Benjamin Stein is co-founder and chief executive of SuperDuper, which builds personalized software to organize family schedules, school communications, and household administration. His career spans communications infrastructure, energy technology, workplace agents, and a central question for AI products: How can software accomplish useful work without demanding exhaustive instructions?
Stein began as a software engineer at Bloomberg before co-founding Mobile Commons, a text-messaging, voice, and campaign-management platform acquired by Upland Software in 2015. He subsequently became a general manager at Twilio and led product and engineering at Twilio.org. In 2017, he introduced Twilio Studio, a visual builder that made programmable messaging and voice workflows accessible to developers and their nontechnical collaborators.
He joined Arcadia as chief product officer in 2022, overseeing product strategy for its energy-data platform, Arc. He also co-founded QuitCarbon, extending his work into household electrification.
By AI Engineer World’s Fair 2024, Stein was affiliated with Perpetual and developing virtual colleagues modeled on familiar workplace roles. He co-founded Teammates that year, launching the product in early 2025 with AI collaborators equipped with identities, memory, personalities, and workplace integrations.
Teammates exposed a practical limitation: Stein estimated that nine out of ten virtual colleagues remained idle because users lacked the time or context to explain what they needed. He calls this the specification burden: the work of translating goals, preferences, and standards into instructions an AI system can execute. SuperDuper applies that lesson by extracting context directly from school emails, activity schedules, and other existing information. Its family dashboard also addresses household coordination that disproportionately falls on one person.
- Personality-driven agent design: Recognizable roles make an agent’s responsibilities immediately legible, but faces and voices also create expectations of conversation, reliability, and humanlike behavior. Stein warns that generated avatars can reproduce professional gender stereotypes and that describing agents as employees can heighten fears of job displacement.
- Specialized agents with bounded tools: Assigning an agent a narrow role, limited inputs, and a defined toolset reduces opportunities for misunderstood instructions and incorrect tool calls. A recruiter handles résumés and scheduling; an engineering assistant handles code and review.
- Composable agent primitives: Drawing on Twilio’s programmable infrastructure, Stein favors reusable capabilities such as document reading, database queries, analysis, and report generation. He has described combining those primitives into an agent that investigates application errors.
- Proactive, personalized software: Stein wants software to infer priorities from information people already possess and adapt to each household or organization. He also argues that apparent AI productivity should be measured against downstream testing, maintenance, review, and coordination costs—not the speed of generating an impressive first draft.
