Latent Space

Latent Space is a technical newsletter, podcast and community focused on building with AI: foundation models, agents, developer tools, infrastructure and AI user experience. Started by swyx, with Alessio Fanelli as an early podcast cohost, it developed an editorial focus on AI engineers—software builders who apply and productize model capabilities. In January 2026, it brought AINews into its email subscription and launched an AI for Science podcast with additional hosts.[1][4][6][7]

That expansion followed a newsletter and interview podcast whose earlier history also included small meetups, hackathons and paper discussions.[1][2][4]

swyx started the newsletter in 2022; the podcast followed in 2023. Both grew out of an informal Discord community running since Covid. In his naming account, swyx recalls initially publishing as “L-Space Diaries” at lspace.swyx.io. The name referred both to latent space and to L-space in Terry Pratchett’s Discworld. He credits Brian McMahan, coauthor of Natural Language Processing with PyTorch, with the friendly domain offer that enabled the move to latent.space.[1]

An early event outgrew its intended format. On February 14, 2023, the organizers reported 200 registrations for a planned 15–20-person Airbnb hackathon. Alessio, through Decibel, and friends of Latent Space contributed support to expand Demo Day and open submissions to local and remote builders. Categories covered prompt engineering, OpenAI applications, developer tools, infrastructure and open source projects; sponsors included LangChain, Exafunction/Codeium, Baseten and HuggingFace, with $2,500 in OpenAI credits offered to one winner. At that point, two podcast episodes had been recorded and two more were lined up.[2]

The June 30, 2023 essay “The Rise of the AI Engineer” articulated the publication’s technical thesis: open source foundation models and model APIs were enabling software engineers to evaluate, orchestrate and turn AI capabilities into products. swyx described this as a “shift right” of applied AI, while explicitly saying, “I am calling attention to this trend rather than starting it.” The essay treats the API boundary as permeable: AI engineers can fine-tune or host models, and research engineers can build atop APIs. Product-specific data and evaluations remain part of the AI engineer’s work, distinct from foundation-model pretraining data and general benchmark evaluations.[4]

The same essay announced the AI Engineer Summit after months of small meetups. swyx teamed up with Ben Dunphy of Reactathon, whom he credited as cofounder of the AI Engineer conference series and network. He also thanked Alessio Fanelli of Decibel and Sarah Guo and Pranav Reddy of Conviction for reviewing drafts and providing feedback and support.[4]

The podcast also collaborated with other AI shows. A Practical AI crossover, published July 2, 2023 and taped at StudioPod in San Francisco, brought Daniel Whitenack into a discussion of favorite episodes and emerging AI trends.[3]

In its March 9, 2024 anniversary account, Latent Space recalled launching the podcast a year earlier with Logan from OpenAI. It reported more than 750,000 downloads, crossing one million unique Substack readers and reaching number 10 in the US Tech podcast charts. Its anniversary event, Final Frontiers, featured ten handpicked teams, including Lindy.ai and Julius.ai, judged by technical AI leaders from LlamaIndex, Replit, GitHub, AMD, Meta and Lemurian Labs. Pixee and RWKV won.[5]

Alongside interviews, swyx and Alessio developed monthly news and research recaps. Their January–February 2024 recap discussed five LLM research directions selected by swyx: long inference, synthetic data, alternative architectures, mixture of experts and online LLMs. Long inference included iterative feedback, chaining, search and planning rather than relying on a single prompt. The community’s LLM Paper Club complements this editorial synthesis with weekly unrecorded Zoom meetings covering an AI paper or reading.[5][1]

In “Scaling without Slop”, published January 23, 2026, swyx welcomed AINews’s 80,000 subscribers and described one email subscription combining daily AINews, weekly podcasts and broader essays, with opt-in and opt-out choices. He announced a podcast-network strategy with more hosts, education and entertainment formats, and most new investment directed toward YouTube and video-native production. Latent Space also had a new studio at Kernel in San Francisco. Saying that writing had fallen behind, he called for renewed investment in essays and recruitment of researchers and writers.[6]

Five days later, on January 28, Latent Space launched its AI for Science podcast, hosted by RJ Honicky of Miraomics and Brandon Anderson of Atomic. The show aims to explain AI applications in the hard sciences to engineers. swyx credited the new hosts’ scientific expertise as essential to extending coverage beyond his and Alessio’s developer-tools background.[7]

The January 23 announcement attributed an estimate of about 1.5 million all-time unique listeners to Flightcast. In its discussion of backing for AI Engineer and Latent Space, swyx described donations, partly from investors who had been told the projects would never sell or sell out, and said operating expenses came from Substack subscribers. He presented that support as giving the projects freedom to curate and experiment; he did not specify how the funding was allocated between them.[6]

On June 3, 2026, Latent Space published a live Microsoft Build crossover with No Priors, featuring Satya Nadella alongside swyx, Sarah Guo and Elad Gil.[8]