Import AI 441: My agents are working. Are yours?
Drexler's institutional framing of AI governance - directing an ecology of AI services rather than controlling a single agent - offers a conceptually useful lens for APS policy design.
Key points
- Jack Clark's essay describes firsthand experience deploying AI research agents to automate large-scale literature analysis and task execution.
- Drexler's 'Framework for a Hypercapable World' argues good AI outcomes depend on building institutional structures, not controlling singular AI entities.
- Content is primarily analytical and reflective; limited direct APS applicability but carries useful framing for AI governance thinking.
Implications for Australian agencies
- Consider AI governance and strategy teams could consider whether Drexler's institutional framing - designing processes to direct AI rather than control individual models - usefully informs APS AI governance architecture.
- Monitor Security and data integrity teams may want to monitor adversarial data poisoning tools like Poison Fountain as a signal of emerging risks to AI training pipelines used by or procured by government.
Implications are AI-generated. Starting points, not advice — see methodology for how they're framed.
View original source
Copied.
Appeared in:
Weekly digest, 19 January 2026
"Import AI 441: My agents are working. Are yours?"
Source: Import AI – Substack (Jack Clark)
Published: 19 January 2026
URL: https://importai.substack.com/p/import-ai-441-my-agents-are-working
Import AI #441 covers several distinct threads. Clark's lead essay reflects on his personal use of AI agents for research automation, observing meaningful productivity multiplication and raising questions about labour, inequality, and the pace of AI capability growth. A separate item covers Poison Fountain, an anti-AI data poisoning tool designed to corrupt training datasets, signalling a growing adversarial dynamic in the AI data ecosystem. Eric Drexler's new paper argues that AI governance should focus on building human institutions capable of directing a diverse AI ecology, rather than attempting to control singular powerful agents - a reframe with genuine relevance for governance design. A final research item describes a human-AI collaborative mathematical proof using Google Gemini and an unpublished DeepMind tool, illustrating frontier AI-assisted scientific discovery.
Implications for Australian agencies:
- [Consider] AI governance and strategy teams could consider whether Drexler's institutional framing - designing processes to direct AI rather than control individual models - usefully informs APS AI governance architecture.
- [Monitor] Security and data integrity teams may want to monitor adversarial data poisoning tools like Poison Fountain as a signal of emerging risks to AI training pipelines used by or procured by government.
Retrieved from SIMS, 18 July 2026.