10 tips to level up your ai-assisted coding - Aleksander Stensby - NDC Manchester 2025

Value of the Information & Strength of the Argument

The talk provides high practical value for developers seeking to enhance their AI-assisted coding practices. Stensby offers concrete, actionable tips that are directly applicable, such as using rule files, managing context windows, and leveraging subagents. His argumentation is coherent and grounded in personal experience, though it lacks formal evidence or citations. He effectively argues that AI should be treated as a collaborator, and he supports this with examples and analogies. The emphasis on iterative feedback and active context management is well-reasoned and aligns with best practices in the field. However, the talk is largely anecdotal, and the speaker does not provide empirical data or comparative studies to substantiate his claims, which weakens the overall argumentation.

Scientific Rigor, Source Quality, Title Accuracy

The talk demonstrates a reasonable level of scientific rigor for a practitioner-oriented presentation. Stensby references tools and concepts like Claude Code, Cursor, MCP, and ‘compounding engineering’ from Every, but he does not cite specific academic sources or empirical studies. The quality of sources is moderate, as he relies on his own experience and industry trends. The title accurately reflects the content, and the talk stays on topic. The speaker does not provide a formal bibliography, but he mentions the concept of ‘compounding engineering’ from Every, which is a credible source. Overall, the scientific rigor is acceptable for a conference talk, but it would benefit from more concrete references.

The title accurately reflects the content, as the speaker delivers 10 tips (plus a bonus) for improving AI-assisted coding workflows.

7/10

The talk provides practical, experience-based advice on AI-assisted coding, with a focus on tools like Claude Code and Cursor. The speaker is a practitioner with two years of hands-on experience, and the content aligns with current industry practices. However, the talk is largely anecdotal and lacks rigorous empirical evidence or citations to specific studies, which limits its scientific reliability.

Markers derived by PSI from the transcript: the creator did not define chapters.

• 02:00Introduction: Speaker introduces himself and the talk's purpose.
• 05:00Discussion on the evolution of AI coding assistants and the importance of mindset.
• 10:00Tip 1: Context is king - the importance of providing sufficient context.
• 15:00Tip 2: Use rule files (CLAUDE.md) to encode preferences and style.
• 20:00Tip 3: Actively manage the context window and reset conversations.
• 25:00Tip 4: Leverage subagents for parallel tasks and context isolation.
• 30:00Tip 5: Use documentation and LLM.txt files for targeted context.
• 35:00Tip 6: Give specific feedback and iterate with 'make it better'.
• 40:00Tip 7: Use MCP to connect AI to external tools like GitHub and Slack.
• 45:00Tip 8: Embrace 'compounding engineering' for long-term improvement.
• 50:00Tip 9: Use slashinit to generate tailored rule files.
• 55:00Tip 10: Treat AI as a junior developer and provide feedback.
• 1:00:00Conclusion and Q&A.

• NDC Conferences — Conference organizer and host of the talk.
• NDC Manchester — Specific conference where the talk was recorded.

• Anthropic's Claude Code documentation — Provides details on subagents and rule files, aligning with the speaker's tips.
• Model Context Protocol (MCP) official site — Explains the standard for connecting AI to external tools, as mentioned in the talk.