# Yunpeng "Spencer" Zhao

> Cofounder at CoreSpeed — the access and control layer for AI agents. Personal site: projects, writing on AI agents and infrastructure, and contact. Based in San Jose, CA.

Spencer builds CoreSpeed (https://corespeed.io): connect an AI agent to 50+ apps through one OAuth, with built-in tools, memory, plain-English policy, and agent pay. He also ships products end to end (BuildLog, Fixo) and open source (Zypher Agent, Claude Max Gateway). Every blog post is available as raw Markdown by appending `.md` to its URL.

## Pages

- [Home](https://spinsirr.github.io/): overview, selected work, and stack
- [Projects](https://spinsirr.github.io/projects/): CoreSpeed, products, open source, hardware
- [About](https://spinsirr.github.io/about/): bio, experience, education, and contact

## Writing

- [AI agent access and control: a field guide](https://spinsirr.github.io/blog/access-and-control-field-guide.md): One map of the agent stack: access through one OAuth, built-in tools, memory, plain-English control, and portability across hosts — and where CoreSpeed fits.
- [The setup tax: why connecting AI agents to apps is hard](https://spinsirr.github.io/blog/the-setup-tax.md): Connecting AI agents to apps means OAuth, token refresh, scopes, rate limits, pagination, and webhooks per service. Here's why that work stalls agent projects.
- [Building AI products fast: my stack and patterns](https://spinsirr.github.io/blog/building-ai-products-fast.md): The stack and patterns I use for building AI products fast — Vercel AI SDK, Deno + Hono, Supabase, Drizzle, Stripe, AG-UI — drawn from BuildLog and Fixo.
- [Rebooting this site](https://spinsirr.github.io/blog/rebooting.md): Why I rebuilt my corner of the internet — and what I am working on now.
- [What an AI agent harness actually does](https://spinsirr.github.io/blog/agent-harness.md): An agent harness is the loop around the model — tool access, error recovery, staying on task. Why that layer decides whether AI agents are useful.
- [Building production AI agents with the Vercel AI SDK](https://spinsirr.github.io/blog/vercel-ai-sdk-agents.md): Streaming, tool calling, structured output, and provider routing — the patterns I use to ship real AI products on the Vercel AI SDK, not just demos.
- [AG-UI: giving your AI agent a real interface](https://spinsirr.github.io/blog/ag-ui-agent-surfaces.md): AG-UI is a protocol for streaming agent state and actions into a product UI. How I use it in Fixo to turn a chat into real actions.
- [MCP vs. agent connectors: how agents reach apps](https://spinsirr.github.io/blog/mcp-vs-connectors.md): MCP is the standard interface for handing tools to an agent. Agent connectors handle auth, hosting, and breadth. Here's how they fit.
- [AI agent guardrails: policy, approvals, and pay](https://spinsirr.github.io/blog/ai-agent-guardrails.md): AI agent guardrails — plain-English policy, human approvals, and Agent Pay spending limits — are what make an agent safe to run past a demo.
- [What I'm building at CoreSpeed](https://spinsirr.github.io/blog/building-agent-infrastructure.md): Most AI agents can think but cannot act. CoreSpeed gives them access to the apps you already use — so they do real work, not just chat.
- [AI agent portability: run one agent across hosts](https://spinsirr.github.io/blog/agent-portability.md): AI agent portability — keep your apps, memory, voice, and policy with the agent, not the host. Switch Claude Code, Codex, and Cursor without re-setup.
- [The AI agent tools every platform should ship built in](https://spinsirr.github.io/blog/agent-built-in-tools.md): App connectors aren't enough. The AI agent tools every agent needs built in: web search, scraping, image generation, code sandboxes, and email.
- [A dual-format gateway for Claude Max](https://spinsirr.github.io/blog/claude-max-gateway.md): Point your existing OpenAI- or Anthropic-SDK code at a Claude Max subscription, no rewrites.

## Projects

- [CoreSpeed](https://corespeed.io): The access and control layer for AI agents. Plug your agent into 50+ apps through one OAuth so it gets real work done — connectors, built-in tools, and memory — with plain-English policy and agent pay to keep it in bounds. Portable across Claude Code, Codex, and Cursor.
- [BuildLog](https://buildlog.ink): Turns your team's commits, PRs, and releases into marketing — AI drafts social posts you review and publish to X, LinkedIn, and Bluesky.
- [Fixo](https://github.com/hmls-autos/hmls): Mobile-mechanic platform — an AI agent diagnoses car faults from a chat, writes the repair estimate, and drives scheduling and Stripe payments.
- [Amazon Order Wizard](https://github.com/spinsirr/order-wizard): Offline-first browser extension for tracking Amazon orders with optional cloud sync — a React 19 extension backed by a Rust (Axum) + MongoDB API.
- [Claude Max Gateway](https://github.com/spinsirr/claude-max-gateway): Dual-format API gateway for a Claude Max subscription — exposes both OpenAI- and Anthropic-style Messages APIs through the Claude Code CLI.
- [Zypher Agent](https://github.com/corespeed-io/zypher-agent): A minimal, open-source framework for building AI agents with full control over tools, providers, and execution flow.
- [FPGA Clock](https://github.com/spinsirr/FPGA_Clock): A multi-function digital clock built in Verilog, running on an FPGA.

## Links

- GitHub: https://github.com/spinsirr
- X: https://x.com/zhao_spenc
- LinkedIn: https://www.linkedin.com/in/spencerzhyp
- Email: mailto:yunpeng@corespeed.io