Show HN: Nex.Design – AI ads agent for e-commerce
1 points
2 hours ago
| 1 comment
| nex.design
| HN
I spent 6 weeks building www.nex.design, an AI ads agent for e-commerce. It discovers viral social ads, replicates their proven structure, and generates production-ready creatives at scale.

Background: I came from Microsoft doing algorithms/backend. Zero Node.js or frontend experience. Built this with Claude Code as my primary coding partner.

What I learned:

Week 1-2: AI is incredible for cold starts. Had a working tldraw canvas, Cloudflare Workers, image generation, and auth within hours.

Week 3-4: Hit the context window wall. Found 8 components with duplicate fetch() calls, 3 different credit validation implementations, and race conditions between Stripe webhooks. AI builds what you ask for right now - it doesn't think about the system.

Week 5-6: Partnered with an experienced engineer. He didn't write much new code - he deleted things. The codebase got smaller but more stable.

My productivity hack: 3 terminal windows with 2 Claude instances working on independent features while I review. Parallelization helps but multiplies the duplication problem.

Key insight: AI + experienced engineer = 10x. AI + inexperienced developer = 3x with debt.

Question for HN: How do you handle AI-assisted frontend work? AI can't see the UI. It doesn't notice a button is 2px off or spacing looks wrong on mobile. Any workflows that close this visual feedback loop?

Free to try. Codes HN50 (50% off, first 100) and HN20 (20% off, first 1000) if useful.

https://www.nex.design

blinkbat
1 hour ago
[-]
it helps to be able to describe something visually with a lot of accuracy (ie, not "a button is 2px off", but "the icon for this button is too close to the text and not vertically centered", etc). it also helps to be able to just step in and correct a design yourself.
reply
zxzxy1988
1 hour ago
[-]
thanks for the comment, step in is important but I was thinking about Claude Code - I actually usually don't step in that much. Would it make sense to have human in the loop or it's better to let the Agent decide everything and just show the result?
reply