I’m from the generation that remembers the movie WarGames. It’s the one where a computer almost starts a nuclear war because it didn’t understand real-world consequences.
Thankfully, the prospect of a machine programmed with Artificial General Intelligence going rogue and deciding to launch a missile strike remains firmly in the realms of science fiction. Even so, modern generative AI models, powerful as they are, have their own set of challenges. Many of us have seen the consequences of hallucinations or model collapse; AI confidently delivering wrong answers, skipping steps, or inventing data entirely. In maritime, where accuracy matters, those risks are real.
At Marcura, we don’t just theorize about AI. We test it. We use it. Internally. First. As I like to say, we eat our own dog food. That way, we know it works before it ever reaches you.
Curious about what AI can do for you?
Join our webinar on 21st May as we dive deeper into how Marcura AI is helping customers boost voyage profitability with smarter port calls and claims
Generative AI vs. agentic AI
Generative AI, like ChatGPT, is useful. It helps draft emails and summaries in the blink of an eye.
But maritime operations need more than a digital assistant. They need AI that handles tasks, follows context, and makes decisions. That’s where Agentic AI comes in.
Agentic AI isn’t built to chat, it’s built to work. It understands context, triggers workflows, and takes action inside your systems.
It’s not just answering questions. AI agents like the ones we’re building at Marcura go beyond that, they retrieve documents, surface insights, and initiate next steps.
What makes up an AI agent?
An AI agent has four core components:
Models: Large Language Models (LLMs) tuned for maritime work—like OpenAI or Claude. We select models based on the task. Sometimes that means fast. Sometimes that means accurate. Sometimes both.
Prompts: Prompting isn’t just asking a question—it’s a craft. There are user prompts, system prompts, and meta prompts, all pre-configured by us to get results you can trust.
Tools: Connections to your systems—email, virtual machines, document handling. This is where the agent becomes useful, not just smart.
Data and workflow: Clean, well-organized data that supports sound decisions. The best agents work quietly in the background. No prompts needed. Just results.
The link between data and risk
If your data is messy, AI won’t fix it. It’s far more likely to make a bigger mess faster make the mess faster.
You’ve probably seen it:
Compounded errors: One step goes wrong. Then that has an impact on five subsequent steps with the issues becoming more serious with each step.
Hallucinations: AI spits out something wrong but sounds confident doing it. You may have heard the story about the AI model that couldn’t count the number of ‘r’s in “strawberry”? Maritime doesn’t have time for that.
Model drift: AI loses accuracy over time if you don’t feed it fresh, quality data.
Maritime data is complex and often defies standardisation. Tariffs aren’t clear. Scans are hard to parse. Calculations get tricky.
Add AI to that mix, and small issues quickly become big ones. That’s why we manage our data tightly. And why we test everything.
That’s also why we always keep a human in the loop.
AI can get close, but it still can’t reason the way a person can. Without oversight, there’s a risk the AI will do something confidently… and completely wrong. We’ve learned that the most powerful combination isn’t the best model or the smartest agent. It’s that model paired with a human who knows the domain.
Like Garry Kasparov said after his matches against Deep Blue: the winning formula isn’t man or machine, it’s man with machine. And it’s not just about the data or the model. You need the right scaffolding, evaluation frameworks, feedback loops, governance layers, that keep the AI in check. We’ve built all of this internally before handing it to you.
Agentic AI in Action
Let’s say you ask a generic AI, “When will my vessel be available next?” It might say, “When discharge finishes.” Technically true. Operationally useless.
Agentic AI gives you the full picture: “Discharge in Singapore starts March 25, takes three days. Vessel ready March 28.” It checks weather. Port delays. Timings. All of it. We apply this kind of context-first reasoning across many workflows. Take port DA screening. Our agents pull 200,000 port call protocols, extract clauses, digitize Statements of Fact, and flag inconsistencies—all without being prompted.
The same goes for laytime calculations or COA-based claims. The agent highlights what matters and keeps the rest out of your way.
The Marcura Way
We’ve spent 20+ years in this industry. We know how to build tools that fit how you actually work. Our AI is tested, grounded in real-world use, and trained on maritime data, not general web content.
Marcura AI agents are already delivering real impact, embedding context-based AI across workflows to automate processes, reduce risk, and enhance decision-making:
Charter party interpretation: Agents extract key clauses in seconds, flag risks, and generate voyage orders automatically.
Contract intelligence: AI benchmarks pricing, flags ambiguous clauses, and surfaces red flags before fixtures happen.
Port call and PDA management: With over 200,000 port calls processed, our agents apply 2,000+ rules per call—delivering faster estimates, instant discrepancy checks, and automated tariff validation.
Claims handling: We process 20,000+ claims annually, detecting overcharges and accelerating settlements with greater accuracy. These aren’t proofs of concept. They’re available now and they’re built to make your daily work faster, clearer, and more predictable.
This isn’t some plug-in to chat GPT. This is context-driven, vertical AI built for maritime, shaped by maritime, and tested inside real maritime operations.
What we’re delivering isn’t about replacing people. It’s about helping you move faster, spot issues earlier, and make better calls. That’s what Agentic AI does. And no missile launches.
Learn more about Marcura AI
