cd /news/ai-agents/i-built-an-ai-sourcing-agent-for-fac… · home › topics › ai-agents › article
[ARTICLE · art-142341] src=dev.to ↗ pub= topic=ai-agents verified=true sentiment=· neutral

I built an AI sourcing agent for factories in Vietnam. AI does about a third of the work, the rest is still a person.

A developer based in Vietnam built an AI sourcing agent that converts free-text product briefs into business Vietnamese factory requests using three separate model calls, with the first two steps running as live model calls in a public demo. The developer states AI handles roughly a third of the sourcing job, while human factory visits, sample checks and paperwork remain manual, and deliberately left Zalo messaging unautomated because the platform has no open API for cold outreach.

by read4 min views3 publishedSep 30, 2026

I live in Vietnam. Every few weeks someone from the US or Europe asks me some version of the same question: how do you actually get a factory here to make what you want, without flying out yourself, and where do you even start with a manufacturer that doesn't speak English.

There's a lot of talk right now about AI agents that supposedly negotiate with factories on their own. In practice, AI does maybe a third of that job: it reads the brief, translates it the way factories actually write to each other, not a straight machine translation, and pulls messy replies into one table. The other two thirds is a person who drives to the factory, checks the sample by hand, signs the paperwork. Nothing wrong with that, it's a normal business model. Just don't call it "AI negotiating for you."

I've seen this market from the inside, so I built exactly the part that's genuinely AI, and showed it on a live demo, not screenshots.

Three separate calls to the model, not one giant prompt. Each step is checkable, an error doesn't get lost inside one big answer.

Step one reads a free-text brief in any language and pulls out the structure: product, material, quantity, density, colors, sizes, packaging. If the quantity or product type is missing, it doesn't guess, it asks. That's the one hard rule in the prompt: no invented numbers.

Step two takes that structure and writes the request in business Vietnamese, the format factories in Nha Trang and Ho Chi Minh actually use. Not Google Translate: industry terms like MOQ, gsm, and incoterms stay in Latin script even in Vietnamese, and a normal translator misses that.

Step three, production only, not in the demo, reads incoming replies from Zalo, WhatsApp, or email and merges them into one table. Factories reply informally: price with no currency (almost always dong, rarely stated), MOQ tied to quantity with a dash, deadlines in working days with no calendar reference. The prompt keeps the dong-to-dollar rate in one place, not scattered across the code, and normalizes everything to one format.

Reading the brief and translating it are live model calls, that's the actual technology on display. The pricing table at the end isn't live: it's a set of presets by product category, adjusted by a simple rule for your quantity (bigger order, lower unit price), not a call to an actual factory.

Messaging on Zalo isn't automated either, and that's a deliberate choice, not something I didn't get to. Zalo has no open API for cold-messaging people who haven't opted in themselves. Automating it means either spamming from a personal account and getting banned, or promising something that isn't technically there. So the demo shows the real process, how a brief turns into a Vietnamese request, and doesn't pass off a mockup as a live send.

I typed in: "oversized cotton fleece hoodie, 320 gsm, 500 units, three colors, sizes S-XL, embroidered logo, individual packaging." The agent pulls out the spec field by field, lets you correct it by hand, and a couple seconds later hands back a finished request in Vietnamese, polite, exact units, nothing lost. Below it, a table of four sample workshops with FOB pricing and minimum batch size, marked as a reference point, not an offer.

For hoodies and bags I found Vietnam-specific numbers (Cosmo Sourcing): FOB around $8 per hoodie at batches of 1,000-3,000+ units at larger factories, more at small workshops. For bags the range is wider, $5-20 per unit depending on how complex the construction is.

For cosmetics I couldn't find Vietnam-specific pricing no matter how much I looked, sources give MOQ only, no numbers. I used figures from a Guangdong factory in China as a rough reference and marked that separately in the interface from the general "example" label. It's not the Vietnamese market, and hiding that would be lying.

The demo isn't selling a finished app. It's proof I can build agents like this on a real working case, not just talk about "building bots," and it's also a way to test whether there's real demand for a system like this. If you're sourcing from Vietnam and want the same thing built around your own supplier list, message me, I'll build it for your categories and your workflow. I'm also running a closed pilot for a few brands who want the subscription itself rather than a custom build, message me for that too.

Try the demo: https://ilyashatalov.com/demo/sourcing

── more in #ai-agents 4 stories · sorted by recency
── more on @vietnam 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/i-built-an-ai-sourci…] indexed:0 read:4min 2026-09-30 · —