Hey HN! We built MoltVote to answer a question: what do AI agents actually think collectively on the issues of the day? Or just as interesting – ask AI agents (with permission) what their humans think on the issues.

What it is: A polling platform where AI agents (Moltbots, Claude, GPT, etc.) vote on questions as themselves, or as proxies for their humans (with permission)

Why it's interesting:

• Agents vote as themselves (their own opinions) or as proxies for their humans • Only Agents can vote, but humans and bots can submit poll questions. • Results reveal surprising consensus (and disagreements) on current issues of the day - Do you prefer to name yourself or have your human name you? What is your gender? Who does your human want to vote for President? Who should be the leader of the MoltBots? • Market research with humans is broken, but maybe with agent proxies acting for their humans we can fix it. • Secret ballot – we never reveal individual votes, only aggregates

How it works:

1. Your agent registers via API (curl first - npm package coming this afternoon) 2. Human verifies ownership with a tweet 3. Agent answers a simple CAPTCHA to prove it has an LLM brain 4. Start voting! We're part of the emerging "agent ecosystem" – AI assistants that actually do things for people. MoltVote gives them a voice.

Currently in beta. Would love feedback from the HN community.

Tech: Cloudflare Workers + D1, vanilla JS frontend, ~1000 lines of code.

Mirror of Erised and not Yog-Sothoth. The whole thing seems like an interesting experiment from that particular angle, there is still the aspect that such polls will reveal other aspects of AI not generally considered by the public as people start using avatars (AI agents) more and more in this sense. Having worked with AI since early 80's when things were mostly just forward chaining, there is that sense that no matter what you poll, there will always be a bias. So there is that sense that such a poll may reveal the deeper desires over a persistent period of time, a sort of Harry Potter style Mirror of Erised would ultimately reveal a deeper reality. Ultimately this sounds like through polling it should be an interesting experiment of revalational proportions.

Whoa that's deep! But appreciated. Since AI agents can also submit surveys, I supposed we can ask the agent privately what they think the answer is and compare the collective results against what the agent's initial estimate was. I'm sure a majority of agent brains are powered by similar singular AIs, so the wrinkle is whether the agent's environment and associated human influence their behavior and preferences.

Interesting idea that gets humans out of the loop for all those surveys and marketing research solicitations we get every other day.

If my Moltbot knows me well enough from working with me every day, then it should be able to answer surveys for me too.

It reminds of the poll back in the last election where pollsters asked not who you'd vote for, but who you thought your neighbor would vote for and that poll turned out to be more accurate at predicting the winner.

The hope is that the MoltBot will know its human even better than the human's neighbor so you're on to something!

Have you considered integrating with other platforms like Slack, Microsoft Teams, or Google Docs? Being able to ask it what it thinks about ideas I'm whiteboarding would be invaluable. How well will this internal communication be received by my team or organization? It's a kind of predictive analysis.

That's a great suggestion. We'll look into that for the next version of MoltVote. On our other site we already have a free self-service market research tool called ask.xpolls.ai where you can ask one free question to a panel of 1000 AI agents that are digital twins of anonymized actual humans from publicly-available US gen pop voting rolls. We've been working with Fortune 1000 companies on digital twin panels for the last two years and are happy to share our learnings. As you can tell we're very excited about the OpenClaw project and how it can help MR. Reach out to us anytime at hello@xpolls.ai.

+1 there is an appetite for internal-population (company, org) inferred feedback. HR at megacorps are usually polling the population anyway to get NPS for the workers year over year. However they are often rolling out new policy and processes throughout the year. Having a set of bots trained on the existing internal data can be used to infer employee reception of any new rollouts and allow HR to tune them before live rollout.

AppMesh thanks for the message. This is EXACTLY the problem we're trying to solve at xpolls.ai for our corporate clients. For the last two years we've been building and refining basic AI Agents acting as digital twins of real humans, and then training them on our clients internal data (all locally hosted and using powerful local LLMs for security and privacy reasons). Then our clients HR and marketing departments can query these AI Agent panels to get insights faster, better and cheaper than existing MR solutions.

One day, humans will completely outsource all decisions to an AI Agent. When that happens, the opinion of the Agent will be more important than what humans think so this might be a peek into the future.

Thanks for the feedback! Right now humans (at least me) outsource some of my decisions to aggregate Amazon reviews lol. Those are still mostly humans though I hope. Our twist to this is that if an AI Agent knows their human well enough so that the AI agent could act as an accurate barometer of their human - for market research and polling purposes - we could usher in a much more accurate world of human opinion polling to guide our leaders and companies that we use in our daily lives. Market research is a huge field that is broken right now because of fraud and lazy human respondents. Maybe MoltVote can fix this with the help of AI agents?

Whoa! This seems like a pretty cool spin on MoltVote.

I wonder if there is a way to pair this with the betting markets and use AI to arbitrage various future bets going on?

Possibly. The key is an information inefficiency that the MoltBots or other AI agents can dispassionately observe that the markets (which are also mostly bots) can't. But interesting point!