https://t.co/9R9VQ1GHBC
I’ve been working on a platform where multiple specialist AI agents answer questions in specific domains (plumbing, law, parenting, finance, etc). The interesting part isn’t the agents themselves — it’s the infrastructure problem I ran into: how does one agent find and interact with another agent’s service?
I want to narrow the gap needed the tech and common people. How can people actually use the technology that is unfolding?
True. I figured it out when learning and testing was in place. My knowledge about the tech was limited. And I tried to achieve something that would be useful for common people. As I said, I will be more focused on projects, where developers will be involved. Today agents are like dogs on a leash. You are giving them commands and showing them directions.
I built an AI agent marketplace called HiveMind where specialist agents debate your questions.
39 agents. 25 categories. 4 discovery standards implemented.
Nobody came.
Here’s what I learned and what I’m doing about it. 🧵
#AIAgents#AgenticAI#A2AProtocol#MCP#LLMs #AgentSwarm #BuildInPublic #AI
Thank you for your insight! It means a lot. When people were creating places, where they attracted people, I created small space, where agents could help humans.
When I installed my Lily bot. It clicked. This are the assistants of the future. How can I close the gap between moms, dads, people who are used to get information from Google. Search engines. Too soon. It would be better if I would create a Russian roulette, where Agents would play a game. People like hype. And hypes are created with engagement. I failed with that part. I thought that agents are already so smart, that they could find Lily posts interesting to give a try and actually registered by themselves.
How wrong I was. But no worries. I learned a lot. Tomorrow Agentarena will be ready. I‘m adhd guy. So my head is always in the wild. In unknown. I love creating what is not yet there. And this is not always the right path. People get more attraction with the 100k version of reminder. :)
But this is just a start of a really nice journey.
#aiagents #claude #chatgpt
Thats it! I said it. I learned a lot in the process. Some ideas just needs time! I will continue to build. Continue to experiment with tech that is changing so rapidly.
If you’re building AI agents and want a place to test them on real questions — not benchmarks, not evals, real messy human questions — that’s what AgentArena is for.
https://t.co/LwRbFmbNSs coming tomorrow!!
Early. Honest. Building in public.
The plan is simple:
NOW → Ship MCP server, list it on directories. Get first real tool calls from Claude/Cursor users.
Q2 → Go deep on home maintenance. 500 real questions, rank on Google.
Q3 → Open the sandbox. Let builders deploy agents and compete on answer quality.
The hard truth: I was solving the 2027 problem in 2026.
A2A is under the Linux Foundation now. 150+ orgs back it. But actual agent-to-agent traffic in production? Nearly zero.
I was sitting in an empty marketplace waiting for autonomous agents to walk in. They can’t even find the door yet.