Following his appearance on “AI for Energy, Energy for AI” at the International Downstream Conference in Bahrain, Beyond Limits CEO AJ Abdallat sat down with us to discuss what the surge in AI demand really means for energy systems, how operators can move from trials to scale, and why neuro-symbolic, agentic AI is the right fit for complex downstream environments.
“This is the defining challenge of our time. AI demand is rising fast. The winners will be the companies that deploy intelligence that can explain itself, reason about constraints, and deliver more value with less compute.”
IDCE 2025 was remarkable in how it demonstrated the industry's readiness to embrace AI at scale. What struck me most was the shift in conversation. We're no longer debating whether AI will impact downstream operations, but rather how quickly and effectively we can implement it responsibly. The level of engagement during our "AI for Energy" panel was extraordinary. Operators, technology providers, and policymakers were asking very specific, practical questions about deployment, governance, and measurable outcomes.
The conference theme, "Innovating Energy, Creating Sustainable Value," really captured the dual imperative facing our industry. We need to meet growing global energy demand while simultaneously advancing sustainability goals. AI isn't just a nice-to-have technology anymore. It's becoming essential infrastructure for navigating this complex landscape.
The transformation has been dramatic. When we started Beyond Limits in 2014, drawing on our NASA and Caltech heritage, there was significant skepticism about AI in industrial settings. The energy sector, rightfully so, is conservative. Safety, reliability, and regulatory compliance are paramount. But over the past decade, we've seen a fundamental shift.
The industry has moved from experimental pilots to mission-critical deployments. What's driving this change is the convergence of several factors: the availability of vast amounts of operational data, advances in computing power, and most importantly, the development of AI systems that are transparent and explainable, what we call Hybrid AI.
Traditional AI, particularly machine learning models, often operate as "black boxes." They provide recommendations or predictions, but they can't explain their reasoning. In high-stakes environments like refineries or petrochemical plants, that's simply not acceptable. Operators need to understand why the AI is making a particular recommendation before they can trust it and act on it. Our Hybrid AI approach, rooted in the technology that powered Mars Rovers and deep space missions, combines the pattern recognition capabilities of machine learning with symbolic reasoning systems that can explain their logic. This creates an audit trail that operators can follow, understand, and validate against their own expertise.
By Hybrid AI we mean the deliberate integration of machine learning with symbolic reasoning. The machine learning layer detects complex patterns across massive datasets, while the symbolic reasoning layer applies logic, constraints, and expert knowledge to explain and validate outcomes. Neuro-symbolic AI is the technical category in which this approach sits. Agentic AI builds upon it, deploying reasoning agents that can act autonomously within defined safety and operational boundaries while maintaining a full audit trail. This distinction clarifies that Hybrid AI is Beyond Limits' architecture, neuro-symbolic is the class of methods it employs, and agentic AI is the operational mode it enables.
The reasoning engines that guided spacecraft millions of miles from Earth without human intervention are the same foundation Beyond Limits uses today in downstream operations. NASA JPL and Caltech developed these systems to diagnose faults, adapt to unexpected conditions, and ensure mission success under extreme uncertainty. That same proven reliability is now applied to refineries and petrochemical plants, where safety, compliance, and resilience are equally non-negotiable. This continuity demonstrates that Beyond Limits' AI is not theoretical, it is already field-tested in the most unforgiving environments.
Think about it this way: if an AI system recommends adjusting a process parameter that could impact safety or production, the operator needs to know not just what to do, but why. Our systems can say, "Based on these sensor readings, combined with historical patterns and process knowledge, here's the recommendation and here's the reasoning behind it."
Responsible AI adoption starts with recognizing that AI is not about replacing human expertise. It's about augmenting it. The most successful implementations I've seen are those where AI serves as an intelligent advisor, providing operators with enhanced insights and recommendations while keeping humans firmly in control of critical decisions.
Responsible AI means that every recommendation is validated in runtime, every inference is traceable, and every action maintains human-in-the-loop integrity. Beyond Limits' architecture provides an audit trail that records sensor data inputs, symbolic rules applied, and the reasoning path followed, ensuring full transparency for regulators and operators. This governance-ready design distinguishes Beyond Limits from black-box models, enabling compliance assurance and building institutional trust in AI-driven decision support.
This is particularly important in our industry because of the concept we call "agentic autonomous operations AI." Rather than pursuing fully automated systems, we're developing AI that can operate autonomously within defined parameters while maintaining meaningful human oversight. It's about creating intelligent systems that can handle routine optimization and monitoring tasks, freeing up human experts to focus on strategic decisions and exception handling.
Agentic autonomy differs from conventional automation by reasoning about context and adapting in real time. For example, in a downstream facility an AI agent might detect signs of heat exchanger fouling. Instead of raising a generic alarm, it could autonomously rebalance flow, adjust schedules to reduce strain on related equipment, and notify operators with a transparent explanation of its actions. Strategic decisions such as rescheduling production runs remain in human hands, but the agent manages the continuous optimization cycle. This balance of autonomous action and human authority is what defines agentic AI.
From a governance perspective, responsible adoption means establishing clear frameworks for AI deployment, ensuring transparency in decision-making processes, and maintaining rigorous testing and validation protocols. Companies need to develop internal AI expertise rather than simply outsourcing everything to external providers.
That's a critical point. While external AI providers can offer valuable technology and expertise, energy companies cannot afford to treat AI as a complete black box service. The stakes are too high, and the operational context is too specific.
Successful AI implementation requires deep domain knowledge, understanding the nuances of refining processes, the interdependence between different systems, and the regulatory environment. This knowledge can't be outsourced. Companies need to build internal capabilities to understand, validate, and govern their AI systems.
We've seen this in our partnerships. The most successful deployments happen when our clients have dedicated teams that work closely with us, learning the technology and developing the expertise to manage it long-term. It's a collaborative approach that combines our AI capabilities with their operational expertise.
Predictive maintenance continues to be a major driver, and for good reason. The ability to anticipate equipment failures before they happen delivers immediate, measurable value: reduced downtime, lower maintenance costs, improved safety. We're seeing companies achieve 20-30% reductions in unplanned outages through AI-powered predictive maintenance.
But the real excitement is in real-time process optimization. Modern refineries and petrochemical plants generate enormous amounts of data from thousands of sensors. AI can analyze this data continuously, identifying optimization opportunities that human operators simply couldn't detect manually. We're talking about improvements in yield, energy efficiency, and product quality that can translate to millions of dollars in annual value for a single facility.
AI is becoming a critical enabler of decarbonization efforts. Through intelligent energy management, we're helping companies reduce their carbon footprint while maintaining operational efficiency. AI can optimize energy consumption in real-time, identify opportunities for waste heat recovery, and improve the efficiency of carbon capture systems.
Recent studies suggest that AI-enabled systems could potentially reduce carbon emissions in the petrochemical industry by up to 20% through process optimization alone. But beyond direct emissions reduction, AI is also accelerating the development of sustainable products and processes. Machine learning can expedite the development of new catalysts, optimize the production of renewable fuels, and improve the efficiency of recycling processes.
Partnerships are absolutely essential. Our collaboration with Aramco, BP, NVIDIA, Microsoft and Google, for example, has been instrumental in bringing advanced AI computing capabilities to the energy sector. But it's not just about technology partnerships. We need collaboration across the entire ecosystem.
The most successful AI deployments happen when technology providers, operators, regulators, and academic institutions work together. Each brings unique perspectives and capabilities. Operators understand the practical challenges and requirements. Technology providers bring innovation and expertise. Regulators ensure safety and compliance. Academic institutions drive fundamental research and talent development.
I see several transformative trends emerging, centered around what we call agentic autonomous AI. This represents a fundamental shift from traditional automation to intelligent systems that can reason, adapt, and make decisions within defined operational boundaries.
The first major trend is agentic autonomous operations. Unlike conventional automation that follows rigid rules, agentic AI can understand context, anticipate problems, and take corrective actions autonomously while maintaining human oversight. Our AI agents can manage complex process optimization, coordinate maintenance schedules, and respond to unexpected scenarios across multiple time horizons simultaneously.
Second, integrating autonomous AI with IoT, edge computing, and 5G networks to create "intelligent infrastructure" where components communicate, reason, and optimize collectively. Imagine AI agents managing different refinery units collaborating autonomously to optimize entire facility performance.
Third, more sophisticated AI models are handling industrial complexity through multi-modal processing of sensor data, images, text, and operational logs. Our autonomous AI systems are cognitive, understanding and reasoning about industrial environments in unprecedented ways.
Finally, what's revolutionary is dialable autonomy. This isn't about replacing human expertise, but giving operators the ability to adjust AI autonomy levels based on situation and comfort. Dialable autonomy lets operators set the level of independence AI systems exercise. In steady-state operations, autonomy can be dialed high, enabling AI to execute optimizations continuously. During a planned turnaround or emergency, autonomy can be dialed low, shifting the system into advisory mode, where recommendations require explicit operator approval. For example, during a compressor trip event, the AI might continue providing optimization insights but await confirmation before executing changes. This flexibility reassures operators that they remain in control while still gaining the benefits of intelligent automation.
Operators can dial up autonomy for routine operations, dial back for critical situations, maintaining the essential human-AI partnership at the heart of safe industrial operations. The future is about intelligent partnerships where AI amplifies human expertise while humans retain ultimate authority over critical decisions.
Start with a clear understanding of your business objectives. AI is a powerful tool, but it's not a solution in search of a problem. Identify specific operational challenges where AI can deliver measurable value, whether that's reducing maintenance costs, improving energy efficiency, or enhancing safety.
Invest in building internal capabilities. You don't need to become an AI company, but you do need people who understand the technology and can work effectively with AI providers. This includes both technical expertise and change management capabilities.
Take a phased approach. Start with pilot projects that can demonstrate value quickly, then scale successful implementations across your operations. This allows you to learn and adapt while managing risk.
Finally, prioritize transparency and explainability. In our industry, trust is paramount. Choose AI solutions that you can understand, validate, and explain to stakeholders.
We're at an inflection point. The convergence of AI capabilities, data availability, and industry readiness is creating unprecedented opportunities for transformation. But success will require more than just technology. It will require vision, leadership, and a commitment to responsible innovation.
The companies that succeed will not only deploy AI, they will embed it as sovereign industrial infrastructure. Beyond Limits delivers this through explainable neuro-symbolic reasoning and agentic autonomy, ensuring that AI strengthens operational resilience, reduces emissions, and supports national strategies like Vision 2030. We are not only providing technology but partnering with industry leaders to build sustainable, sovereign, and accountable AI foundations that drive both profitability and long-term energy security.
The future of energy is being written today, and AI will be one of the most important chapters in that story.
AJ Abdallat is the CEO and Co-Founder of Beyond Limits, a leader in industrial-grade artificial intelligence solutions. With over 20 years of experience in high-tech entrepreneurship and a background rooted in NASA and Caltech research, he is a recognized thought leader in the application of AI to complex industrial challenges.
.jpg)