Scaffolded LLMs as natural language computers

Recently, LLM-based agents have been all the rage – with projects like AutoGPT showing how easy it is to wrap an LLM in a simple agentic loop and prompt it to achieve real-world tasks. More generally, we can think about the class of ‘scaffolded’ 1 LLM systems – which wrap...

Recently, LLM-based agents have been all the rage – with projects like AutoGPT showing how easy it is to wrap an LLM in a simple agentic loop and prompt it to achieve real-world tasks. More generally, we can think about the class of ‘scaffolded’ 1 LLM systems – which wrap a programmatic scaffold around an LLM core and chain together a number of individual LLM calls to achieve some larger and more complex task than can be accomplished in a single prompt. The idea of scaffolded LLMs is not new, however with GPT4, we have potentially reached a threshold of reliability and instruction following ca

related reading

• LLM Powered Autonomous Agents | Lil'Loglilianweng.github.io
• 2025 LLM Year in Review – karpathykarpathy.bearblog.dev
• What We’ve Learned From A Year of Building with LLMs – Applied LLMsapplied-llms.org
• Guardian Angels: LLM Personalization for Productivity and Security · Gwern.netgwern.net
• As Rocks May Think | Eric Jangevjang.com
• Malleable software in the age of LLMsgeoffreylitt.com
• Patterns for Building LLM-based Systems & Productseugeneyan.com
• LLM Daydreaming · Gwern.netgwern.net
• GenAI Handbookgenai-handbook.github.io
• Language Models can Solve Computer Tasksarxiv.org
• LLMs as Parts of Systems - Marc's Blogbrooker.co.za
• Andrej Karpathy — AGI is still a decade awaydwarkesh.com