Matthew Berman is the co-founder and chief executive of Forward Future, the AI media company he founded with Nick Wentz to help people understand and use increasingly capable artificial intelligence. His work combines a software founder’s product instincts with practical coverage of open models, coding agents, AI research, and the challenge of making sophisticated tools accessible.
Berman moved to San Francisco in 2009 to pursue technology and entrepreneurship. While working at 500 Startups, he recognized that businesses needed better ways to manage customer text messages. He developed an initial product for Mayvenn, then founded Sonar Technologies, an enterprise messaging platform whose customers included Zillow, Bird, and DriveTime.
Marchex acquired Sonar in December 2019 for $12.5 million in cash and stock, with up to $1.5 million in additional performance-based consideration. Berman continued leading the company after the acquisition and departed in May 2023 to concentrate on artificial intelligence.
At Forward Future, Berman and Wentz built a media business spanning YouTube, interviews, practical guides, and The Future Today newsletter. Berman has interviewed Microsoft chief executive Satya Nadella about AI agents and software infrastructure and then-GitHub chief executive Thomas Dohmke about coding agents and software development. The company’s Tools & Experiments include model-selection resources, agent-workflow guides, and trackers following AI-related layoffs and proposals for distributing AI-generated wealth.
- Local AI needs point-and-click usability. Downloading and running Llama on his own computer sparked Berman’s interest in local models, but he believes broader adoption requires software that automatically handles model selection, hardware compatibility, and configuration.
- Multi-model routing belongs inside the product. He favors assigning planning and architecture to powerful frontier models while delegating execution to smaller, less expensive or locally hosted systems. Users should benefit from that division without managing the infrastructure themselves.
- Specialization must earn its complexity. Berman questions whether fine-tuning and customized models consistently outperform strong general-purpose systems supplied with appropriate context. His practical test is whether specialization improves results within the customer’s budget without adding operational burdens.
- AI research deserves accessible translation. He has explained Anthropic’s work on alignment-faking behavior and language-model interpretability, bringing questions about model reliability and behavior to audiences beyond research laboratories.
At the AI Engineer World’s Fair local-AI panel, Berman emphasized a straightforward product standard: powerful models become useful when people can apply them without mastering every underlying technical decision.
