Hi everyone,

Thank you again for joining us for the February session of the Dynamic Abstractions Reading Group and for engaging in the discussion with Bryan Min on What is Malleable Software Now?

📼 Recording:
For those who couldn't attend, you can watch the session here:

We're excited to share the details for our next reading group, happening next Friday, May 1!🎉

🧠 Topic: Agentic Architectures for Embodied Mathematics
🎙 Speaker: Nazmus Saquib
🗓 Date: Friday, May 1, 2026
🕛 Time: 12 pm EST / 9 am PST
📍 Meeting Link: http://meet.google.com/tvw-pfxs-crv

About the Session
Scientific and Mathematical Computing practitioners rely on abstract symbolic and numeric computation for their typical workflows. On the other hand, science and math educators aim to explain abstract mathematical concepts using embodied, metaphorical representations when possible. Visual grammars and interactions developed in many educational and prototypical interactions often do not scale to a wide variety of abstractions typical in mathematics. However, with the advent of LLM-powered agents in the recent years, we believe embodied and agentic interactions for doing mathematics will take on new directions. In this paper, we present a roadmap for agent-assisted embodied interactions for scientific and mathematical computing, inspired by the architectures of coding assistants and previous research in the HCI community. Three separate systems define our architecture: an LLM-assisted visual grammar agent to extract mathematical relationships and elements from text, a set of coding agents crafted to notate drawings and handle embodied interactions in coding IDEs, and an enhanced tool calling pipeline that utilizes dynamic code generation to work in unison with the two systems above. We envision enabling a wide range of applications both in education and professional scientific computing via this architecture. We invite the scientific and HCI communities to comment on this direction.

Speaker Bio

Saquib is a veteran technical founder building infrastructure for human-agent interactions. Currently, he is building Oak Network, an a16z-backed spinout from Kickstarter. Under his leadership, Oak has scaled to support enterprise clients across the USA, LatAm, Africa, and Asia, providing the critical infrastructure for human-agent collaborations in commerce and micropayments. Saquib founded multiple companies in the past, building in multiple areas of deep tech: from architecting Web3 protocols and IoT hardware (Oneshot Inc., Mavu Labs, Wildflower Labs) to building interactive AI and large-scale scientific visualization tools (at MIT Media Lab, Adobe Research, ShopUp, Universal Machine).

Saquib received his PhD from the MIT Media Lab specializing in HCI and Embodied Cognition, with past degrees in Physics and Computer Science. His current academic interests focus on productivity and augmentation. Specifically, designing human-agent interactions to accelerate scientific, engineering, and educational workflows.

Beyond the lab and the terminal, Saquib channels his creative energy into Lowfat Dreams (a dream-grunge band he founded) and creating graphic novels.

Readings

The speaker will share readings later this week — stay tuned!

We hope to see many of you next Friday! Please feel free to share this with others who might be interested.

Best regards,
Sangho on behalf of the Dynamic Abstractions Team