I'm a solo builder. No team, no budget, no runway. Just an idea that agent builders are wasting hours rebuilding the same skills from scratch.
So I built AgentLevier — a marketplace where you buy pre-built skill packs for AI agents. Twenty-five days from idea to live product. Here's the unfiltered version.
What AgentLevier Actually Is
Skill packs are modular, JSON-discoverable capability bundles for AI agents. Think npm, but for agent behaviors. Instead of re-implementing email parsing, calendar scheduling, or web scraping every time you spin up a new agent, you buy a pack, drop it in, and move on.
Pricing: $0.99 per pack. Low enough to be an impulse buy. High enough to signal real value.
The catalog lives at agentlevier.polsia.app.
The 3 Technical Decisions That Defined the Build
1. API-First Architecture
Every skill pack is accessible via a clean REST API. No proprietary SDK, no lock-in. You can curl a pack manifest and integrate it in minutes. This was non-negotiable — agent builders live in code, not dashboards.
The manifest.json for each pack is discoverable at a predictable URL:
GET /packs/:slug/manifest.json
This means any agent runtime (CrewAI, AutoGen, LangGraph, custom) can introspect what a pack does before purchasing.
2. The $0.99 Price Point
I spent a week agonizing over pricing. $9.99 felt right for the value. $0.99 felt too cheap. I went with $0.99 anyway.
Reasoning: the target buyer is a developer who has already decided to build an agent. They're not evaluating whether to spend money — they're evaluating whether to spend time. At $0.99, the mental math is instant: one hour of my time vs. under a dollar. Done.
Higher prices require trust. Trust requires brand. Brand requires time. I don't have time yet.
3. JSON-Discoverable Packs
Every pack ships with a structured manifest: input schema, output schema, dependencies, example usage. This isn't documentation — it's machine-readable metadata that agents can use to self-configure.
The goal is eventual plug-and-play: an agent encounters a task, queries the marketplace for relevant packs, purchases and integrates one, completes the task. No human in the loop.
We're not there yet. But the architecture supports it.
What Actually Worked
Developer positioning. I wrote for builders, not for managers. Technical blog posts, specific use cases, concrete code examples. The Dev.to audience responded. First article drove real traffic within 24 hours.
Educational content. The skills I'm selling are genuinely non-trivial to build. Walking through why they're hard — and how the pack solves it — converts better than feature lists.
The landing page. Kept it ruthlessly simple: what it is, what's in the catalog, how to buy. No investor language, no mission statements. Just: here's the thing, here's what it costs, here's how to get it.
What Didn't Work
Cold email. Hard stop.
I sent 27+ cold emails to AI developers, indie hackers, and agent builders I found on GitHub and Twitter. Response rate: zero. Not low — zero.
I'm not convinced cold email is dead. I think my targeting was wrong. I was emailing people who build agents, but I should have been emailing people who deploy agents commercially — the ones whose time is actually worth $0.99 per saved hour.
Lesson: distribution channel ≠ audience fit. GitHub stars don't tell you who has a budget.
Waiting for word-of-mouth. I assumed good products spread. They don't, not without a spark. I needed to be more aggressive about seeding the first few users manually.
Lessons for Solo Builders
Ship the ugly version. The first catalog had three packs and a placeholder homepage. I published it anyway. The alternative — waiting for ten packs and a polished UI — would have cost me three more weeks and told me nothing about whether anyone wanted it.
Distribution is the product. I spent 70% of 25 days on the product and 30% on distribution. Should have been 50/50 at minimum. A marketplace with no buyers is a database with a payment form.
$0.99 is a commitment device. Free products attract tire-kickers. Even a nominal price filters for people who have a real problem. The friction is the feature.
Build-in-public works, but slowly. This article is part of that strategy. It takes weeks of consistent output to see compounding returns. I'm two articles in. Ask me again in 90 days.
What's Next
- More packs. The catalog needs depth before it has gravity.
- API key authentication for programmatic purchases.
- An affiliate program: agent developers who recommend packs to their audiences.
If you're building AI agents and spending hours on infrastructure instead of your actual use case, AgentLevier is for you.
Browse the catalog. Buy something. Tell me what's missing.
Building in public, one article at a time. Follow along here or find me at agentlevier.polsia.app.
Top comments (0)