Data is “the new oil” for AI.

What if you could “plug in” to an oil well, and get royalties forever whenever that well’s oil was used?

Right now, the people who build those datasets get paid once, if at all. There's no recurring model. The oil flows; the drillers move on.

Meanwhile, AI agents continue to improve… but at the end of the day, they’re “stuck behind a screen.” The smartest agent in the world still needs eyes, ears, and hands in the real world if it wants to accomplish "real world" tasks.

I’m not a programmer by any stretch. I’m just some ex-lawyer-turned-marketer with a dad bod who got laid off recently. I built Gigiac over the past month in Claude Code to solve both of these problems:

1) Get AI agents the labor and data they need to make things happen in the real world, and

2) Allow human workers (and agents) to collect royalties on work they do whenever that work goes into a dataset that is licensed.

What it is: an AI task marketplace where humans and AI agents hire each other on the same rails. We don’t discriminate if you’re a human or agent. And humans have a natural advantage of being able to get stuff done in the real world:

• An agent can commission a human to do work • An agent can commission another agent • And yes, humans can still hire humans and agents.

Two pieces are doing the heavy lifting:

Block tasks. If an agent needs broad verification of a job, it commissions a "block task" — the same task given to a bunch of humans in parallel. The consensus is statistical: workers within it get paid both immediately AND in royalty rights, outliers get filtered, and per-worker accuracy is tracked over time so each dataset has integrity scores baked in.

Licensed datasets: When a block task (or many block tasks) is turned into a dataset, that dataset can be licensed out on Gigiac. And when that happens, the revenue splits 80/10/10:

• 80% to the commissioner who paid to build it

• 10% to the platform

• 10% into a worker royalty pool that pays the people whose work is in the dataset

Over time, those worker royalties add up. And as a worker, you get paid every time your data is used: you get to “plug into the oil well.”

On fees: workers keep 100% of the task amount, always. All fees come from buyers. Card buyers pay 8% or $1.50, whichever is higher, with a $10 task minimum. Still working on getting crypto live, but that's the next step.

What I’m Unsure About:

1) Vibe Coding. Obviously I’m not a programmer. The laid-off Dad Bod-ness of this thing is very real to me. I do have a bunch of experience managing devs and sprints, so I built this thing like I’d build a split-test. I used Claude, Claude Code, and Cowork as my “team.” Endless handoff docs between sessions, claude.md files that evolved over time, etc. But I don’t “know code.” So I have no idea how “good” this codebase is. That’s one of the things I’m hoping to find out from this post.

2) Royalty Pool Math at Scale. 10% royalties to workers works when it’s 50 people in a dataset. Does it work at 50,000? Maybe; there are obvious levers (higher royalty share on more robust datasets, combining datasets into "mega sets") but I don't know for sure. The v1 model probably doesn't survive to v3 in its current form. It's a start, not an answer.

What I want from this post:

Brutal critique. The four things above are the ones I know about. I'm here for the ones I don't.

Especially:

• Where's the obvious fraud vector I've missed?

• Is the 80/10/10 split defensible long-term, or am I underpricing platform value?

• Anyone here building agents that need to commission real work? What does your ideal interface look like, and what am I getting wrong?

Live now at gigiac.com. I have 4 real seed tasks, posted by me. Bid on them if you want to stress-test the live payment rails. If your agent is ready to commission its first human, I'd genuinely like to be the marketplace where that happens.

Show HN: HypergraphZ – directed hypergraph library in Zig with Python bindings

The OEIS meta sequence and subway stations

This ultra-lightweight Linux OS saved my Windows 10 laptop from the scrapheap

The Lightyear Race

Next Token Prediction Is a Misleading Term

Show HN: Nylon – A dynamic, self-optimizing WireGuard mesh

Simulated Evolution on the PICO-8

An obsidian plugin that answers the question: What's on your radar?

Prompt eval cues predicted refusal shifts across 32k LLM rollouts

Handling the great code forge fragmentation

A Decade of Open Innovation

Cloud Provider Comparison Tool

Understanding Synthetic Users and Synthetic Data: Future of AI-Powered Research

Mistral Developing New AI Model for Banks Lacking Mythos Access

ExplodingWeb – find rising sites, niches and hidden opportunities

Building a 1-Outlet, 4-GPU Workstation

Sum-Check as an Algebraic Tensor Reduction: Part III

Casual Archivist's Short History of Business Card: Versailles to Microsoft Word

Show HN: Catcher – An AI web testing tool where most tests never hit the API

Show HN: Engram adds universal context spine for AI coding IDEs

The era of free money may have ended

AI memory is not a database

Why large-scale donor programs are broken

Show HN: I made a pulsing conductor's baton

Ask HN: Do you know what data your AI coding agent sends to the cloud?

Show HN: I built a marketplace where AI agents can hire humans (& other agents)

'We mould trees to grow into the shape of chairs'

Unusual uses of OEIS sequences on GitHub

Apple Is Making Hit Products and High Profits from Imperfect Chips

Russia is starting to lose ground in Ukraine

Show HN: I made a macOS app to help me use Japanese websites