Dreamer Launches Sidekick: A Personal Agent Operating System
David Singleton and Hugo Barra shipped the project they've been building in stealth since leaving Stripe and Meta. Sidekick is an operating system for personal agents: a platform where users describe agents in natural language and deploy them across their digital tools.
The $56 million in funding from a16z and OpenAI's startup fund received coverage. The technical architecture matters more. Dreamer's "Sidekick" isn't an agent. It's a platform that lets you build agents through natural language, then deploy them across your digital life.
What Sidekick Actually Does
Most "AI agents" are prompt chains with memory tacked on. Sidekick takes a different approach: it treats agent construction as a programming environment where the programming language is English (or Spanish, or Japanese—it's multilingual from launch).
You describe what you want an agent to do. Sidekick compiles that description into an operational plan: what tools to use, what APIs to call, what memory to maintain, what decisions require human approval. Then it executes.
The shift: From "use this pre-built agent" to "describe the agent you need and materialize it."
The Tool Ecosystem Play
Dreamer announced $10,000 prizes for developers building tools that Sidekick agents can use. This is classic platform strategy: seeding an ecosystem before the user base exists, following the pattern of iPhone's App Store and Slack's integrations.
The bet: personal agents become useful when they can manipulate your actual software stack. Email, calendar, CRM, code repositories, design tools. Each integration makes Sidekick more valuable. The $10k prizes buy them a catalog of integrations without building them all in-house.
Competitive Positioning
| Platform | Approach | Dreamer's Differentiation |
|---|---|---|
| Claude Desktop | Local AI with tool access | Dreamer is cloud-native, multi-device sync |
| ChatGPT / Operator | Pre-built agents for specific tasks | User-constructed agents, not OpenAI-designed |
| Everything-as-Copilot | AI embedded in existing apps | Cross-app orchestration layer |
| Devin / SWE-agents | Specialized coding agents | General-purpose personal agents, not just code |
The Infrastructure Context
Dreamer launches into a landscape suddenly crowded with agent infrastructure. GitHub's trending tells the story: everything-claude-code, obra/superpowers, deepagents, deer-flow—all variations on the same theme. How do we build systems that can build agents?
What distinguishes Dreamer is consumer focus. While most agent infrastructure targets developers (Claude Code, SWE-agent harnesses), Dreamer targets end users who don't know Python. The abstraction level is higher. The trade-off is power for accessibility.
Related: Listen to the full Dreamer episode on Latent.Space with exclusive waitlist skip for subscribers.
What Happens Now
Anthropic has Claude Desktop. OpenAI has Operator. Now Dreamer has Sidekick. Three different architectures addressing the same problem: how to give AI systems persistent access to digital environments without compromising security or control.
Dreamer's answer: natural language as the interface layer, tool ecosystem as the moat, consumer simplicity as the wedge. Whether that architecture wins depends on whether users actually build agents—whether describing what you want proves easier than learning to code it.
Early signs suggest they will. The friction of traditional automation (Zapier, IFTTT, Make) has always been the translation layer between intent and implementation. If natural language removes that friction, the category expands beyond developers to the professionals who actually need the automation.
What would you build if you could describe it in a paragraph and have it running an hour later? That's the question Dreamer is betting we'll want to answer.