SlipstreamJobs tracks this role from the company's public career site. Apply directly on the employer's site.
Sandbar is building Stream, a private voice AI assistant that extends human thinking through conversation. The company, founded by engineers from Meta, CTRL-labs, Apple, Peloton, and Equinox, has been featured in WSJ, Bloomberg, Wired, and Time's Best Inventions of the Year.
You will own the integration layer that connects Stream to the software and services people already use daily. Stream is an AI agent that listens, remembers, and helps users think through conversation—and its utility scales with how deeply it can reach into users' existing tools and workflows.
Most software wasn't built to work with AI agents. Your job is to understand how each application is architected, find the most reliable way to connect to it, and make that connection dependable for users who rely on it every day. You'll work end-to-end on this problem: designing tool definitions and result shaping, managing OAuth and token lifecycle, handling webhooks and rate limits, building MCP servers and clients in production, and instrumenting every integration so failures are caught before users notice them.
You'll ship on the same real-time Python services the rest of Stream runs on, work directly with the founders, carry a share of the core backend and its on-call rotation, and help shape how other engineers build on Stream's platform.
This is a generalist role that requires comfort across backend, auth, client code, and agent logic—taking open-ended problems to shipped results without a detailed spec.
REQUIREMENTS:
- 5+ years building software, with strong Python and demonstrated experience owning a live, real-time service
- Shipped third-party integrations in production: OAuth 2/OIDC, token refresh and revocation, webhooks, API versioning, and vendor-specific quirks
- Deep understanding of software architecture to work with applications you didn't write
- Experience integrating tools into an LLM agent that real users depend on, with accountability for action completion
- Hands-on MCP (Model Context Protocol) experience
- Generalist range across backend, auth, client code, and agent logic; ability to take an open-ended problem to a shipped result without a spec
- Product sense for software that acts on someone's behalf, including when the user isn't looking at a screen
- Precise communication and ability to make tradeoffs explicit
NICE TO HAVE:
- Connector or plugin work at an AI product
- Browser automation (Playwright, Chrome DevTools Protocol)
- Integration security: scoped credentials, secret storage, prompt-injection defenses, audit logging
- Voice or multimodal assistant experience
- Experience writing developer documentation that other engineers have built on
