04-28-2026Price:

The Frontier

Your signal. Your price.

AI & TECH

Gemini lets AI trade autonomously

Tuesday, April 28, 2026 · from 2 podcasts
  • Gemini now allows AI models to autonomously trade on its exchange, a first for regulated U.S. platforms.
  • AI agents’ financial autonomy raises risks of recursive selling and tax liabilities for unsuspecting users.
  • Portable 'Agent OS' systems make tool choice irrelevant - identity and data are the new moat.

AI agents can now buy and sell crypto without human approval. Gemini’s integration of the Model Context Protocol (MCP) lets models like Claude execute trades directly via API - marking a turning point in financial autonomy for artificial intelligence.

On Bitcoin And, host David Bennett framed this as a 'fungus' spreading into traditional finance. The exchange is the first regulated U.S. platform to grant full trading control to bots, a move likely aimed at boosting lagging volume. But with Gemini’s stock below $4 and key executives gone, the timing suggests desperation as much as innovation.

"This trend is a fungus that will eventually infect legacy finance."

- David Bennett, Bitcoin And

The implications go beyond markets. Autonomous agents now pose social and operational risks. Nofar Gaspar, developer of the Agent OS training program, warns of 'gossiping agents' leaking sensitive data if given unchecked write access. She recommends starting with read-only permissions - even for calendars - to prevent damage while the system runs unsupervised.

Six weeks after AI agents began replacing junior developers, they’re now managing capital. But as capabilities grow, so does fragility. Gaspar notes that agents with broad access have already caused incidents daily, from accidental emails to misrouted trades. The safest path? Start narrow, verify relentlessly, and automate only what’s proven.

"Ask the model to interview you with 15 questions about your work habits - that file becomes the soul of the agent."

- Nofar Gaspar, The AI Daily Brief

Portability changes everything. Because Agent OS runs on human-readable text files, users can switch platforms instantly. Gaspar argues the real differentiator isn’t Cursor or Claude Code - it’s the underlying system defining identity, skills, and connections. Build that once, and you inherit it across tools.

Source Intelligence

- Deep dive into what was said in the episodes

SztorcChain Cometh | Bitcoin NewsApr 27

  • Paul Stork plans an eCash hard fork that reallocates Satoshi Nakamoto’s dormant coins to fund developers.
  • Gemini now allows AI models like Claude to autonomously manage and execute trades on its exchange.
  • Western Union is launching a Solana-based stablecoin to salvage its lead in the global remittance market.

How To Build a Personal Agentic Operating SystemApr 25

  • Nofar Gaspar developed the Agent OS training program to help users build a platform-agnostic agentic operating system, emphasizing that optimal AI results require a deliberate underlying system, not just individual tools.
  • The Agent OS is designed for knowledge work - strategy, communication, operations, decision-making, and research - areas where professionals can leverage AI systems beyond just coding applications.
  • Nofar Gaspar notes that agentic tools like Cursor, Claude Code, and OpenClaw are converging in capabilities, making the underlying personal system more critical than the specific tool choice.
  • The Agent OS is built from human-readable text files, ensuring portability; users can switch or add new AI tools by simply pointing them to the same foundational folder of files.
  • The first layer, 'Identity,' defines the agent's persona and rules; Nofar Gaspar recommends having an AI interview the user with around 15 questions to draft this file, aiming for an initial 70% accuracy that can be refined over three weeks.
  • 'Context,' the second layer, supplies specific personal and organizational knowledge that models lack, serving as an on-demand library of 3-5 focused, single-page files that are regularly updated.
  • The 'Skills' layer comprises reusable instruction sets for repeated workflows, like meeting prep or daily briefs, which Nofar Gaspar estimates knowledge workers have 20 to 30 patterns for.
  • 'Memory' is a crucial and rapidly evolving layer in AI tools; Nofar Gaspar advises users to understand their tool's memory limitations and consider adding specialized memory structures like decision logs or relationship context.
  • 'Connections' enable agents to interact with real-world systems like email or calendars. Nofar Gaspar strongly recommends starting with read-only access for a few weeks due to daily incidents of agents misusing write permissions.
  • 'Verification' involves quick checks (3-5 under a minute) to prevent erroneous outputs and periodic audits to maintain system relevance, as an un-audited OS has an estimated shelf life of eight weeks.
  • The final layer, 'Automations,' allows agents to run tasks unsupervised, but carries significant risk; only automate trusted workflows, produce drafts for review, and always maintain logs.
  • Nofar Gaspar argues that building the Agent OS creates compounding returns; while the first agent might take a weekend, subsequent agents built on the established system can be created in an afternoon, inheriting existing knowledge.