Michael Fester

Michael Fester co-founded Markprompt with Marie Schneegans, bringing language-model question answering to documentation and support teams. Released in late March 2023, it combined document retrieval with embeddable chat components. Fester advocated giving content writers control over AI behavior while preserving useful conventional interfaces, rather than making every interaction conversational.[2]

In his April 2023 interview with Tom Johnson, Fester explained that customers of Motif, the collaborative documentation platform he had co-founded, were using Markprompt alongside regular search. He cited AngelList and Cal.com and said that this reception led the team to release Markprompt as a standalone, open-source platform so others could build on it.[2]

Markprompt’s architecture treated sections, rather than pages, as the units of retrieval. Each section received an embedding representing its meaning; a user’s question was also embedded. Similarity search selected up to ten relevant sections, excluding those too distant in meaning, and inserted them into a prompt for the language model. This supplied context for answering questions rather than retraining the model’s weights. Fester questioned whether putting an entire documentation repository into GPT-4’s larger context would improve results enough to justify its cost.[2]

The team used Supabase vector storage and Vercel edge functions, with APIs, a dashboard, and React and Web Component integrations. Fester also described practical complications of generating answers from documents: relative links needed conversion to absolute links because an answer could appear anywhere on a site, and one generated sentence could combine material from different source files. The team addressed link handling through prompt engineering. Checksums identified changed content so that only those portions needed reprocessing.[2]

Fester described the default answer controls as instructions to use the supplied context, say “I don’t know” when uncertain, and use a temperature of 0.1. Users could change the prompt template to permit answers beyond that scope. These were intended safeguards against fabrication, not a guarantee of correctness. He linked the open-source decision to transparency around analytics and deployment on organizations’ own premises; in April 2023, he said the company did not examine analytics events apart from token usage required for billing.[2]

Despite building a documentation chat tool, Fester cautioned against rushing to “everything becomes a chat”. Writing a natural-language question takes effort, he argued, while browsing a well-designed page can quickly yield a code snippet. His example was Tailwind CSS’s color grid: it offered dense information, inspiration, and room for exploration, with no obvious benefit from replacing it with chat.[2]

He instead imagined documentation tailored to a visitor’s project and experience level. His hypothetical example was a Stripe documentation site assembled for someone building a Swift photo-generation app with a Ruby backend and three pricing tiers. This was a proposal, not a shipped feature. He also saw prompt templates as a way for content writers to shape responses without coding—for example, adjusting tone to a user’s apparent distress or humor. He called this “programming behavior.”[2]

Fester participated in the 2023 AI | UX San Francisco meetup. The Latent Space recap, published April 26, listed his and Marie’s MarkPrompt demonstration as an AI-native CMS that ingested Markdown and exposed it through a chat component.[1]

In a joint launch account retained on Y Combinator’s GloGlo page, Fester and Schneegans described building 14.ai as an AI-native customer-service agency. Their approach combined automated actions, such as generating shipping labels and triggering refunds, with human engineers responsible for support tickets and feeding interventions back into the agents. As of September 20, 2026, YC listed both as active GloGlo founders. GloGlo describes its AI glucose companion as turning continuous glucose data into personalized insights and practical guidance.[3]

Before the documentation and customer-service work, Fester co-founded Snips, a privacy-focused voice AI platform acquired by Sonos. He subsequently co-founded Motif, which supported collaborative documentation in Markdown, MDX, and Markdoc. This earlier documentation work supplied the customer setting from which Markprompt emerged; Fester also cited the team’s experience with human-in-the-loop tools at Motif when describing its ambitions for AI-assisted content creation.[2][3]