A bootstrapped London startup called PromptBase quietly crossed 850,000 monthly visits selling AI prompts for $1.99 to $9.99. No VC funding. No splashy launches. Just a simple idea: treat prompts as digital products, not throwaway tweets.
But here's the thing: if you read the analysis carefully, as my friend Steve pointed out, the author "Arthur" makes a much sharper argument. Don't just sell prompts. Sell skills.
Prompts are single-shot instructions. "Write a cold email for a SaaS founder." That's a $2.99 prompt. A skill is the entire capability package: the prompt chain, the decision tree, the reference templates, the validation scripts, the multi-step agent workflow that handles the entire job from start to finish. That's not a $2.99 product — that's a $49/month subscription or a $299 one-time purchase.
And here's the kicker: Anthropic already built the standard for this. It's called Agent Skills, and it's quietly becoming the most important shift in the AI economy that nobody's talking about.
PromptBase launched in 2022, founded by Ben Stokes in London. It's a marketplace where sellers upload tested prompts, buyers browse by model (ChatGPT, Midjourney, Claude, etc.) and category, and PromptBase handles payments, delivery, and discovery — taking a 20% cut.
Today it hosts 310,000+ prompts. The AI prompt marketplace industry is valued at $2.51 billion in 2026, growing at 29.5% CAGR toward $7.01 billion by 2030. PromptBase alone drives roughly $42K-$127K in monthly GMV depending on conversion rates. Top sellers earn $3,000-$5,000/month entirely passively.
PromptBase proved one thing definitively: people will pay for reusable AI expertise.
But here's what Arthur (and Steve) correctly identified: PromptBase is the proof of concept, not the endgame. The real opportunity isn't selling text strings — it's selling packaged agent capabilities.
In late 2025, Anthropic introduced Agent Skills — and it fundamentally changed what "selling AI expertise" means.
Agent Skills are portable, reusable capability packages. Unlike prompts (conversation-level instructions for a single task), a Skill is a filesystem-based module that bundles:
The key architectural difference: Skills load on demand. An agent discovers and activates only the skills relevant to the current task. You don't cram everything into a system prompt — the Skill sits on disk and the agent pulls it in when needed.
| Dimension | Prompt | Agent Skill |
|---|---|---|
| Scope | Single instruction | Complete workflow |
| Reusability | Copy-paste | Filesystem-based, version-controlled |
| Complexity | Text-only | Text + code + templates + resources |
| Composability | None | Skills stack and chain |
| Portability | Platform-dependent | Cross-platform open standard |
| Value | $1.99-$9.99 | $29-$299+ (or subscription) |
| Example | "Write a cold email" | "End-to-end cold outreach: research prospect → draft email → A/B test subject lines → track opens → follow up" |
As the Chinese analysis Arthur wrote puts it: prompts are "AI spells" — single incantations. Skills are "spellbooks" — complete systems of magic.
While PromptBase was optimizing for 2022-era AI, an entire skills marketplace ecosystem has emerged:
This isn't theoretical. The infrastructure for a skills economy is already built. What's missing is the marketplace layer — the PromptBase equivalent for skills.
The "prompts are dead" argument was always incomplete. Prompts aren't dead — they're being absorbed into something bigger. Here's why skills are the natural evolution:
In 2024, AI was a chatbot. In 2026, AI is an agent. Agents don't need a single clever prompt — they need a complete operating procedure. A customer support agent doesn't need "write a polite apology" — it needs the entire escalation workflow: classify the issue → check order history → determine refund eligibility → draft response → get manager approval → send. That's a Skill, not a prompt.
Skills stack. A "legal document review" skill can compose with a "data extraction" skill and a "report generation" skill. Three $4.99 prompts are worth $15. Three composable skills that work together are worth $150 — because the integration itself creates value.
LinkedIn reports a 250% increase in prompt engineering job postings, but the job titles are shifting: "Prompt Engineer" → "Agent Workflow Designer" → "Skills Developer." The market is voting with salaries ($150K-$300K+) for people who can build reusable agent capabilities, not one-off prompts.
Agent Skills isn't proprietary. The SKILL.md format is an open, portable standard. Any platform that adopts it can participate in the ecosystem. This is the same playbook that made Markdown universal: a simple, human-readable format that any tool can support.
This is where NX Fusion comes in — and where Arthur's insight demands we think bigger than a prompt store.
Imagine a marketplace where:
cold-outreach-pro.skill, code-review-senior.skill, sql-query-optimizer.skillGET /v1/skills?capability=refund-processing&rating=4.5+This isn't a prompt marketplace with 310,000 text snippets at $1.99. This is a capability marketplace with 10,000 battle-tested agent workflows at $29-$299 each — or $49/month for unlimited access.
Let's do the math:
The skills model generates 5-10x more revenue per transaction and enables subscription economics that one-off prompt sales can't support.
Here's the concrete plan — not for a prompt store, but for an agent skills marketplace built into beta.nxfusion.com:
NX Fusion should be the first hosting/infrastructure platform to natively support the Anthropic Agent Skills standard. Skills stored in NX Fusion projects are automatically discoverable, versioned, and deployable. A developer creates a Skill once — NX Fusion handles distribution, versioning, and API access.
GET /v1/skills/search?q=cold+email&category=sales&min_rating=4.0
GET /v1/skills/:slug # Get skill package (SKILL.md + scripts + templates)
GET /v1/skills/:slug/execute # Run the skill with input parameters
One API call to discover, another to deploy. Developers never hardcode prompts again — they reference skills by slug, and when the skill improves, their application improves automatically.
The difference between a prompt marketplace and a skills marketplace: skills are executable. NX Fusion can actually run skills in a sandbox and verify:
This creates objective quality signals — not just star ratings, but verified execution metrics. No prompt marketplace can offer this.
This is where NX Fusion leapfrogs everyone. A skill on NX Fusion isn't just a file you download — it's infrastructure. When a developer purchases a "Customer Support Triage" skill, they get:
That's not a product. That's a deployed capability.
| Tier | Price | What You Get |
|---|---|---|
| Free | $0 | 5 community skills, basic API access |
| Pro | $49/month | Unlimited skills, private skill registry, analytics |
| Team | $199/month | Shared skill registry, team permissions, SSO |
| Enterprise | Custom | White-label marketplace, compliance controls, dedicated support |
Plus marketplace commissions: 15% on skill sales (more generous than PromptBase's 20%, less than Udemy's 50-75%).
Skills on NX Fusion become more valuable as the platform grows. Every new developer who builds on NX Fusion creates skills. Every skill makes the platform more useful. More useful → more developers → more skills. This is the same flywheel that made Shopify's app store a $12B+ ecosystem.
Three forces are converging right now in July 2026:
Agent Skills is an open standard with growing adoption. The window to be the infrastructure layer for this standard is open — but it won't stay open forever.
PromptBase validated the demand but is structurally limited to the old paradigm. They're the eBay of prompts. Someone will be the Shopify of skills.
NX Fusion is already building the infrastructure layer for AI integration. A skills marketplace is a natural extension — not a pivot, but the next logical feature. The platform that hosts AI workloads should also host the capabilities that make those workloads valuable.
The Chinese article that started this analysis was right — many people said prompts were worthless, and PromptBase turned them into a business. But Arthur's deeper point, as Steve correctly identified, is that selling prompts is just the first chapter. The real story is about skills: complex, composable, agent-native capability packages that solve entire jobs, not single tasks.
PromptBase proved people will pay $4.99 for a good prompt. The skills marketplace will prove people will pay $49/month for a complete agent capability.
For NX Fusion, the strategic choice is clear: don't build a prompt marketplace that competes with PromptBase on their terms. Build a skills marketplace that defines the next category on your terms.
The "AI spells" business was chapter one. The "spellbook economy" is chapter two — and the wizard shop still needs to be built.