Quick Overview
This video is an enterprise feedback showcase hosted by Romain Huet, Head of Developer Experience at OpenAI. Technical leaders from six partner companies share their initial impressions and screen recordings of GPT-6 Astra performing real-world tasks. The video serves to demonstrate the model's practical capabilities in code editing, data reasoning, research, and computer interaction.
Key Points
- 1.Industry partners from Box, Figma, cxo.dev, Ramp, BAM, and Cognition share early impressions of GPT-6 Astra across complex real-world workflows.
- 2.Box demonstrates how GPT-6 Astra handles nuanced business calculations across spreadsheets, avoiding double-counting errors in complex financial data.
- 3.Figma shows GPT-6 Astra generating complex vector animations by coordinating moving elements to create optical motion effects.
- 4.cxo.dev and Ramp highlight the model's autonomous computer and browser use capabilities to construct multi-step workflows and implement code changes directly.
- 5.BAM and Cognition demonstrate Astra's capacity to navigate ambiguous research trees and optimize lower-level GPU kernels for significant runtime speedups.
Summary
OpenAI Head of Developer Experience Romain Huet speaks with engineering and AI leaders from Box, Figma, cxo.dev, Ramp, Balyasny Asset Management, and Cognition to discuss their practical experiences with GPT-6 Astra. The participants note that Astra distinguishes itself from previous models through its frontier reasoning, deliberate task assessment before execution, and deep integration with computer use and software engineering workflows.
Yashodha Bhavnani, Head of AI at Box, details how Astra tackles difficult, nuanced tasks in entertainment finance. She presents a case study involving international film profitability and production rankings, where the model parses multiple spreadsheets, checks definitions across documents, and accurately factors in a twenty percent incentive for production costs without double-counting. Bhavnani explains that previous model generations struggled with these subtle calculation rules, whereas Astra successfully verified data across multiple interconnected tables.
John Milinovich, Senior Director of AI at Figma, showcases a concept animation titled Paper Cinema, where Astra constructed a vector illusion of a bird in motion. Rather than animating independent frames manually, the model manipulated a moving vertical grille across six static underlying poses to simulate smooth flight across the canvas, demonstrating advanced geometric understanding. Claire Vo, founder of cxo.dev, shares how Astra and Codex autonomously operate the node-based video and image editing tool Flora, placing nodes and generating thumbnail assets via direct computer use rather than reading manual documentation.
Veeral Patel, Head of Applied AI at Ramp, demonstrates Astra implementing internal-use UI badges across web navigation components. The model examined existing banners, added amber styling and lock icons across both desktop and mobile layouts, and returned visual screenshots confirming the implementation. In financial research, Hinal Jajal of Balyasny Asset Management shows Astra mapping complex research trees from massive text corpora, rigorously verifying assumptions and balancing task breadth with node-level depth.
Silas Alberti, SVP of Research at Cognition, outlines how Devin uses Astra to tackle complex research and engineering problems. The model evaluated past GPU optimization experiments and modified a norm-sampling loop in a register-based GPU kernel, achieving a 3.3 percent end-to-end speedup on GB200 hardware. Milinovich concludes that Astra's ability to execute parallel multi-step engineering tasks will substantially accelerate product shipping cycles.
Initial Impressions Across Industry Leaders
Representatives from Box, Figma, cxo.dev, Ramp, Balyasny Asset Management, and Cognition discuss their initial experiences using GPT-6 Astra. Leaders highlight the model's frontier capabilities, heightened confidence, thoroughness before executing tasks, and practical execution in coding and computer use.
Spreadsheet Reasoning and Complex Calculations at Box
Yashodha Bhavnani demonstrates Box AI leveraging Astra on nuanced financial datasets in media and entertainment. Astra verifies data across spreadsheets, correctly factoring in regional tax incentives without double-counting, which previously caused calculation errors in earlier models.
Visual Creativity and Dynamic Motion in Figma
John Milinovich presents a static vector animation created in Figma, where Astra animated a moving bird using a moving grille over six static poses. The model demonstrates an understanding of visual physics, layering, and canvas mechanics in design software.
Autonomous Computer Use in cxo.dev and Ramp Workflows
Claire Vo shows Codex and Astra interacting directly with Flora to construct node-based video and thumbnail workflows without human intervention. Veeral Patel shows Astra navigating internal tools at Ramp, modifying application code, and returning verified desktop and mobile screenshots.
Deep Research and Low-Level GPU Kernel Optimization
Hinal Jajal from BAM explains how Astra steers smaller models across large text corpora while maintaining depth and breadth in complex research trees. Silas Alberti from Cognition shows Devin using Astra to discover and apply GPU kernel optimizations, yielding a 3.3 percent speedup on GB200 workloads.
The Bottom Line
The video establishes that GPT-6 Astra provides significant gains in reasoning, computer use, and domain-specific execution across enterprise environments. Industry partners showcase validated workflows ranging from financial spreadsheet verification and design generation to automated frontend development and GPU kernel acceleration. The discussion highlights the model's reliability in handling ambiguous problems, leaving open the broader long-term implications for end-to-end software development pipelines.
FAQ
What is GPT-6 Astra and what are its primary business capabilities?
GPT-6 Astra is an advanced AI model developed by OpenAI that demonstrates enhanced reasoning, autonomous browser and computer use, spreadsheet analysis, and code optimization across complex enterprise workflows.
How does Box use GPT-6 Astra for complex data analysis?
Box uses GPT-6 Astra to analyze nuanced media and entertainment financial spreadsheets, performing multi-step verification and accurately applying tax incentive rules without double-counting.
What visual design demonstration did Figma present using GPT-6 Astra?
Figma demonstrated a static vector animation where Astra animated a bird in motion by orchestrating a moving line grille across six static underlying poses.
How did Ramp use GPT-6 Astra to modify user interface code?
Ramp prompted Astra inside an inspection tool to add internal-use warning badges, edit frontend components, and provide desktop and mobile screenshots of the finished UI.
What performance optimization did Cognition achieve with GPT-6 Astra in Devin?
Cognition used Astra within Devin to explore historical optimization experiments, ultimately improving a GPU norm-sampling kernel to deliver a 3.3 percent end-to-end speedup on GB200 hardware.
Worth watching for
Software engineers, AI developers, and technical product leaders interested in early enterprise use cases and agentic capabilities of OpenAI's GPT-6 Astra.
- gpt-6-astra
- openai
- computer-use
- ai-agents
- software-engineering
- enterprise-ai