Natasha Maniar is an AI researcher whose work spans human-centered multimodal AI, assistive technology, and the redesign of software development around coding agents. She helped create MemPal, a wearable memory assistant for older adults, before contributing to McKinsey research on AI-enabled engineering organizations.
From wearable assistants to software organizations
At the MIT Media Lab’s Fluid Interfaces group, Maniar developed MemPal, which combines wearable cameras, computer vision, language models, and voice interaction to help older adults locate misplaced belongings, recall recent activities, and receive safety reminders. She designed and evaluated the system alongside older adults, caregivers, and clinicians, including trials in participants’ homes. As first author of the peer-reviewed MemPal research, she documented its use of visual activity recognition, room localization, and retrieval-augmented generation.
In 2025, Maniar worked at McKinsey & Company as a business analyst, contributing to research on enterprise adoption of open-source AI and AI-enabled software organizations. Her focus shifted from building AI applications to understanding how engineering teams must change when agents can generate code faster than existing review processes, planning cycles, and organizational structures can absorb.
- Specification-driven development: Agents need explicit requirements and acceptance criteria, including security and observability expectations; vague instructions create downstream review and rework.
- Task-specific human-agent collaboration: Legacy modernization can support tightly specified automation with human oversight, while new product development benefits from iterative experimentation. Maniar and Martin Harrysson connected these approaches to smaller multidisciplinary teams in their AI Engineer Code 2025 appearance.
- Outcome-centered engineering measurement: Tool adoption matters less than delivery speed, code quality, security, resilience, developer experience, and business results. One practical resilience measure is the time required to resolve high-priority defects.
- Interoperable, context-aware AI: Maniar has highlighted the Model Context Protocol, simulations of human and agent behavior, model limitations, and wearable interfaces as promising areas for practical AI development in a McKinsey conversation about emerging technologies.
