How We Built Stashed Using AI (And Almost Lost Our Minds)

Stashed is a modular tech bag brand. Made in Chennai. Sold at stashed.co.in. Nav (me) and Sujal built it. 18 months from idea to first shipment. 80% of the work was AI. 20% was the factory. Here's what worked, what didn't, and what we'd do differently.

The brief we gave ourselves.

September 2024: a bag for Indian commuters. Removable Velcro front panels so you can swap designs. 22L, fits a 16" laptop, water-resistant, ₹4,999. Made in India, sold direct, no middleman.

The brief to ourselves was: "build the brand, the product, the photos, the copy, the launch — all with AI. Spend under ₹50K on tools. Take under 6 months." We missed the 6 months (took 18) but hit the budget (₹32K total for everything AI-related).

What AI did.

Naming: 200+ candidates generated by Claude, narrowed to 12, picked "Stashed" in 4 hours. Without AI: 4-6 weeks of arguing.

Product photos: Nano Banana 2 generated 30 product shots. We paid for one professional shoot (₹25K) and it was worse than 25 of the 30. The AI shots shipped.

Instagram: 90 posts in 3 months, all captions written by Claude, all images by Nano Banana 2 Lite, all scheduling by OpenClaw. 14K followers, 4% engagement rate. Without AI: 2-3 posts/week, half the engagement.

Customer service: WhatsApp agent (OpenClaw + Claude + the Nano Banana 2 product ID) handles 40% of inbound. Saves 6 hours/week.

Launch reel: Veo 3.1 made 12 clips in 2 hours. We edited them in CapCut. 250K views on the launch reel.

What AI didn't do.

Manufacturing. Finding the right factory, negotiating pricing, doing QC. That's a human job. We went through 4 factories before we found SN Bags (yes, our own SN Bags, but we'd have used them anyway).

Trust. Customers still want to feel the bag before they buy. We had to ship 50 free bags to influencers, 20 to journalists, 30 to early customers. The AI can build the brand, but the brand has to be tried.

What we'd do differently.

Start with OpenClaw day 1, not month 6. The Instagram pipeline alone would have saved us 30 hours/week. We could have launched in 4 months, not 18.

Use Nano Banana 2 Lite for lifestyle, full for product. We over-spent on the full model for shots where Lite was 95% as good.

Document the AI workflow from day 1. We have all the prompts now but we had to reconstruct them. The full playbook is in /services/ugc-for-ai-tools — for anyone building a physical product with AI.