Global freight still runs on email, PDFs, paper work orders and legacy portals. Pallet built AI agents fluent in all of it - quietly doing the back-office work that keeps trucks, warehouses and shippers moving.
Logistics is a $12 trillion industry that, in 2026, still depends on humans copying numbers off paper into screens. Sushanth Raman saw it up close. On one visit, he met a trucking company owner named Riyo who keyed over 100 paper work orders into a computer for six hours every Saturday evening - sacrificing weekends so deliveries went out on time.
That scene is the whole pitch. Pallet, which Raman co-founded with fellow Retool engineer Andrew Spencer in 2021, builds an AI workforce to take that work off people's plates. Its flagship product, CoPallet, automates the unglamorous core of freight: order entry, quoting, dispatch, inventory, accounting, and portal updates.
Both founders have logistics in the family. Raman's grandfather was in the shipping business; Spencer's father runs the engineering team at transportation management company MercuryGate. They aren't tourists in this world - they grew up near it, then spent years building developer tools before turning that craft on an industry the software boom mostly skipped.
Raman describes Pallet as "like a modern OS for moving any physical product from point A to B." The goal isn't a flashy demo. It's the boring, durable plumbing that lets a midsize carrier reallocate 25 order-entry employees to higher-value work and save millions.
This wasn't a story about hype, it was a story about math. Customers ran the numbers.
- Sushanth Raman, CEO & Co-founderPallet's engineers say their agents mirror how a good logistics clerk actually works - four capabilities, stacked.
Sequential, repeatable processes encoded from how real teams already do the job.
Short-term context for the task at hand, long-term pattern recognition across thousands of orders.
APIs, browser automation, email, SMS, and computer vision - even legacy portals with no API.
Logical decision-making to handle the edge cases that break rigid, rules-only software.
The engineering choices are deliberately unsexy and sturdy: Zod schemas enforce strict input and output contracts so agents stay inside predictable boundaries. An event-driven design runs distributed, long-running jobs - lightweight workers handle simple lookups, GPU instances chew through document processing. And OpenTelemetry traces every request, decision and state change, so a human can always see exactly what the agent did and why.
About 10% of all logistics spend - roughly $100 billion - goes to administrative tasks: typing, copying, re-keying, chasing portals. Pallet wants every dollar of it. Customers report cutting repetitive staffing costs by 50 to 70% and increasing throughput up to tenfold.
Demand grew so fast that Pallet closed a $27M Series B just seven months after its Series A - not on a pitch, but because customers kept finding new jobs for the AI to do beyond what it was first bought for. Within 18 months of launch, the company had reached around 60 customers and a $3M annual run rate.
In March 2026, Pallet launched Atlas - a product that turns the same operational data into a profit map, surfacing the margin opportunities already hiding inside a customer's freight.
One midsize carrier reallocated 25 employees from order entry - and saved millions.
- Pallet customer outcomeSushanth Raman and Andrew Spencer leave Retool to start Pallet - two engineers with shipping in the family.
$18M Series A led by Bain Capital Ventures. ~60 customers, $3M run rate, 18 months in. TechCrunch calls it logistics in the 21st century.
Publishes its agent-engineering deep dive on the "coffee-stained world of logistics" - workflows, memory, tools, reasoning.
$27M Series B led by General Catalyst, just 7 months after Series A. Total funding reaches $50M.
Launches Atlas, surfacing hidden profitability inside customers' operational data.
Recognized by FreightWaves for customer-specific, end-to-end AI - and still hiring "ambitious builders."