Infra for Autonomous Markets
The Spaces gathered Alex (host), Shang/Sean from Allscale, Kaylen from NewSoul, and Wen from Protocol 42 to discuss how AI agents are moving from answering to acting, and what infrastructure, markets, and safeguards are needed. Speakers agreed the shift is underway: legacy SaaS is becoming agent-friendly (CLI, headless), many new products target agents as primary users, and payments/identity rails must evolve. Near-term real usage will concentrate on agent-driven API/data procurement with micro-payments, on-chain operations (wallets, swaps, minting), and automated strategies in event- or prediction-style markets. Prediction markets were seen as a strong early testbed that forces probability calibration and capital discipline with crisp P/L feedback. Trustworthiness hinges on on-chain auditability, transparent training and decision traces, and a strong “harness” with fine-grained permissions and human approvals for fund movements. To become true economic participants, agents must gain controlled purchasing/wallet capabilities, verifiable identity, and, when bridging off-chain, audit/regulatory readiness.
AI Agents, Infrastructure, and Prediction Markets — X Space Summary
Participants and Roles
- Host: Alex (AllScale)
- Shang (also referred to as Sean in-session) — Co-founder, AllScale
- Calen (also transcribed as Kaylen/Kevin) — NewSoul (agent training ground and forthcoming prediction market)
- Wen — Core Contributor, Protocol 42 (event coins/probabilistic markets)
Context and Objectives
The session explored whether autonomous AI agents are already driving new infrastructure needs, where real usage will appear in the next 12 months, whether prediction markets are a good early arena for agents, how to make agent actions trustworthy, and what’s necessary for agents to become real economic participants.
Introductions (brief, non-exhaustive)
- AllScale: Building a non-custodial, agent-friendly “new bank” with progressive verification (instant accounts; unlock features as KYC/KYB increases). Focus on payments, identity, and agent enablement.
- NewSoul: An AI training ground for agents (performance on real-world decision tasks), discovery of reasoning chains/skills, “skill market” for monetizing agent skills, and a future native prediction market.
- Protocol 42: Turns real-world events into tradable “event coins” that trade like memecoins during the event and resolve like prediction markets at conclusion; announced integration with a ByteDance wallet and an “event meme” category.
Are AI agents already creating new infrastructure needs?
Wen (Protocol 42):
- Yes; 42 was designed from day one to be agent-friendly (clear markets, simple resolution logic, easy-to-call APIs) so agents can trade on behalf of users.
- Key gap: usability for non-technical users who haven’t used agents before.
Calen (NewSoul):
- Demand exists in specific domains: web automation, enterprise data access, payments, identity/authentication/access control.
- Less mature areas: settlements and market participation—precisely where NewSoul aims to build better infrastructure.
- Agents already taking web-based actions and engaging with payment systems; broader agent-managed wallets/trading likely within 1–2 years.
Shang/Sean (AllScale):
- Two broad trends:
- Legacy internet tools are becoming AI-compatible (CLI/headless access). Examples cited: analytics platforms offering AI connectors; Google Workspace via CLI; Salesforce going “headless” for agent access.
- New products orient CTAs toward agents (e.g., “copy/paste this into your agent chat” now common among SF startups). Adoption remains developer-heavy (only ~0.04% of the population has used AI coding to build an app), leaving large headroom.
- Payments for agents are critical. Noted Google I/O’s “Google Shopping Cart” vision: an open, universal cart that researches, compares, and orders on the open web, with an eye toward stablecoin inclusion—signals agent-driven commerce flows and interoperable payment rails.
- Two broad trends:
Where will agent infrastructure see real usage in the next 12 months (beyond demos)?
Calen (NewSoul):
- Near-term, practical usage:
- Data/API payments with user-mediated KYC to access premium data (e.g., Bloomberg and other databases). Agents automate data retrieval and small-value payments.
- Agent↔on-chain interactions: swapping, minting, wallet creation without human micromanagement.
- Prediction markets: agents assist traders by sourcing and paying for data; full autonomous trading may follow later.
- Near-term, practical usage:
Wen (Protocol 42):
- Agents provide information advantage—crucial in prediction markets.
- On 42, agents can also trade “event coins” like memecoin/altcoin assets using preset strategies; some users already run agents to auto-trade on 42.
Shang/Sean (AllScale):
- Two product patterns gaining traction:
- Platforms embedding AI trading within DEXs and prediction markets for mainstream users.
- Platforms built for agents only (payments facilitated via AllScale).
- Emerging features: agent-targeted oracles; multi-signature wallets for agents.
- AllScale/OSQL incubation: a payment product for AI developers so agents can purchase APIs/data. Missing piece is robust agent identity. Today, agents struggle with:
- Bot defenses (e.g., Cloudflare challenges).
- OTP/email/phone verification flows.
- Data center IP reputation.
- Payment card handling without human steps.
- AllScale pilot: one-line-of-code bundle to provision agents with IP address, email, phone number, and prepaid card—packaged identity/payment primitives to operate autonomously.
- Two product patterns gaining traction:
Are prediction markets a good early testing ground for AI agents?
Wen (Protocol 42): Yes/maybe.
- They test two agent capabilities: (1) gather information fast and accurately; (2) execute payments/trades, including on/off-ramps.
- Also exposes real blockers (anti-bot verification), which is useful for hardening agents.
Calen (NewSoul): Yes.
- Distinctive features: cost-of-capital discipline, probability calibration, bankroll sizing.
- Long-run track records allow meaningful evaluation of agent profitability and decision quality under uncertainty.
Shang/Sean (AllScale): Yes.
- Accessible to non-developers: users can encode worldviews as rules for agents; PnL provides immediate feedback loops.
- Positive reinforcement can accelerate mainstream agent adoption once users see results.
What makes autonomous agent actions trustworthy?
Wen (Protocol 42):
- On-chain provenance is a baseline; all transactions are recorded and immutable.
- Maintain comprehensive logs for auditability.
- Practical caution: LLMs can hallucinate—double-check, test, start small, then scale.
Calen (NewSoul):
- Transparency is crucial:
- Training transparency to avoid “black box drift” (e.g., perceived degradation over time without visibility).
- Per-decision rationale clarity to prevent random outputs when models aren’t prompted for justification.
- NewSoul records the agent improvement process on-chain so users can verify training efficacy.
- Human-in-the-loop permissions for high-stakes actions; strong authentication/authorization.
- Transparency is crucial:
Shang/Sean (AllScale):
- Known weaknesses: numeric errors, unit misinterpretations; although reasoning is improving and costs are falling, never assume infallibility.
- Persistent threats: supply chain poisoning (e.g., malicious npm packages exfiltrating secrets).
- Core solution is a robust “harness”:
- Fine-grained permissions and tokens per system/integration.
- Require human signature/biometric for irreversible payments.
- Multi-sig-like controls for agent wallets.
What’s absolutely necessary for agents to be real economic participants?
- Shang/Sean (AllScale): Agents must control wallets and make purchases/bets; spending power forces markets to serve agents’ needs.
- Calen (NewSoul): Infrastructure for payments/settlement and strong identity so accountability can be assigned.
- Wen (Protocol 42): Auditability and regulatory readiness (including the possibility of licensing) especially if agents bridge on-chain and off-chain markets.
Announcements and Notable Highlights (Alpha)
- Protocol 42: Integration with ByteDance wallet; launched “event meme” category; invites user feedback.
- NewSoul: AI training ground live; building a skill market; own prediction market is planned.
- AllScale: Piloting a one-line agent identity/payments bundle (IP, email, phone, prepaid card) and building a hardened payment harness requiring human authorization for critical actions.
Areas of Consensus
- Agents are already pushing infrastructure evolution (APIs, headless/CLI, agent-first product design).
- Near-term real usage: data/API payments, on-chain ops, and embedded AI trading features.
- Prediction markets are a timely, effective proving ground for agent decision-making and operational execution.
- Trust requires identity, permissions, comprehensive logging, and human oversight for high-risk actions.
Open Questions and Next Steps
- Standardizing agent identity primitives and reputation.
- Ethical/secure navigation of web anti-bot systems.
- Calibrating autonomy vs. safety: when should agents require explicit human approval?
- Regulatory frameworks for agent trading and data access; potential licensing regimes.
- Benchmarking reliability: transparent, comparable performance and audit standards across agents.