A demurrage report that once consumed a full shipping executive's day now takes seven minutes to compile. That figure, cited by Matthew Talbot, co-CEO of Complexio, at last week's Splash Singapore, captures both the productivity promise of AI in shipping and the problem it creates: if the repetitive junior work disappears, where will the industry find its next generation of experts?

The "AI, Digitalisation and the Maritime Workforce" panel brought together executives from shipping, technology, classification and mining, and the debate moved quickly past technology to the harder question of what happens to the people expected to work alongside AI.

Moderator Cynthia Worley, VP strategic accounts at Sedna, framed the moment as a foghorn rather than a siren song — a warning to proceed with clarity rather than at full speed. "We may not have jobs if we can't keep up," she said.

Talbot argued that change management, not technology, remains the principal obstacle. "People's jobs aren't going to be directly replaced by AI. The person next to you who's using AI is going to replace your job," he said. "AI is a multiplier effect. It makes everyone better. It makes us faster, makes us stronger, makes us more productive."

Small gains or full redesign?

The sharpest disagreement centred on implementation strategy. Talbot advocated starting with repetitive tasks — moving information between systems, gathering reports, work employees already dislike — to build confidence before wider deployment.

Chakib Abi-Saab, chief technology and innovation officer at Lloyd's Register, pushed for something more fundamental. "If you are automating bad processes, which is what we are doing, you get bad results faster, but still bad," he said.

For Abi-Saab, competitive advantage will come from redesigning workflows around AI rather than bolting copilots onto processes built for another era. He identified five requirements for scaling the technology: governed data, redesigned processes, accessible organisational knowledge, governance and change management.

"You never needed more maritime knowledge and expertise from people than today," he stressed, arguing that an employee merely clicking an AI-generated approval button does not constitute meaningful human oversight.

The unpaid quality controller problem

Janani Yagnamurthy, SVP of product and strategic growth at Marcura, made a similar point on trust. "If you do not evidence your observations, you are actually turning your end user into AI's unpaid quality controller," she said.

Marcura now starts with the business decision and works backwards into the technology rather than inserting another system into already crowded workflows. "Stop admiring AI as long as AI makes you look better and makes your judgments even more effective," she advised.

She cited a chartering example in which AI assesses a Santos-Rotterdam sugar voyage, examines charterparty and unpaid-time risk, and suggests adding $2 per tonne to a quotation — a concrete illustration of how tools move from data retrieval to commercial judgement support.

Talbot's demurrage example shows the same shift: structured and unstructured data brought together cut a 90-day report from a full day of compilation across multiple systems to seven minutes.

Onboard, the brakes stay on

Offshore adoption runs at a different pace. Rajesh Divakaran, GM and head of group crewing services at BW Group, said systems must first prove reliable ashore, then run on a limited number of ships before fleetwide deployment. "We need to have a wait and see approach rather than rushing into deploying onboard," he said.

Accountability, he stressed, remains human. If an AI-supported navigational or operational system fails, masters and crews will still exercise seamanship and take control.

Menand Karsan, GM – RT Marine at Rio Tinto, said the miner sees similar dynamics through its relationships with owners and operators. AI-powered cameras can initially make crews uncomfortable, but become powerful safety tools once their purpose, controls and governance are clear.

The question nobody could answer

The panel's most consequential moment came from the audience. Pradeep Chawla, who heads MarinePALS, asked: if AI removes the repetitive junior tasks through which generations of shipping professionals learned their craft, who becomes the expert capable of challenging AI in 20 or 30 years?

Abi-Saab called it the best question he had received on any AI panel — and admitted he had no answer.

For carriers, operators and forwarders racing to deploy AI across chartering, documentation and claims workflows, the exchange signals a looming structural risk: the commercial gains of the next five years may hollow out the expertise base the industry will need to supervise, audit and override those systems in the 2040s. The question is no longer whether machines can learn maritime expertise, but how humans will continue to acquire it.