AI Is Eating Logistics

Flexport uses AI to optimize container shipping, achieving 2% cost savings while simultaneously improving transit times by 20%—a typically impossible trade-off. The company aims to reduce ocean freight costs by 8-10% over the next few years through AI automation, which represents 10% of total freight forwarding costs.

• •AI-powered planning saved 2% on ocean freight spend while improving transit time 20%
• •Target: 8-10% reduction in ocean container shipping costs globally through AI
• •Labor represents 10% of total freight forwarding costs, making it a major automation opportunity
• •Scale economy shared model: bigger scale + automation = lower costs passed to customers

Flexport runs bi-annual hackathons where 90% of projects now use LLMs, compared to just a handful 18 months ago. These hackathons generate production-ready features that leadership wouldn't have conceived, forcing a balance between top-down direction and bottom-up innovation. The company schedules hackathons before roadmap planning to incorporate winning ideas into the budget.

• •90% of hackathon projects now LLM-based vs. ~5 projects 18 months ago (out of 50 teams)
• •Hackathon ideas becoming real product features, not just toys
• •Timing hackathons before roadmap planning to budget winning projects
• •Engineers close to business problems generate ideas leadership wouldn't conceive

Flexport created a 90-day AI boot camp giving non-engineers one day per week to learn cursor, Streamlit, and workflow automation. The program aims to make participants 10x more productive by enabling domain experts to automate their own repetitive work without waiting for engineering resources.

• •90-day program: 20% time allocation (1 day/week) for non-engineers to learn AI tools
• •Focus on cursor, Streamlit, and workflow automation for 'freight email forwarding' tasks
• •Goal: return participants as 10x more productive than peers
• •Started in Amsterdam engineering office, now expanding globally

A hackathon project became a production feature allowing customers to query supply chain data in natural language instead of SQL or dashboards. This automation saves 25% of account management time previously spent generating reports for customers.

• •Natural language interface for supply chain data (orders, SKUs, costs, customs)
• •Eliminates need for SQL knowledge or dashboard building
• •Saves 25% of account management time on report generation
• •Started as hackathon project, now production feature

Machine learning models optimize which ship each container goes on, balancing cost and transit time. The system automatically rebooking 2,000 canceled containers per week (10x daily) by moving forward containers originally scheduled later, achieving the 20% transit time improvement.

• •2,000 containers canceled weekly—AI automatically finds replacement cargo to move forward
• •Optimization runs 10x daily across entire container inventory
• •Classical solver for optimization, potential for LLM agents to use solver as tool
• •Scale makes human optimization impossible—requires automated decision-making

LLM agents handle warehouse address verification and delivery appointments via email and voice calls. For any location not delivered to in 3 months, agents proactively confirm addresses and schedule appointments—work previously too expensive to do manually.

• •AI agents verify warehouse addresses via email and voice before delivery
• •Triggers for any location not visited in last 3 months
• •Prevents costly delivery failures from bad address data
• •Automates work that was too expensive for humans to do consistently

Flexport automated 20% of work at year start, will hit 50% by year-end, and initially targeted 80% for next year. They now believe 90-95% automation is achievable with current technology, with even higher potential as LLMs progress.

• •Automation progression: 20% → 50% (current year) → 80% target → 90-95% revised estimate
• •Customer sentiment detection in messages creates automatic escalations to managers
• •AI spell-checker reduces customs filing errors from 2% industry standard
• •Infinite list of automation opportunities still being discovered

Discussion of AI's potential to increase GDP 7% annually (doubling in 10 years) and philosophical implications. Petersen argues companies exist to deliver goods/services efficiently, not employ people, and that human desire for 'more' will create new jobs as AI automates existing work—similar to historical technological transitions.

• •AI could increase GDP 7% annually, doubling economy in 10 years
• •Companies' role is delivering goods/services efficiently, not employment
• •Human nature's infinite desire will create demand for new work as AI automates existing jobs
• •Historical parallel: Axial Age (500 BC) when coins created societal disruption and new philosophies emerged

Petersen advises founders to focus on price per share and control, not ownership percentage. His key lesson: money wants to spend itself, leading to bloat. He recommends a 90-day hiring freeze immediately after raising large rounds to maintain problem-solving culture instead of defaulting to hiring.

• •Only metrics that matter: price per share and control (legal/cultural)
• •Biggest mistake: using money to solve problems by hiring instead of executing
• •Recommendation: 90-day hiring freeze after large fundraise to preserve culture
• •Flexport experienced headcount bloat from defaulting to hiring for every problem

Flexport's vision is to ship anything, anywhere, by any means, via code—making logistics a utility like electricity. Currently shipping to 147 countries with employees in 22, they plan to cover 95% of container trade with own employees by 2028 and be in every legal country by 2035.

• •Vision: ship anything, anywhere, by any mode, in any quantity, all via APIs/voice
• •Current: 147 countries served, employees in 22; Target: 95% of container trade by 2028
• •Easier to automate own employees than third-party partners
• •Goal: logistics becomes invisible utility so companies focus on making products and talking to users