Latent Space: The AI Engineer Podcast
Dreamer: the Personal Agent OS — David Singleton
Guest: David Singleton (Co-founder & CEO, Dreamer, former CTO of Stripe)
Host: Swyx
Recorded: March 20, 2026
Episode Overview
This episode features David Singleton, CEO and co-founder of Dreamer, in a deep dive into the design, vision, architecture, and community behind Dreamer—a consumer-friendly platform for discovering, building, and deploying AI agents and agentic apps. David shares examples, demonstrates real use-cases, highlights Dreamer’s technical underpinnings, and discusses broader implications for the future of agentic software, small teams, and AI-powered productivity.
Key Discussion Points
1. What Is Dreamer? (00:22–01:54)
- Dreamer is a platform where anyone—regardless of technical background—can discover, build, and use AI agents and agentic apps.
- Design goal: Focused on consumers, making AI tools accessible to all.
- “I think often of my sister. She's very smart. She's not in the slightest bit technical... Dreamer is a place where she can come in, grab some intelligent apps...start using them right away and solve real problems in her life.” — David Singleton (00:41)
- Core feature: Every user has a customizable "Sidekick," a personal agent that helps across all Dreamer experiences.
- Launch: Recently released in beta—anyone can join the waitlist.
2. Inspiration from Android and the App Ecosystem (02:24–03:47)
- Dreamer’s founding team (David, Hugo Barrett, Nicholas Chetkov) worked on early Android; drew lessons from building developer ecosystems.
- Analogy: The transition from early mobile ecosystems to the open ecosystem of agentic apps and tools.
3. Dreamer’s UX: Demo and Onboarding (03:48–06:02)
- Onboarding: Users are greeted by Sidekick, which learns about them to recommend and help build agents/apps.
- Dashboard: Personalized feeds, widgets, background updates, and agents tailored to the user.
- Example: Calendar Hero—integrates with Apple Podcasts to produce a daily briefing podcast using data from meetings and contacts.
4. Voice vs. Podcast, Multimodality, and Integrations (05:52–06:51)
- Dreamer started with voice interactions but expanded to embedding output (like podcasts) in users’ existing apps.
- “It's incredibly powerful to make things show up in the other apps that you already use in your life.” — David Singleton (06:05)
- Emphasizes multimodal output and seamless integrations.
5. Community Gallery, Tools, and Tool Builder Economy (06:52–09:29)
- Gallery: Hundreds of community-built agents covering use cases from productivity to live sports.
- Tools: Foundation of the platform; Dreamer provides direct, high-quality integrations (e.g., real-time sports data, Google Search, Gmail).
- Open Platform: Developers can contribute tools. Notably, tool builders get paid based on usage—a first-time announcement on this podcast.
- “Tool Builders on Dreamer get paid. So if you publish a tool to the platform and a lot of agents use it, you'll actually get paid in proportion to their usage.” — David Singleton (06:52)
- Tools can be free or premium/pay-per-use, and Dreamer provides trials.
6. Agent Studio: Building Apps Using Natural Language (09:30–14:09)
- Agent Studio: Allows building apps or agents via conversational prompts with Sidekick.
- Example: Built a custom conference app for the AI Engineer conference in under 30 minutes using Sidekick and conference site data.
- “This is my dream conference app...I was able to build this yesterday morning...I spent a total of 25 minutes of wall clock time on it.” — David Singleton (12:37)
- Remix/Forking: Any agent/app can be forked, customized, and published by others.
7. How Agents Are Built and Deployed (14:10–18:34)
- The build process: Sidekick plans, builds, and tests the entire agent.
- Agents are fully hosted—no need to manage databases, LLM API keys, or infra.
- Sharing: Agents can be shared publicly or privately; common in teams/personal life.
- Example: Dreamer internally uses its own agents for operations (like waitlist management).
- Sidekick tasks: Powerful sub-agents with full access to tools/data.
8. Dreamer as the Agent OS: Security, Trust, and Architecture (20:21–25:42)
- Sidekick: Acts as the kernel of an agent OS—central traffic cop, enforcing permissions and user intent.
- “The model that we have...is that the sidekick is at the core of everything here...it is both your companion...but it's also the traffic cop in the system.” — David Singleton (21:14)
- All agent-to-agent/tool requests are mediated by Sidekick for privacy, safety, and intent alignment.
- Dreamer aspires to be a true platform, maximizing value for builders/users (Bill Gates’ definition).
- Announced a Builders in Residence program (apply via dreamer.com/latentspace): get fast-tracked, build unique agents, and get paid for creativity.
9. Tool Builder Program, Monetization & Commerce (25:43–29:43)
- Open call for tool builders; best new tool by mid-April earns $10k.
- Dreamer uses Stripe Connect to pay out builders—supports both fiat and (potentially) stablecoin protocols in the future.
- Dreamer aims to be OS/app store for agentic software, abstracting model selection and infra.
10. Advanced Example: Bespoke Apps for Short-Lived Needs (30:21–32:54)
- Ski Trip App: Built a custom agent for a ski weekend, integrating live ski data, group expenses, and more; easily shared with friends.
- Showcases the power and joy of making custom, ephemeral agents for unique life moments.
11. Partnerships and Curated Tools (33:22–34:37)
- Dreamer partners directly for first-party tool integrations (e.g., Attain Finance for banking, image gen, RSS, recipes).
- Maintains interface stability and state-of-the-art quality by auto-updating models/tools on the backend.
12. Technical Deep-Dive: SDK, CLI, and Engineering Flexibility (38:41–43:37)
- SDK and CLI: Power users can build with CLI, bring code out of Dreamer, work locally, and enjoy fully general VMs.
- “At the core of everything here is an SDK with a powerful command line interface. And we built that first.” — David Singleton (40:53)
- TypeScript-first: Agents built in TypeScript for strong typing and optimal LLM interaction.
- “TypeScript is an amazing language for AI because...it's strongly typed...and there's nothing better for working with coding agents.” — David Singleton (42:09)
- Dreamer has custom versioning, not Git, for efficiency and flexibility.
- Agents can be triggered by webhooks, filtered emails, and expose their internals for interoperability.
13. Identity, Auth, and Multi-user Experiences (46:50–47:49)
- Auth is platform-native; multi-user data privacy handled automatically by Sidekick.
14. Memory, Personalization, and Agentic OS Moat (49:40–51:51)
- Sidekick builds long-term memory of user preferences, enabling personalized suggestions (e.g., recommending parade events for an Irish user).
- Explored multiple memory architectures (vector DBs, embeddings, RAG, knowledge graphs). Settled on a custom system; actively evolving.
15. Company Building: Small Teams, AI-Native Workflows (52:02–54:40)
- David’s first time as CEO: Focused on cross-functional leadership, user engagement, and rapid learning.
- Dreamer’s team is small (17 people; core built by 6) but is highly efficient due to heavy agent usage and high talent density.
- “The core team that built everything I just showed you was honestly about six people. We're larger now. We're about 17 people at the company now.” — David Singleton (54:37)
- Agents used across Dreamer for ops, marketing, code review, and more.
16. Agentic Collaboration & Engineering Hiring in an AI Age (57:07–59:13)
- New hiring focus: Engineers must collaborate well with coding agents.
- Interview includes a live build session with LLM tools, observing how candidates work with agents.
17. On LLM Limitations: Taste, Individuality, and Creativity (59:23–61:45)
- LLM capabilities have rapidly improved, but outputs still lack "taste"—human-level creativity and delight.
- “What we've had to do is work on Dreamer to make it possible for the experiences that we build to not look like AI generic slop...LLMs do not do that by default.” — David Singleton (60:43)
- Dreamer invests in templates, harnesses, and prompts to enforce taste; this is a frontier for future research.
Notable Quotes & Moments
-
On Dreamer's mission:
“Dreamer is a new product where everyone, literally everyone, can discover, build and enjoy and use AI agents and agentic apps... It's really aimed at everyone.” — David Singleton (00:31) -
On first announcement of tool builder payments:
“Tool Builders on Dreamer get paid. So if you publish a tool to the platform and a lot of agents use it, you’ll actually get paid in proportion to their usage.” — David Singleton (06:52) -
On building high-quality, ephemeral software:
“This was all through Dreamer...it just helped us all settle up. We didn’t move the money on Dreamer. You could do that...if you’re a tool builder, come build a tool.” — David Singleton (31:30) -
On platform value & Legos analogy:
“You can only be a platform if you create more value for the folks participating...than the platform itself creates.” — David Singleton (24:33) -
On real-world company scale:
“The core team that built everything I just showed you was honestly about six people. We're larger now. We're about 17 people at the company now.” — David Singleton (54:37)
Important Timestamps
- [00:22] — What is Dreamer? Mission & design
- [03:48] — Live product demo & onboarding
- [06:52] — Gallery, tools, and tool builder payments Announced first time
- [09:30] — Agent Studio: Building with Sidekick
- [12:37] — Custom conference app demo
- [20:21] — Agent OS, Sidekick, security/permissions
- [25:43] — Builders in Residence & tool builder programs
- [29:43] — Dreamer as agent commerce platform
- [30:21] — Ski trip app: fast, personal, throwaway agents
- [33:24] — First-party/canonical tools
- [38:41] — SDK, CLI, advanced developer features
- [46:50] — Auth & multi-user data handling
- [49:40] — Memory and personalization
- [52:02] — Company building: small teams, agent-native execution
- [57:15] — How they hire engineers (agent-centric)
- [59:23] — What LLMs still can't do—taste and creativity
Memorable Moments
- Launch of monetized open tool-builder ecosystem (06:52)
- Rapid user personalization and memory as Dreamer’s moat (49:40–51:51)
- Live demo of building a conference app in "25 minutes" via Sidekick (12:37)
- “This is my dream conference app. It's exactly what I've always wanted” — David Singleton (12:37)
- Explanation of trust/safety model with Sidekick as the "traffic cop/kernel" (21:14)
- Insight that strong taste and creative delight are still outside LLM reach (60:43)
How to Join/Participate
- Apply for early access and builders’ programs via dreamer.com/latentspace
- Tool and Agent builders can monetize—best new tool by April wins $10,000.
Summary Takeaways
Dreamer is an ambitious take on democratizing AI agent creation by making building, remixing, and using agentic apps accessible to everyone. It stands out due to its focus on consumer UX, community monetization, agent OS-like architecture (with Sidekick as a personal OS kernel), and a thriving builder economy. With a strong technical foundation (TypeScript/CLI/SDK, open tool platform), Dreamer aims to accelerate bespoke, delightful, and trustworthy agentic software for all—while pushing the limits of what LLMs can do, and what they still can’t.
For full episode details and transcript, visit latent.space.
