Thoughtworks Technology Podcast What we're talking about when we talk about context engineering
97 snips
Oct 2, 2025 Bharani Subramaniam, a ThoughtWorks technologist specializing in AI systems, and Alessio Ferri, an expert in developer tooling, delve into the fascinating world of context engineering. They unpack what context means in AI, emphasizing its crucial role in improving model quality and reducing hallucinations. The duo also shares innovative techniques like KV Cache and RAG, exploring how context management could evolve akin to operating systems. Their discussion highlights the broader implications for developers and users alike, advocating for a thoughtful approach to context in technology.
AI Snips
Chapters
Transcript
Episode notes
Context Means The Model's Input Universe
- Context is everything fed into a model, including memory, conversation history, and external data.
- Engineering shapes what the model sees and how we test and maintain that input.
Beyond Single Prompts
- Context engineering extends prompt engineering when systems become multi-hop and stateful over time.
- The focus shifts from single prompts to what the model actually sees across histories.
Compress Conversation History
- Compact and curate conversation history rather than sending raw, lengthy transcripts to the model.
- Summarize and compress past interactions to improve response quality and reduce token costs.