Zeronorth.com
ZeroNorth, the Danish maritime technology firm that already optimizes more than 5,500 vessels worldwide, launched Propel by ZeroNorth on Tuesday — an agentic AI system that autonomously executes voyage coordination tasks that human operators have handled manually since the industry began. Unlike the advisory software that preceded it, Propel does not surface a recommendation and wait. It generates voyage plans, contacts the vessel master, incorporates the master’s feedback into a revised plan, and updates routing as weather and operational conditions shift — keeping the shore-side operator informed without waiting for each step to be manually approved. Founding partners Cargill, Ultrabulk, and CMB.TECH received early access on the launch date, with broader commercial rollout planned for later in 2026.
For the shipping industry, this is a qualitatively different kind of software — and the technical architecture behind it matters more than the headline claim.
What “Agentic” Actually Means in a Voyage-Optimization Context
The word “agentic” is contested enough in enterprise AI circles that Gartner issued a June 2025 warning about “agent washing” — the rebranding of existing automated systems as autonomous AI agents without a meaningful change in underlying capability. The distinction is specific: an advisory system generates an optimal route recommendation and delivers it to a human who then sends it to the ship. An agentic system executes the multi-step coordination loop that follows — communication with the vessel master, plan revision, condition-responsive rerouting — as an autonomous workflow with the human monitoring, not driving.
ZeroNorth’s existing platform already sat on top of a technically sophisticated optimization engine. The core mechanism combines naval architecture principles with machine learning and historical vessel-specific data to produce fuel consumption predictions with 93.8% median accuracy, independently verified by DNV. Every voyage runs through a routing simulation that scores each candidate route across three dimensions: bunker cost, transit time, and emissions exposure. External weather forecasts feed into the system every six hours, triggering automatic recalculation when a weather window shifts — before the operator would otherwise know conditions had changed.
That underlying engine is unchanged in Propel. What Propel adds is an execution layer. Previously: the optimization engine produced a recommendation, and a human operator read it, sent it to the vessel master, received the master’s feedback, produced a revision, and sent it again. Each iteration consumed time — the coordination loop across a large fleet compounds quickly. According to the Maritime Executive, voyage planning takes 3.5 hours per vessel. Propel automates that loop. The agent handles the communication sequence; the human operator receives updates and can intervene, but is not required to execute each step.
The Financial Times, analyzing the state of enterprise AI agents in May 2025, compared agent autonomy to the SAE self-driving classification scale — noting that most real-world applications operate at Level 2 or Level 3, where the system handles defined subtasks while a human remains ultimately responsible. Propel’s design maps precisely to that description: bounded autonomy within operator-defined constraints, with humans informed rather than bypassed. That is an explicit architectural choice, and one that differentiates Propel from unconstrained agentic systems that have generated well-documented failure modes in other industries.
The boundary constraint — agents act “within the limits operators define” — also addresses one of the documented failure modes in agentic AI deployment elsewhere: agents executing consequential actions without human confirmation in high-stakes environments. In a domain where a routing error on a single laden voyage can cost hundreds of thousands of dollars in fuel and charter party penalties, that constraint is commercially essential, not a limitation.
Why the Timing Makes Commercial Sense
ZeroNorth chose the moment carefully. Since January 1, 2026, the European Union’s Emissions Trading System has applied 100% maritime shipping coverage — all CO₂, methane, and nitrous oxide emissions from vessels over 5,000 gross tons operating to or from European ports must now be offset by tradable EU Allowances (EUAs, each covering one tonne of CO₂ equivalent). At current EUA prices of roughly €75–80 (approximately $86–92 USD) per tonne, a mid-size bulk carrier on a 10-day laden leg from a non-EU port into Europe now faces approximately €37,800 (approximately $43,000 USD) in bulk carrier ETS carbon costs on that single leg, before port stay emissions. The cost of getting the routing or speed wrong has grown materially.
Simultaneously, the International Maritime Organization’s Carbon Intensity Indicator (CII) framework — mandatory since January 2023, with annual thresholds tightening through 2030 — assigns ships a letter grade from A to E based on grams of CO₂ per cargo-carrying capacity per nautical mile. Ships that receive three consecutive D ratings or one E rating must file a corrective action plan. Speed and routing decisions made voyage by voyage accumulate directly into a vessel’s annual CII score. In this environment, the hours of lag time in the manual coordination loop are not just operationally inefficient — they are financially and regulatorily consequential.
ZeroNorth’s platform already claims 5–10% fuel savings per voyage from its optimization engine. At VLSFO prices of $840 per metric tonne in Singapore on August 4, 2026, and a typical fuel burn of 25–35 tonnes per day on a bulk carrier, a 7% improvement compounds quickly across a fleet and across a year’s voyages. Propel is not changing the optimization math — it is removing the coordination latency that allowed some of those savings to be missed when conditions changed faster than a human operator could respond.
The Technical Stack Beneath Propel
The optimization engine that Propel sits on top of is one of the more sophisticated in maritime software. ZeroNorth’s fuel models combine three data sources: naval architecture principles (how a vessel’s hull form responds to speed, heading, and sea state), machine learning trained on historical voyage data, and live vessel telemetry from the SMARTShip system. The resulting models are vessel-specific — not generic fleet averages — and DNV has independently verified their 93.8% median fuel model accuracy.
The system’s integration with Charter Party terms is a distinguishing technical feature that most public descriptions of maritime AI overlook. Charter Party clauses typically specify speed bands, fuel consumption limits, and weather routing windows within which a vessel must operate. A voyage optimization engine that recommends routes outside Charter Party constraints is commercially useless — the recommendation may be mathematically optimal but legally and contractually invalid. ZeroNorth’s routing simulation runs within CP constraints from the outset, producing plans that are commercially defensible before voyage commencement. The same logic applies to Propel’s execution layer: when the agent revises a plan, it revises it within the same constraints.
ZeroNorth’s February 2024 Alpha Ori merger added high-frequency IoT sensor data and vessel-level digital twins to the platform, improving the accuracy of AI prediction models with each vessel onboarded. The platform now tracks more than 5,500 vessels, giving the fuel models a large base of real-world performance data across vessel types, trade routes, and weather conditions.
Voyage Optimization is not ZeroNorth’s only product — the platform spans vessel reporting, emissions analytics, bunker procurement, and Charter Party selection. Propel is designed to connect across these modules: routing decisions in the voyage agent flow into the bunker procurement module for stem planning and into emissions analytics for CII compliance tracking, without manual re-entry of data at each step.
ZeroNorth’s Trajectory and Who’s Already Using It
ZeroNorth was spun out of Maersk Tankers in early 2020 as a six-person team. By January 2025, it had grown to more than 600 employees across ten global locations and reached just under $40 million ARR, a 38% increase from 2023. In December 2024, the company recorded its first month of positive EBITDA; it confirmed full-year positive EBITDA for 2025. In February 2025, the company raised $20 million to continue scaling.
The deployment record is substantial. In 2024, ZeroNorth’s platform optimized 72,000 voyage legs by generating 1.5 million routes, with the company claiming the activity cut over one million CO₂ tonnes that year and by more than 5.2 million metric tonnes since the platform launched in 2022. In January 2026, Maersk Tankers signed a five-year renewal covering approximately 240 vessels across all tankers and gas carriers under its management — expanding a relationship that began when ZeroNorth itself was still an internal Maersk Tankers tool. In March 2026, Western Bulk, the Norwegian dry bulk operator, rolled out the ZeroNorth platform fleet-wide across approximately 120 commercially controlled vessels in a six-week global implementation.
The three founding partners for Propel — Cargill Ocean Transportation, Ultrabulk, and CMB.TECH — have been ZeroNorth customers since the company’s founding. Their involvement in shaping Propel before wider release reflects ZeroNorth’s stated approach: test each new capability against real operational conditions across different shipping segments before broader deployment.
James Lewis, Vice President of Global Operations at Cargill Ocean Transportation, said the company was proud to work with ZeroNorth to explore technologies that can modernize shipping, improve operational efficiency, and help move shipping forward.
Oskar Fabricius, CFO at Ultrabulk, was more specific about the distinction Propel represents: “The potential for AI that can take action — not just surface insights — is significant, and as a partner since ZeroNorth’s earliest days, we are glad to be at the front of shaping what that looks like in practice,” Fabricius told Ship & Bunker.
Søren Andersen, who joined ZeroNorth as CEO in March 2026, framed the launch as a structural shift in what maritime software can do: “There is immense potential to move towards operations that are more connected, continuous and precise, where technology takes on more of the coordination work and people can focus their judgement where it creates the greatest value,” Andersen said at the launch.
What’s Next: Vessel, Fuel, and Fleet-Level Agents
The voyage optimization agent is the first Propel release. ZeroNorth has confirmed that vessel performance and fuel workflow agents are in development and will be released on a rolling basis, with each new capability tested against operational data from partners before wider deployment. The company’s stated longer-term vision is for Propel to monitor maritime operations continuously across voyage, vessel, and fuel dimensions simultaneously — maintaining coordination across all three layers rather than handling them sequentially as separate software products.
Whether that vision materializes at scale depends on how well the execution-layer architecture performs under conditions the founding-partner testing period does not fully surface: novel weather events, mechanical anomalies, geopolitical disruptions to trade routes, and edge cases in charter party interpretation. ZeroNorth’s deliberate approach — three months of internal testing with its Professional Services team before any partner access, followed by a controlled founding-partner phase before broader commercial release — is the right architecture for finding those cases before they compound. The broader shipping industry will be watching how Cargill, Ultrabulk, and CMB.TECH use Propel through the remainder of 2026.
Currency conversions in this article are based on a EUR/USD rate of approximately $1.15, as of August 4, 2026, and are approximate.
Frequently Asked QuestionsWhat is agentic AI, and how is Propel different from ZeroNorth’s previous software?
Agentic AI refers to AI systems that pursue goals and execute multi-step tasks autonomously, rather than generating a recommendation for a human to act on. ZeroNorth’s previous voyage optimization software produced the best route and speed plan for a voyage — but a human operator still had to send that plan to the vessel master, receive feedback, revise the plan, and re-send it, repeating this cycle as conditions changed. Propel adds an execution layer: the AI agent handles the communication-and-revision workflow autonomously, keeping the operator informed and able to intervene, but no longer requiring the operator to execute each step manually.
How does voyage optimization reduce fuel costs and emissions?
Voyage optimization calculates the most fuel-efficient combination of route, speed, trim, and departure timing for a specific vessel on a specific voyage — accounting for live weather, ocean currents, charter party contractual constraints, and regulatory requirements including CII and EU ETS carbon costs. ZeroNorth’s platform claims 5–10% fuel savings per voyage through a combination of dynamic routing, speed optimization, and continuous re-routing as conditions change. For a bulk carrier burning 30 tonnes of fuel per day on a 15-day voyage, a 7% fuel reduction amounts to roughly 31.5 metric tonnes of fuel and roughly 99 metric tonnes of CO₂ avoided — per voyage.
What does the EU Emissions Trading System mean for ship operators in 2026?
Since January 1, 2026, every vessel over 5,000 gross tons operating to or from European ports must purchase EU Allowances (EUAs) to offset 100% of its verified CO₂, methane, and nitrous oxide emissions on those voyages. At current EUA prices of approximately €75–80 (about $86–92 USD) per tonne, the carbon cost on a single EU-linked voyage of a mid-size bulk carrier can reach five figures. This is a direct financial incentive to minimize fuel burn on every EU-touching voyage — which is one reason voyage optimization software adoption has accelerated rapidly across the industry in 2025 and 2026.
Does letting an AI agent manage voyage coordination create risks operators should understand?
The documented failure modes of agentic AI in other industries — agents executing consequential actions beyond their intended scope — are real. ZeroNorth’s explicit response to this risk is architectural: Propel’s agents act only “within the boundaries operators define,” and the system is designed to keep the human operator informed rather than bypassed. The three-month internal testing period before any external partner access, followed by the controlled founding-partner phase, reflects a deployment philosophy calibrated to the error cost of getting things wrong in maritime logistics. Operators adopting Propel should understand what the agent can and cannot decide autonomously, and confirm those boundaries are correctly configured for their specific fleet operations and charter party obligations.