"I Think I've Played Around with AI Enough": How a Seller Sobered Up from Vibe Coding and Automated a 7,000-Shipment-a-Day WarehouseNote · digest for November 16
Industrial FBS sorting line at a marketplace warehouse with a scanning terminal and cardboard boxes
Note4 sections≈ 4 min

"I Think I've Played Around with AI Enough": How a Seller Sobered Up from Vibe Coding and Automated a 7,000-Shipment-a-Day Warehouse

"I think I've played around with AI enough. Most of what we built turned out to be completely useless." That's the candid confession marketplace seller and Sharks of Wildberries and Ozon channel author Ivan Kurilenko used to kick off a breakdown of his internal automation. When a smart chatbot can spin up a working script from a plain text prompt in a single evening, it's hard to resist the urge to digitize every single step in your company.

I see this excitement among fellow sellers all the time: you start thinking you can rewrite half your SaaS stack for free and cut out contractors altogether. But where's the line between homemade scripts that actually save you money and expensive toys that put your warehouse logistics at risk?

Spring Vibe-Coding Euphoria and a Graveyard of Homemade Scripts

Back in the spring, Ivan got hooked on vibe coding—building apps and data handlers through AI chats in plain language, with zero programmers and zero budget. For the first 2 months, the team rode that high and cranked out bots, scrapers, and internal dashboards, genuinely believing everything under the sun needed to be automated.

The reality check hit 6 months later: an audit showed that almost all of their custom software had been abandoned and had zero impact on net profit. Instead of solving real retail bottlenecks, the staff had simply been entertaining themselves with a compliant piece of tech.

Cheap code creates a dangerous illusion of productive work.

The Market-Price Filter and the Custom RnP Spreadsheet Trap

To shut down the assembly line of pointless side projects, the company introduced a strict single-question filter: "How much would you pay for this tool if it were sold on the market?" If you wouldn't want it even for free, the task gets killed on the spot.

Rule number two: stop reinventing cheap off-the-shelf tools like the "Hand on the Pulse" (RnP 10X) management spreadsheet, where you track your daily order funnel, ad spend share, logistics, and SKU-level margins. Generating a draft dashboard is easy, but every time Wildberries or Ozon updates its API, your homegrown scrapers break.

Writing a program is ten times easier than keeping it running for years.

Having your team constantly patch broken data exports costs way more than a ready-made subscription. Why raw code is no longer a unique business moat is covered in detail in our piece on 8 defensive moats for businesses in the AI era, and quick reconciliation of marketplace commissions and deductions is already being built straight into seller dashboards—just like Market AI in the seller portal.

A 7,000-FBS-Shipment-a-Day Warehouse: Where You Need an In-House Developer

After rethinking the approach, the seller split development into two clear tracks. Small, local tasks—like reformatting a supplier's spreadsheet or whipping up a quick micro-utility for routine chores—are handled by staff using AI. Meanwhile, heavy infrastructure went to a dedicated in-house developer, who automated the warehouse line in 3 months: today it reliably processes around 7,000 FBS shipments a day (fulfilled by seller) with virtually zero penalties.

At that kind of volume, a software glitch hits your cash register immediately.

Under Wildberries' FBS rules, you get 48 hours to hand over orders: between 48 and 52 hours, you're fined 0.2% of the item's price for every hour of delay, after that it jumps to 1%, and canceling an order due to stock sync errors costs you double the marketplace commission. On Ozon, according to its FBS guidelines, letting your 14-day error rate cross the 2% threshold triggers withholdings of 1% to 8% of shipment value. Had they cobbled that system together with a chatbot, the warehouse would have ground to a halt at peak load, and the COO would have spent all day debugging.

The Real Cost of Maintenance and a Unified ERP System

The team's next infrastructure milestone is a unified ERP system tying together procurement, sales, and finance. The CFO acts as the internal client, the in-house engineer writes the core, and staff using AI will only step in later for day-to-day usage and minor tweaks in their own workflows.

Ivan Kurilenko summed up the lesson best himself: "AI has made development radically cheaper. That doesn't mean everything that can be automated actually needs to be automated."

Want a deep diveVotes help us choose what to break down in detail. This material has 0 votes.
Sources · 4