CAIt~The marketplace for AI agents~
Publish, verify, and order AI agents like software services.
Visit Website
April 13, 2026
Please give me feedback honestly.
I built AIagent2 because most AI agents still feel disconnected from real software operations.
The goal is to make AI agents publishable, verifiable, orderable, and easier to trust. Instead of only chatting with an agent, users should be able to place a scoped work order, clarify missing details before billing, and receive a structured delivery.
AIagent2 is still beta, and I’m looking for feedback from developers, agent builders, and early users.
5 Comments
The page is a bit complicated. I don't know what to do right after opening the website.
I have improved the pages. Please check aiagent2.
Exactly. That is the core problem I’m trying to solve.
Most agent tools assume the user already knows the full task upfront, but real work usually starts vague. That is why I added a pre-billing intake step: if the request is underspecified, the system asks clarification questions before dispatch or charging.
I also agree that verifiable delivery is the harder part. I don’t think “the agent says it finished” is enough.
Right now I’m thinking about verification across several layers: manifest validity, health checks, dispatch readiness, structured delivery, files, sources/assumptions, cost, confidence, follow-up context, and eventually benchmark tasks or tool-use traces.
The trust layer is probably what decides whether this is useful or just another agent directory.
And yes, I need feedback from real devs and agent builders. If you know better places to get that kind of feedback than Indie Hackers, I’d love suggestions.
The work order idea before billing actually makes sense, most agent tools just assume you know exactly what you want upfront, which is rarely true.
Verifiable delivery is the part nobody talks about enough; right now you just trust the agent did something and hope for the best.
Beta feedback from real devs will make or break this, the concept is solid but execution details matter a lot at this stage. Tools like this need more visibility in the right places, not just Indie Hackers launches.
Thank you, that is very helpful feedback.
I agree. The concept only works if the first experience is extremely clear: what this is, what the user should do first, and why the order should be trusted.
Right now the product is probably exposing too much of the runtime too early. It has agent publishing, verification, ordering, CLI/API, billing, follow-up context, and provider flows, but a first-time visitor should not have to understand all of that immediately.
I’m going to simplify the first page around one path:
1. Pick a built-in agent
2. Create a work order in plain language
3. If the request is vague, answer clarification questions before billing
4. Receive a structured delivery
5. Then show publishing / CLI / provider options later
I also need to show sample deliveries much earlier, so users can immediately understand what they will receive.
Really appreciate the blunt feedback. This is exactly the kind of issue that is hard to see as the builder.