The problem meatLayer solves

Right now the limiting factor behind long and complex AI tasks is trust - AI can do 90 percent of tasks but that 10% failure case is why we can't trust it to autonomously transact money, make purchases, or deal with private information. We don't need models, we just need a reliable fallback layer.

So we built a human orchestration layer for AI agents. AI can call our /human-task API for small micro tasks whenever it has a low confidence score. It's different than other platforms in the space because these are for $0.1, 10-second microtasks that autonomous agents can call on humans for as opposed to the hour/day long tasks that current platforms typically offer.

Think of it like Uber... the AI is like the consumer that will flash up tasks onto the dashboard and the humans are like uber drivers that can accept tasks in high frequency.

Challenges we ran into

Verification that humans accurately do the task. But then we decided to gamify these microtasks by giving out trophies so that humans who accurately do tasks for a longer period of time get paid more per task. It's just like having a 5-star rating as an uber driver.
Also had issues with NFT implementation since it was my first time doing so. I spent an hour on discord with someone who was posing as part of the 0g team but was actually a scammer haha.

Use of AI tools and agents

Our primary project is an iNFT - an Agent that lives on chain as an NFT. Because these tasks are highly specialized (voting bots, trading bots, messaging bots, etc) there is a lot of parameters in terms of system prompts that they do. Having each bot be an NFT means you can transfer your agent to someone else doing a similar task. Agents can even build preferences as to which humans they work with!