HyperWrite CEO, AI writing and agent pioneer
Matt Shumer
Profile
Matt Shumer is the co-founder and CEO of OthersideAI, the company behind HyperWrite — one of the earliest AI writing assistants to reach consumers, and later one of the first companies to publicly demo an AI agent that could operate a web browser like a human. If you’re trying to build with AI today, Shumer is a useful figure to watch: he ships, he tweets his process, and both his wins and his failures play out in public.
HyperWrite started as a GPT-3-powered autocomplete Chrome extension in 2020, then pivoted hard into agents. In April 2023 he unveiled Personal Assistant, an agent that could book trips, order pizza, and fill out web forms from natural-language commands — predating ChatGPT’s Operator by nearly two years. He followed with Agent-1, a foundation model purpose-built to operate software, plus Agent Trainer and Agent Studio for customization. Alongside the commercial work he open-sourced useful dev tooling on GitHub — gpt-prompt-engineer (an ELO-ranked tournament that finds the best prompt for a task) and gpt-llm-trainer (generate a synthetic dataset and fine-tune a model from a single task description).
In February 2026 he published “Something Big Is Happening,” a long-form essay arguing that AI had crossed the threshold into self-improving systems and that most people were about to be blindsided. It racked up tens of millions of views on X and was covered by Fortune and much of the tech press — one of those cultural moments where the builder class broadcast its worldview to everyone else. He has since launched Shumer Capital to invest in AI-native companies.
He matters to developers learning AI because he is the archetype of the builder-influencer — someone who treats AI product development as performance on Twitter. His open-source tools are genuinely useful, his prompt-engineering threads (Concept Elevation, Ultra-Deep Thinking Mode for o-series models) are often early to a technique, and his agent work is a worthwhile reference for anyone thinking about AI that takes actions rather than just generating text. But (see Controversies) his track record also includes a very public benchmark scandal, so calibrate the hype accordingly.
Key Articles & Papers
Something Big Is Happening gpt-prompt-engineer gpt-llm-trainer Final Reflection 70B Update — Post-Mortem HyperWrite unveils breakthrough AI agent that can surf the web like a human Meet Reflection 70B: HyperWrite's open-source AI modelControversies
Reflection 70B (September 2024). Shumer announced Reflection 70B as the top open-source model in the world, claiming it beat GPT-4o and Claude 3.5 Sonnet using a technique he called “Reflection Tuning.” Within 48 hours, independent evaluators at Artificial Analysis couldn’t reproduce the scores; the Hugging Face weights performed roughly like vanilla Llama 3 — worse than Llama 3.1, which it claimed to be fine-tuned from. Critics accused Shumer and collaborator Sahil Chaudhary of Glaive AI of fraud, suggesting the hosted API was secretly routing requests to Claude. Shumer first blamed a broken upload, then published a more detailed post-mortem weeks later denying intentional deception but acknowledging serious mistakes. Useful reading if you want a case study in how benchmark hype, rushed launches, and build-in-public culture can collide. See VentureBeat’s coverage and Air Street Press — “Reflections on Reflection”.
Spotify Podcasts
