafting compelling visual stories through big data analysis. With her sharp expertise in data science and a forward-thinking vision for data management and integration, Chloe brings a unique perspective on the challenges and opportunities of building AI agents for enterprise applications. Today, we’ll explore the hurdles of moving AI from prototype to production, the critical skills and team dynamics needed for success, and the organizational support required to make these initiatives thrive.
How do you see companies struggling to transition AI agents from impressive demos to real-world production environments?
The biggest struggle I’ve noticed is the gap between a flashy demo and a reliable production system. Demos are easy to whip up—developers can create something that wows in a controlled setting within days. But when you move to production, you’re dealing with real-world variability and edge cases that the demo never accounted for. Most teams aren’t prepared for the sheer volume of work—about 90% of the effort happens after the demo. They often lack the expertise to handle the probabilistic nature of AI, where the same input might yield different outputs, unlike the predictable software they’re used to.
What makes AI agents so different from traditional software, and why does this catch teams off guard?
AI agents operate on a probabilistic model, which is a complete shift from the deterministic systems traditional software developers know. In regular software, input A always leads to output B. With AI, you might get a range of outputs for the same input, and that unpredictability throws teams for a loop. They’re not used to building systems where “working” means achieving a certain reliability rate—say, 85%—across thousands of scenarios. Without systematic evaluation frameworks, they can’t even measure if their agent is production-ready or just a cool experiment.
What core skills do you think a team needs to successfully build and deploy AI agents?
First and foremost, you need strong machine learning expertise. An AI/ML engineer who understands non-deterministic systems and can build robust evaluation frameworks is critical. Beyond that, backend and frontend developers are essential for the surrounding infrastructure—think containerized systems and user interfaces. But there’s also a newer skill, context engineering, which every engineer should learn. It’s about designing the entire information environment for the AI, from prompts to data retrieval systems. Finally, domain knowledge is non-negotiable—someone who gets the business context can shape how the agent solves real problems.
Can you explain what context engineering involves and why it’s become so vital for AI projects?
Context engineering goes beyond just crafting prompts; it’s about creating the whole ecosystem that guides an AI agent’s decision-making. This includes the prompts, sure, but also the logic for selecting tools, the data it pulls from, and the reasoning frameworks it uses. It’s vital because AI doesn’t just need instructions—it needs a reliable structure to think through problems across countless variations. Without this, agents fail unpredictably in production, even if they shine in demos. It’s like teaching the agent how to think, not just what to say, and every engineer needs to grasp this to build scalable solutions.
Why is it so important to have team members who understand the specific business or industry the AI agent is serving?
Domain expertise is a game-changer. A finance expert building an accounting agent will outshine a brilliant AI engineer who doesn’t know the ins and outs of debits and credits. It’s not just about reviewing the final product; these experts need to be involved from the start, designing how the agent approaches problems and what tools it uses. I’ve seen projects fail because the team didn’t understand the nuances of the workflow they were automating—resulting in agents that solved the wrong problems or created more work than they saved.
What kind of mindset or personality traits do you look for in a team working on AI agents?
I look for people who thrive in ambiguity and aren’t fazed by failure. Building AI agents isn’t like traditional software development—there’s no set playbook. You might spend weeks figuring out why an agent worked one day and flopped the next. The best team members are “learn-it-alls” who get excited by unsolved puzzles. They’re tinkerers, eager to experiment, fail, and try again without getting discouraged. That iterative mindset is what separates teams that succeed from those that get stuck.
How critical is support from other parts of the organization when developing AI agents?
It’s absolutely essential. Even the best AI team will hit roadblocks without buy-in from legal, security, and business stakeholders. I’ve seen projects stall because security teams blocked API integrations or business users didn’t provide feedback during the messy early stages when the agent fails half the time. Unlike traditional software, AI agents need constant collaboration with the people whose work they’re automating. Legal teams must adapt to new AI policies, and executives need to provide cover for early setbacks, treating them as learning opportunities rather than failures.
What role does leadership play in ensuring AI agent projects don’t get derailed by internal challenges or early failures?
Leadership is the glue that holds these projects together. They need to clear internal roadblocks, whether it’s securing data access or aligning security protocols with AI needs. More importantly, they set the tone for handling failure. Early versions of AI agents will mess up—that’s a given. Leaders who view these missteps as part of the learning curve, rather than reasons to pull the plug, empower teams to keep iterating. Without that executive air cover, projects can lose momentum or get shelved before they even have a chance to deliver value.
How can organizations effectively upskill their existing teams to handle AI agent development?
Upskilling doesn’t mean starting from scratch or relying on outdated formal courses. I’ve seen teams transform in just 3-6 months by focusing on hands-on learning. Bring in practitioners who are already building in your domain to run internal workshops or hackathons. Let your developers and data engineers experiment with real prototypes that tie into your systems. Data engineers often transition well to ML roles, and software engineers can become great agent builders with the right support. The key is to ditch the PowerPoint mindset and learn by doing—create something tangible while you’re at it.
What’s your forecast for the future of AI agent development in enterprise settings?
I see AI agent development becoming a core competency for enterprises over the next few years, but the pace of change will keep challenging even the best teams. New frameworks and models will continue to emerge monthly, pushing the boundaries of what agents can do. Organizations that build a culture of constant learning—through weekly reviews, shared documentation, or multiple small teams tackling different problems—will stay ahead. The talent equation will remain the biggest differentiator; those who invest in the right people and skills now will ship agents that deliver real ROI, while others risk watching competitors pull ahead.
