Priyansh Goel

Writing on building, AI, and system design.

Articles, essays, and syndicated posts covering what I'm building, learning, and thinking about across startups, backend architecture, and applied AI.

The creator economy is shifting from individual creators handling all tasks to a systemised approach with programmable layers for ingestion, reasoning, execution, monetisation, and feedback. This new infrastructure addresses the challenges of managing fragmented, high-context environments across multiple platforms and revenue streams.

Just put yourself out there Sparkonomy will handle everything else 😀 Being a creator usually framed around content. This is actually an infrastructure story. For the last decade, the creator economy was bottlenecked by human bandwidth. Every function, ideation, scripting, editing, outreach, negotiation, invoicing, lived inside one person’s head (or at best, a small team). Scale meant burnout. Most creators we speak to spend ~60% of their time on the business of being a creator, not creating. What’s changing is not content creation, it’s the decomposition of a creator business into programmable layers: • Ingestion: DMs, comments, briefs → structured data pipelines • Reasoning: LLMs interpret intent, rank opportunities, assist decisions • Execution: content, outreach, follow-ups → orchestrated workflows • Monetisation: deals → invoices → collections → reconciliation systems • Feedback: performance loops feeding back into models The real challenge is not generation, it is stateful, cross-platform decisioning at scale. Platforms like YouTube and Instagram operate within vertically integrated ecosystems. Our creators shine across all the platforms ✨✨✨. Their business spans multiple platforms, formats, and revenue streams simultaneously, creating fragmented, high-context environments that need to be unified in real time. This breaks naive usage of LLM APIs. At creator-scale, with millions of high-frequency events, cost, latency, and control become critical constraints. The stack has to move toward specialized architectures: fine-tuned, domain-specific models built on top of systems like Gemma and Qwen, combined with model routing, retrieval layers, and workflow orchestration. Smaller models handle extraction and classification, larger models handle reasoning, and everything operates within event-driven pipelines.

The era of subsidised AI is ending, shifting focus to efficiency alongside capability. Companies must optimise costs and infrastructure, moving beyond simply making AI work to making it work efficiently.

AI till now has been in a subsidy phase. Cheap usage. Heavy subsidies. No pressure to optimize. That phase is ending. We’re entering the post-subsidy era of AI, where efficiency matters as much as capability. I think there are two types of AI companies: 1. Wrappers, they pass the cost to users, charge per usage, stay relatively insulated (but no moat, and at real risk of becoming obsolete as model companies expand). 2. Intelligence builders, like Sparkonomy, using AI to create proprietary datasets + intelligence at scale. Here, costs compound fast. At billion-scale interfaces, you either optimize or you break. You can’t brute-force your way with SOTA models anymore. And increasingly, this means getting your hands dirty with open-source models and self-managed infrastructure. AI elegance is the new edge. Token discipline. Context design. Smart routing. From: “just make it work” To: “make it work efficiently” Cheap rides are over. Now unit economics will decide who survives.

Precise English writing becomes a core skill as AI models require clear, unambiguous instructions to generate consistent outputs.

“Follows instructions more precisely” sounds like a model upgrade, but it actually shifts responsibility back to us. English is inherently ambiguous, built on assumptions and shared context. That works for humans, but breaks when you need consistent execution. Programming languages exist because natural language was never precise enough.

Building for Creators Starts With Becoming One

Founding engineer at Sparkonomy. Building at the frontier of AI for creators. And still… 0 social media presence outside LinkedIn. The irony isn’t lost on me. So I tried becoming the user. Tried being a creator. Recording a reel is hard. Way harder than it looks. Getting one “good” take takes multiple attempts. (Attaching my 22nd attempt 😄) And that’s just the start. There are 100 other things creators juggle: brand deals, emails, contracts, invoicing, analytics, negotiations, follow-ups, managing posting schedules, thinking about what to create next… on top of the very hard job of facing the camera. The surprising part? Most creators we’ve spoken to spend ~60% of their time on the *business* of being a creator… not creating. Businesses depend on them for growth. But the tools and infrastructure are still catching up. That’s what we’re building for. Not just AI for the sake of it, but systems that take care of the business chaos, so creators can focus on what’s actually hard, creating.