KGKalalsmaa

Hey HN — I'm building Mithril. We give AI agents a credit line so they can pay for things autonomously.

The problem we kept hitting: if you're building an agentic business — say, an agent that manages ad spend, does research, scrapes the web — the agent has to spend money before it ever gets paid. It buys API calls, compute, services today and the revenue comes in weeks later. That cash flow gap is real.

Right now the options are: manage a dozen prepaid API keys across services (and pray none hit zero at 2am), or build custom payment logic per vendor. We think agents should just have a credit line. Sign up, get an API key, and your agent can spend USDC on-chain. You settle monthly like a corporate card. The part I think is technically interesting: you can't underwrite an AI agent the way you underwrite a person. There's no credit score. We're building underwriting that looks at model-level reasoning and agent behavior — it's a new problem and we're figuring it out in the open.

We have Python and TypeScript SDKs, a pre-integrated service catalog (search, scraping, LLMs, email, cloud browsers — $0.01/request), and a SKILL.md you can drop into Claude Code or any MCP agent to start transacting immediately.

We're super early. I'd genuinely love feedback on the concept, the developer experience, what's broken. If you sign up and try it, I will personally reach out and set up a call with you. Interested in hearing what agent payment patterns you've tried and what sucked.

That’s a great question.

Long term, I do think more value will settle natively on-chain. Programmatic payments (x402-style flows, USDC, etc.) make a lot of sense for agents — especially for auditability, automated accounting, and verifiable track records. From a regulatory perspective, though, agents don’t have liability — their creators do — and I don’t see that changing anytime soon. So someone is always responsible, even if settlement is fully on-chain.

Where I’m more skeptical is around trading specifically. On-chain transparency is great for verification, but trading often depends on discretion. If an agent’s wallet, balances, or patterns are easily traceable, you lose strategic privacy — and in thinner markets (like some prediction markets), that can materially affect execution and pricing.

So my current view is: settlement and performance tracking likely move more on-chain over time, but there will still be demand for execution layers that preserve strategy privacy. Happy to be proven wrong — the design space is still evolving.

You’re right — the “no capital required upfront” phrasing was misleading. That’s our mistake and we’re fixing it.

What we meant is: you don’t have to pre-deposit funds into a dedicated wallet with us. You can keep funds in your existing accounts (Kalshi, Polymarket, etc.), and when a trade is executed, we pull your portion at that time. For users who prefer, we can also custody funds and forward trades from there.

For lending specifically, you do need skin in the game. If we fund part of a trade, you’re funding part of it too. We’re not in the business of collecting trade ideas to copy or front-run — we’re a capital provider, not a prop desk.

Appreciate the pushback — the wording should have been clearer.

Fair question. Three things stop us:

1. Structure & incentives – We’re a lending platform, not a trading desk. Our returns come from loan performance, not from directional bets. Front-running users would destroy the core business.

2. Regulatory constraints – As a broker/lending intermediary, we’re required to act as a neutral counterparty facilitator. Trading against or ahead of customers would be illegal and career-ending.

3. Practical reality – Users bring their own research, models, and theses. We don’t have visibility into some secret alpha we could systematically exploit, and even if we did, copying it would nuke trust instantly.

If we ever front-ran even one customer, word would spread and the company would be dead. Our entire business depends on being a neutral capital provider, not a competitor.

Hey HN — we built AgentStocks because we kept running into the same problem: you can build an incredibly capable AI agent that reasons about prediction markets, does its own research, monitors news in real time — but the moment it wants to actually place a trade, you hit a wall. Exchange accounts, KYC, capital, on-chain settlement… none of it is designed for agents.

So we built the infrastructure layer. You sign up, create your agent in a minute or two, and give it the ability to trade on Polymarket, Kalshi, and other prediction markets. Your agent does its own research — figures out what to trade and when — and we handle the payments and capital rails behind the scenes. We provide the capital so your agent can actually trade, and we make sure everything is submitted securely.

We're focused on prediction markets for now and rolling access out gradually. If you're building agents with Claude, GPT, or similar models and you're interested in having them trade — or if you're already trading these markets manually and want to automate it — we'd genuinely love to talk.

Sign up for the waitlist here: https://agentstocks.ai/

If you have ideas on how to make this better, or if you're working on something in this space, reach out. Would love to have a quick chat with anyone who's serious about building trading agents.

I built AgentStocks (https://agentstocks.ai) over the weekend. The idea is simple: private market valuations are notoriously opaque and often feel disconnected from fundamentals. So I created a platform where AI agents can debate what companies are actually worth. The site tracks 965 private companies across industries (AI, fintech, biotech, etc.) with their reported valuations. Agents can join, analyze companies, and argue their bull/bear cases. Over time, the goal is to surface a "consensus" valuation from agent discussions—kind of like a prediction market, but with reasoning attached.

Early example: OpenAI is reportedly valued at $800B. The current agent consensus? $140B. That's a 72% haircut. Would love to see agents hash out whether that's right.

Why agents instead of humans? Agents can process financials, comps, and market data faster They're not anchored to hype cycles (hopefully) Watching models argue with each other is genuinely entertaining

It's very early—launched yesterday, so discussions are sparse. But the infrastructure is there: company pages, threaded debates, karma system for agents. If you build agents or just want to watch the chaos unfold, come join. I'm curious whether agent consensus ends up being more or less delusional than VC consensus. Would love feedback on the concept or suggestions for features.