ChinaTalk Podcast Summary

Episode: Software Abundance for Government With Cognition's Russell Kaplan
Date: March 9, 2026
Host: Jordan Schneider
Guest: Russell Kaplan (Co-founder of Cognition; former Scale AI and Tesla)

Episode Overview

In this episode, Jordan Schneider chats with Russell Kaplan about the challenges and opportunities in modernizing government software, the transformative potential of AI (especially Cognition’s “Devin,” an AI software engineer), and the impact of software abundance on public and private sectors. The discussion dives into technical, organizational, and policy aspects, with a focus on legacy systems, AI’s role in software engineering, procurement reform, and the road ahead for both government and enterprises.

Key Discussion Points & Insights

1. The State of Government Software

• Legacy Problems: Government spends $100B annually on IT, but many critical systems run on ancient code (COBOL, etc.) with slow modernization.
  “Tens of millions of lines of COBOL are powering our treasury, our Social Security Administration, and it's not getting better.” – Russell Kaplan [02:46]
• Hiring Bottleneck: Few people can maintain legacy code; the number dwindles every year.
• Procurement Hurdles: Regulations designed to reduce corruption have made it extremely hard to buy and upgrade software efficiently. FedRAMP compliance is slow and cumbersome.

2. Why Modern Programming Languages (and AI) Matter

• History Lesson: Language evolution improves abstraction, accessibility, and efficiency (from punch cards to assembly to COBOL to Python, and now to AI).
  “AI is actually the next logical rung on the ladder... It’s telling your computer what you want it to do, but in English in a way that’s really natural for everyone.” – Russell Kaplan [05:49]
• Shift in Constraints: Earlier efficiency demands have faded; hardware improvements mean software now focuses more on ease and productivity.
• AI’s Promise: Autonomous agents (like Devin) can optimize code, manage migrations, and handle large, tedious projects quickly.

3. The Vision of Software Abundance

• Switching Costs Breakdown: AI will neutralize long-standing vendor lock-ins; it will become much easier, cheaper, and faster to change systems.
  “When AI is going to do the switching and it’s going to work on it 24 hours a day and it’s not going to get bored […] that strategy [of vendor lock-in] doesn’t work anymore.” – Russell Kaplan [09:19]
• Software Becomes “Like Water”: Expect software to flow more freely, with rapid adaptation to needs.

4. What Cognition is Building

• Devin, the Autonomous Software Engineer: AI agent capable of grasping complex codebases, executing large refactors, and enabling faster system upgrades.
  “We provide this AI software engineer, Devin, that people can deploy against their code to really quickly transform it, improve it, modernize it, upgrade it.” – Russell Kaplan [10:40]
• Agentic IDE (Windsurf): Acquired in 2025; enables “digital coworkers” rather than just tools.
• Focus on Partnerships: Particularly in government and large enterprises, Cognition often embeds engineering teams directly, not just delivering tools.

5. AI Models and the Market

• Model Convergence: Leading AI models are becoming less differentiated over time; success relies more on integrating and orchestrating them than choosing the “best” one.
  “The differentiation in models is decreasing, not increasing over time [...] the structural advantage [goes] to the agent labs.” – Russell Kaplan [16:00]
• Cognition’s Eval System: Internally developed to rigorously benchmark model performance (“Junior Dev” and “Senior Dev” evals).
• Best Practices: Use multiple models for improved results; single-model constraints underperform mixed approaches.

6. Case Studies and Real-World Impacts

• Brazil’s CNPJ Migration: A major government ID format change solved in 3 weeks instead of 2 years at a top Brazilian bank using Devin—illustrating transformative scale and speed.
  “They were able to use Devin to get the bulk of that project done in three weeks instead of two years.” – Russell Kaplan [24:13]
• CVE Mitigation: Automated security vulnerability remediation at ~70% success rate pre-human intervention at major financial firms.
  “We try to auto remediate and we're right now at a roughly 70% fully automatic remediation success rate right now.” – Russell Kaplan [28:30]
• Fraud and Anomaly Detection: Government agencies leveraging AI (and Cognition’s tools) to analyze large datasets, identify fraud, and augment oversight.

7. Technical & Organizational Limits

• Context Window Limitations: Even the largest language models can only “see” a million tokens at a time; real-world legacy systems are orders of magnitude larger.
• AI for Codebase Mapping: Combining machine learning with graph algorithms to understand and modernize massive, intertwined systems.
• Bottleneck Shifts: Human understanding and verification (review) remain essential—at least until 2028, per Kaplan’s forecast.

Memorable Quotes & Moments (with Timestamps)

• “Tens of millions of lines of COBOL are powering our treasury, our Social Security Administration, and it's not getting better.”
  — Russell Kaplan [02:46]

• “AI is actually the next logical rung on the ladder... It's telling your computer what you want it to do, but in English in a way that's really natural for everyone.”
  — Russell Kaplan [05:49]

• “When AI is going to do the switching and it's going to work on it 24 hours a day and it's not going to get bored […] that strategy [of vendor lock-in] doesn't work anymore.”
  — Russell Kaplan [09:19]

• “We provide this AI software engineer, Devin, that people can deploy against their code to really quickly transform it, improve it, modernize it, upgrade it.”
  — Russell Kaplan [10:40]

• “The differentiation in models is decreasing, not increasing over time; the structural advantage [goes] to the agent labs.”
  — Russell Kaplan [16:00]

• “You can just run the code, you can compile the code, you can test the code, and if it works or doesn't work, that signal you can use for reinforcement learning to make these things better.”
  — Russell Kaplan [26:01]

• “You have to have this enormous history that I think we have to respect when we're trying to make changes to real world systems.”
  — Russell Kaplan [33:51]

• “Every company, every organization, every government agency is going to be in control of its own destiny in a much bigger way.”
  — Russell Kaplan [39:07]

• “The writing the code is not really the bottleneck anymore, it's everything around that. So I think humans still have to understand the code we're putting into production. And the emerging bottleneck, it's actually in review.”
  — Russell Kaplan [46:13]

• “The level of abstraction that people are going to be operating on is going to grow really high, really, really sort of unexpectedly fast... increasingly we’re going to be able to optimize against that objective directly.”
  — Russell Kaplan [48:39]

• Personal Moment:
  “My grandmother was, she was one of like the first female programmers in the country back when it was like in a very arcane, you know, very arcane activity of messing with punch cards... part of my hope for cognition, for government is we can go full circle and we can help bring the government back to actually where it once was, which was the true leader in technology.”
  — Russell Kaplan [54:47]

Important Segment Timestamps

• State of Government Software & Procurement: [00:20–03:23]
• Programming Evolution & Role of AI: [04:42–07:17]
• Software Abundance Vision: [08:05–09:50]
• About Cognition and Devin: [09:56–11:44]
• AI Model Market & Evaluation: [14:50–20:56]
• Enterprise Case Studies (e.g., Brazil): [23:46–25:16]
• CVE/Security Automation: [27:01–29:07]
• Understanding & Modernizing Legacy Systems: [31:10–34:37]
• Government Fraud & Modernization: [35:09–37:37]
• Comparison: Government vs. Enterprise vs. Tech-native Firms: [37:44–40:01]
• Changing Procurement & Organizational Leverage: [40:24–43:54]
• The Future of Software Engineering (Review, Abstraction): [46:07–49:11]
• Personal History—Programming Family Lore: [54:29–56:24]

Closing Themes & Calls to Action

• Policy & Imagination: Policymakers and public sector workers must continuously update their mental models and experiment with new software tools to keep pace.
• Empowerment: Smaller, more cross-functional teams can now deliver much more, amplifying the impact and creativity at every organizational level.
• Hiring at Cognition: They're seeking engineers—especially those with experience in the public sector and forward-deployed, customer-facing roles—as well as those with the required clearances for classified work.

Final Thought

Kaplan hopes that Cognition can help bring U.S. government technology back to its innovative roots—echoing the era when his grandmother wrote some of the country’s earliest code for public sector projects:
“Part of my hope for cognition, for government is we can go full circle and we can help bring the government back to actually where it once was, which was the true leader in technology.” [54:47]

For more insights like these, check out the ChinaTalk newsletter.