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

Let's get one thing out of the way: we didn't have a fancy startup budget. No VC money. No marketing team. No design agency on retainer. We had a jewellery shop on the ground floor, a room on the first floor, two laptops, and an internet connection that dropped every time it rained.

And somehow, we built Stashed — a bag brand — using AI.

This is not a "10 ways AI can help your business" listicle. This is the messy, real, sometimes-embarrassing story of how two boys from Vaniyambadi used AI to do things that should've taken a team of 15 people.

The Name Came From Claude (Sort Of)

We knew we wanted a bag brand. We knew it had to feel Indian but not "ethnic." Modern but not pretentious. Something that a college kid in Chennai would carry without thinking twice.

We spent three days trying to come up with names manually. Absolute garbage. "UrbanKarry." "BagWala." We're not joking.

Then we opened Claude and wrote something like:

"We're building an India-first bag brand for young people who use public transport. The bags have removable Velcro front panels so you can swap designs. Give us 30 brand names that feel modern, short, and have a hidden meaning about carrying or keeping things."

Claude gave us a list. Most were mid. But "Stashed" was in there. The moment we read it, we both looked at each other. Yeh hai. That's the one. You stash your stuff. You keep your things close. It's casual. It's real.

We didn't just take the first output though. We went back and forth — probably 40 messages refining the brand positioning, the tagline, the voice. Claude became our brand strategist that night.

Visual Identity Without a Design Team

Here's where it gets interesting. We needed product shots, social media content, and a visual language — but we couldn't afford a photographer or a graphic designer.

Enter Nano Banana Pro.

We started generating product mockups using Nano Banana. The early ones were... interesting. AI-generated bags that looked like they were designed by someone who had never seen a bag in real life. Zippers in weird places. Straps that defied physics.

But we learned. We got better at prompting. We figured out that specificity is everything:

"Minimalist crossbody bag, matte black cordura fabric, removable velcro front panel with a geometric pattern, Indian metro background, natural lighting, product photography style"

That prompt, refined over probably 50 iterations, gave us visuals that people genuinely thought were real product shots. When we posted them on Instagram, the DMs started coming: "Where can I buy this?"

We hadn't even manufactured the first batch yet.

The Copy Machine

Every product description, every Instagram caption, every email — Claude wrote the first draft. But here's the thing people don't understand about using AI for copy: the first draft is maybe 60% there. The magic is in the editing.

We'd write prompts like:

"Write an Instagram caption for Stashed. The bag has a removable front panel. The audience is 18-25 year olds in Indian cities. Tone: like a friend telling you about something cool they found. No corporate speak. No hashtag spam. Hinglish is fine."

And Claude would give us something decent. Then we'd rewrite it. Add our voice. Remove the parts that sounded too polished. Make it sound like us — two guys who are excited about this thing they built.

The best performing caption we ever wrote was 80% Claude, 20% us. The worst performing one was 100% Claude, 0% us. That taught us the rule: AI writes, humans vibe-check.

Video Content: The Game Changer

Static posts are fine. But reels? That's where the attention is.

We started using Veo 3.1 for video content. Product reveal videos, behind-the-scenes style clips, lifestyle footage. The quality blew our minds — and this is coming from guys who were shooting on a phone six months earlier.

For social-first short videos, Higgsfield.ai became our secret weapon. Quick, punchy, trend-format videos that we could push out daily without spending hours editing.

One video we made — a 15-second clip of the Velcro panel being swapped — got more engagement than anything we'd ever posted. It cost us zero rupees and about 20 minutes of prompting and editing.

Manufacturing: Where AI Couldn't Help (Much)

Here's the honest part. AI is incredible for the digital side. But when you're dealing with physical products, reality hits different.

AI can't tell you that your zipper supplier is going to delay by two weeks. It can't feel the difference between 600D and 900D polyester. It can't stress-test a bag on a crowded Chennai bus during rush hour.

That part — the manufacturing, the materials, the quality control — that was all us and our team at SN Bags. Years of B2B manufacturing experience that no AI can replace. We wrote a whole post on that physical product journey — the zipper decisions, the Velcro testing, and everything that happens before a bag reaches a customer.

But AI did help us think about manufacturing problems:

• We used Claude to research materials, compare durability specs, understand fabric treatments
• We generated design variations quickly before sending final specs to our production team
• We wrote SOPs and QC checklists using AI, then refined them based on real factory floor experience
• We used AI-generated RFQs when finding our manufacturer — that process alone saved us weeks

The combination of AI-powered digital operations and hands-on manufacturing knowledge — that's what made Stashed possible.

The Chaos Was Real

We're not going to pretend this was smooth. There were nights where:

• Claude hallucinated a fabric type that doesn't exist, and we almost ordered it
• An AI-generated product shot went viral but the actual bag looked nothing like it (we had to quickly update the design to match)
• We automated our email marketing and accidentally sent the same welcome email three times to our first 200 subscribers
• A competitor copied our AI-generated visuals because they were too generic — lesson learned on adding brand-specific elements

Every one of these mistakes taught us something. We documented them. We built systems. We got better.

The Real Takeaway

Building Stashed with AI didn't mean AI built Stashed. It meant two guys with no budget and no team could move at the speed of a funded startup.

Here's what AI actually did for us:

• Compressed time: What would take a week took a day
• Reduced cost: What would cost lakhs cost almost nothing
• Expanded capability: We could do things we had zero training in (design, copywriting, video production)
• Forced clarity: Writing good prompts forced us to think clearly about what we actually wanted

Here's what AI didn't do:

• Make strategic decisions for us
• Replace the manufacturing knowledge we'd built over years
• Create genuine human connection with our customers
• Handle the 4 AM anxiety about whether this whole thing would work

If you're sitting in a small town right now, thinking you need to move to Bangalore or raise funding to build something — you don't. You need a laptop, an internet connection, the right AI tools, and the willingness to look stupid for a while.

That's what we had. That's what worked.

What We'd Tell Our Past Selves

1. Start prompting badly. Your first 100 prompts will be terrible. That's the learning.
2. Don't hide the AI. Your audience doesn't care if AI helped. They care if the product is good.
3. Build the system, not the shortcut. AI as a one-off trick is useless. AI as a repeatable workflow is powerful.
4. Keep the human in the loop. Every AI output needs a human vibe-check before it goes live.
5. Document everything. Your prompts, your workflows, your mistakes. That documentation becomes your competitive advantage.

Stashed exists because AI exists. But more importantly, Stashed exists because two guys from a small town decided to figure it out.

Your turn.

---

This is the story behind Stashed. If you want to learn the AI skills we used to build it, check out 2BFT Academy.