Mar 20, 2026 · 1h 4m · latent-space
Dreamer: the Agent OS for Everyone — David Singleton
0:00 / 0:00
▶ Watch on YouTube →
gold bands on the timeline = statements, start to end. Hover to read, click to jump. CC turns on captions
In this in-depth interview, Dreamer co-founder and CEO David Singleton introduces Dreamer, an Agent Operating System that enables both everyday consumers and technical developers to build, run, and monetize autonomous AI agents. Singleton demonstrates live agent workflows, details the platform's kernel-like security architecture, and discusses how agentic software is reshaping application development and modern engineering teams.
How this conversation actually went
Every chapter scored 0–10 on four independent dynamics. Hover any point for the reasoning behind the score. The hosts hold 15.4% of the talking time here. How this is scored →
The hosts as informed peer 6.2 Guest teaching 3.2 Guest disagreement 1.2 The hosts pushing back 2.0
speaking balance: gold is the hosts, purple is the guest (3 minute bins)
Sharpest disagreement ▶ 49:00 David rejects opening Sidekick's internal harness to third-party coding agents
David pushes back against Swyx's suggestion that users want to swap in Claude Code or Codex inside Sidekick, firmly asserting that Dreamer's target consumers need full abstraction rather than engine tinkering.
Hardest push from the hosts ▶ 29:28 Swyx challenges AI platforms shifting to proprietary Robux-style token mechanicsSwyx directly questions the economic trajectory of agentic startups moving from seat-based pricing to proprietary credit currencies, calling out the friction and potential chaos of obscured margins.
Biggest teaching moment ▶ 23:30 David explains why vibe-coded apps fail security without an OS kernelDrawing on his operating system background at Android and Stripe, David educates on why isolated vibe-coded apps inevitably leak data unless regulated by a central kernel-like permission ring.
The host holds their own ▶ 35:48 Swyx articulates the Agent Lab thesis and routing layer abstractionSwyx synthesizes industry architecture by framing Dreamer as an exemplar of the 'Agent Lab' model—organizations that eschew foundation model training to master dynamic routing, evals, and multi-modal composition.
the scores for every segment, with the reasoning behind each
| Chapter | Topic | The hosts as informed peer | Guest teaching | Guest disagreement | The hosts pushing back | Why |
|---|---|---|---|---|---|---|
| Dreamer Interface: Sidekick, Dashboard, and Multimodal Outputs | 5 | 3 | 1 | 2 | David walks through the Dreamer dashboard and podcast generation feature. Swyx demonstrates familiarity with the product by asking why they pivoted from direct interactive voice to pre-downloaded podcasts, showing proactive engagement. | |
| The Tool Ecosystem and Monetization for Builders | 4 | 4 | 1 | 1 | David explains Dreamer's tool ecosystem, direct data feeds for sports, and how tool builders get monetized on the platform. Swyx listens and observes as David lays out the developer platform vision. | |
| Live Demo: AI Engineer Conference App | 6 | 2 | 1 | 1 | David presents a live demo building a personalized conference app using Swyx's AI Engineer Conference public JSON and llms.txt. Swyx adds color commentary about the speakers and his philosophy behind publishing raw data feeds. | |
| Agent Studio: The Planning, Coding, and Self-Testing Loop | 6 | 3 | 1 | 2 | David explains the plan-build-test loop of the Sidekick agent studio and inspecting generated prompts and code. Swyx jumps in with industry context, remarking that prompt management and hosting are whole startups on their own. | |
| Multi-Agent Collaboration and Sidekick Tasks | 6 | 3 | 1 | 2 | Swyx inquires about enrichment APIs like ZoomInfo and Clearbit for background agent tasks. David explains how Sidekick tasks run inside VM harnesses using public web data and highlights multi-agent coordination with tools like Granola. | |
| Security and Architecture: Sidekick as the OS Kernel | 6 | 4 | 1 | 1 | David details how Sidekick acts as an OS kernel mediating security between agent rings to prevent rogue vibe-coded access. Swyx connects this to platform economics and the Lego master builder analogy. | |
| Agentic Commerce, Protocols, and Model Routing | 7 | 3 | 2 | 4 | Swyx presses David on agentic commerce, micropayments, stablecoins, and the tendency of AI platforms to create proprietary token economies (like Robux). David counters by drawing a parallel to early web protocols like Gopher and UUCP before HTTP won. | |
| Episodic Apps Demo: Ski Bum and Expense Splitting | 5 | 4 | 1 | 2 | David shows a bespoke Ski Bum expense-splitting app made for a weekend trip. Swyx brings up his pain point with Bank of America integrations, prompting David to highlight the community-built Plaid integration tool Attain Finance. | |
| Curated Primitives, Partnerships, and the Agent Lab | 8 | 3 | 2 | 3 | Swyx lays out his 'Agent Lab' framework, explaining how platform builders act as dynamic routing and curation layers rather than foundation model training labs. He also pushes on how strict curation might break down as native multimodality advances. | |
| Agent-as-a-Tool and Custom Routing Infrastructure | 8 | 3 | 1 | 3 | Swyx digs deeply into the technical stack under the hood, checking SQLite multi-tenancy, custom versioning vs Git, and TypeScript dominance in coding agents. David explains their architecture choices, including why they avoided standard Git for internal agent versioning. | |
| Auth, Data Isolation, and Coding Agent Philosophy | 7 | 4 | 2 | 3 | Swyx tests Dreamer's design assumptions regarding database-level auth and whether external engines like Claude Code or Codex can replace Sidekick internally. David defends their closed coding agent harness by emphasizing that consumer creators need an abstracted loop rather than pluggable engines. | |
| Designing Agentic Personalization and Memory | 6 | 4 | 1 | 2 | Swyx probes what memory architectures Dreamer tested and discarded, specifically asking about knowledge graphs and vector RAG. David details how early embedding-based retrieval added unnecessary complexity compared to their current persistent profile model. | |
| Startup Leadership and Operating with Lean AI Teams | 6 | 2 | 1 | 1 | Swyx introduces his thesis on 'tiny teams' running multi-million dollar revenues with lean staff. David agrees, detailing how Dreamer built its full platform with roughly 6 to 17 people using their own internal agents. | |
| Hiring Engineers and User Innovations | 6 | 3 | 1 | 1 | David describes their technical interview loop, evaluating candidates on round-robin multi-agent workflows and product sense. Swyx makes lighthearted jokes about tool SEO naming conventions. | |
| Human Taste, Creativity, and the Frontier of LLMs | 7 | 3 | 1 | 2 | David and Swyx discuss the remaining frontiers of LLMs, focusing on human taste and verifiable code generation versus unverifiable creative aesthetics. Swyx reflects on how benchmarks shift as human intelligence benchmarks get automated. |
Statements from this episode (24)
Dreamer targets non-technical consumers for building and using AI agents
“Dreamer is a new product which everyone can come and play with today. It's a place where everyone, literally everyone, can discover, build, and enjoy and use AI agents and agentic apps. And we really did design it for consumers, for folks who are not necessari…”
Singleton: Stripe deployed some of the world's first production AI agent systems
“I was working at Stripe, as you mentioned, and we had the opportunity to put some of the very first AI agent systems in the world into production.”
Singleton: AI agents work best when integrated into users' existing apps
“We've actually find that making things show up in the other apps that you already use in your life is incredibly powerful.”
Dreamer pays AI tool builders proportionally based on agent usage
“And we're actually sharing something for the first time on this podcast, which is 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.”
Singleton built an AI Engineer Conference schedule app in 25 minutes
“And I was able to build this yesterday morning. I did it between some meetings. I think I spent a total of 25 minutes of wall clock time on it. I did it over the course of a couple of hours.”
Self-testing feedback loops unlock the magic of state-of-the-art coding models
“Anytime you can get any modern state-of-the-art coding model into a loop where it can make changes and perceive its own output and then fix bugs, magic happens.”
Dreamer uses autonomous AI agents to research and screen waitlist signups
“So at this point, we pretty much run the company on Dreamer agents for all kinds of important things. Maybe a good example of that is our wait list. People are signing up. Every time someone signs up for our waitlist, a Dreamer agent will actually research tha…”
Dreamer isolates application sub-agents within dedicated virtual machines
“Any application on Dreamer can kick off a sub-agent to do a particular task. So this actually is a powerful agentic harness that runs inside of its own VM.”
Swix: Sam Altman's Top AI Wish Is a Self-Completing To-Do List
“Do you know this is Sam Altman's number one ask for an AI app? It's the self-completing to-do list.”
Singleton: Standalone AI apps cannot scale safely without an OS-like core
“Because if you try to pick off just one piece of this, you can't actually make it work for people at scale because you could build little vibe coded apps, but they're going to grab all your data willy nilly. They won't be able to work together. You actually ha…”
Singleton predicts one or two dominant agentic commerce protocols will emerge
“You know, on Dreamer, we hope that folks will build tools that can make use of all of these things, but I'm sure that at a certain .1 or two will emerge as the winners, and then we'll be able to build, like, really deep support in it.”
Dreamer uses continuous evaluations to dynamically route tasks across AI models
“Dreamer actually uses all of the state-of-the-art models. As a user, you don't have to think about, should I be using, you know, Opus four six, or should I be using the five four model from OpenAI? We are continually doing evals and so forth to make sure that …”
Singleton: Dreamer keeps tool interfaces stable while updating underlying models
“So the point behind these, though, is that we'll keep the interfaces stable, so they'll always work. But, you know, the best translation model, and, you know, there are people using this translation tool to translate Chinese podcasts into English. It's pretty …”
Swix defines Agent Labs as routing and evaluating models without training
“One observation I have, this kind of massive thesis I've been pursuing, which is what I've been calling an agent lab. Where you're sort of different than a model lab in the sense that you never train your own models, but you are the router evaluation layer, su…”
Dreamer architectures AI agents as modular, composable tools
“To some extent, Dreamer does make its own tools in that agents appear to the system as tools, so they can be used to accomplish things. So you can build an agent that is essentially a tool”
Singleton: Developers must optimize their CLIs for AI agent discovery
“I was chatting with folks at Stripe last week and saying, hey, you got to make the Stripe CLI actually tell agents what they can do on Stripe because that way they're going to use more stuff on Stripe. I think this is a real trend for the entire industry.”
Singleton: TypeScript is the optimal language for AI-driven development
“TypeScript is an amazing language for AI because there's tons of training data. In the models and it's strongly typed. And actually at the company, we built most of the stack in TypeScript, and we have this amazing property, which is we have type safety all th…”
Swix: Personalization and memory are the most important jobs of an Agent OS
“I think personalization and memory is probably like the single most important job of the OS.”
Singleton: Dreamer replaced Vector DB RAG for agent memory due to complexity
“Very early on, we were putting lots of facts into a vector database and doing embeddings and pulling them back out using, you know, reverse look of embeddings. Rag that actually worked, but turned out to be much more complexity than was actually required. So, …”
Dreamer built its initial AI platform with a core team of six
“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”
Singleton primarily evaluates engineering candidates on their ability to use coding agents
“One of the main things that I look for now when hiring engineers is how well do you work with coding agents?”
Singleton: Hands-on engineering managers have the ideal skill profile for AI orchestration
“It turns out being an engineering manager, as long as you stay very close to the code and are able to continue to craft it yourself, is actually a great skill profile for being able to make agents work for you and for your team in this in this age.”
Singleton: Orchestrating concurrent coding agents in a round-robin pipeline multiplies productivity
“Anytime I'm working on code these days, I always have more than one agent going at the same time, because while one agent is going and reviewing the output of the next one, and if you get them in a nice round robin, you can be very, very productive. You can al…”