Latest posts
-
13 min read
The cost of the wrong abstraction in agent tools: using Notion as a case study
Notion Agent CLI: a Claude Code benchmark against the official Notion MCP, showing how task-level tools reduce agent turns, context drag, and cost.
Mean cost per session, 95% CI
- S1 Read+Summarize −20% −23%
- S2 Query+Report −79% −84%
- S3 Copy+Modify −61% −76%
- S4 Set Property −21% −23%
- S5 Replace Section −54% −51%
- S6 Merge Pages −88% −89%
- S7 Batch Update −33% −36%
- S8 Copy+Append −80% −81%
- S9 Create Table −53% −47%
- S10 Import DB −23% +19%
-
6 min read
Turning multimodal inputs into structured actions: a pipeline with strict validation
Kiroween 2025 hackathon, trying Kiro's Spec-Driven Development. A multimodal AI fitness app where text, voice, and image inputs route through typed validation before any action reaches the database.
Architecture
-
6 min read
Streamlining Programming Workflows: How AI coding assistants change the game
How tools like GitHub Copilot and ChatGPT are transforming the coding process and boosting developer productivity.
Time to finish the task
with Copilot 1 h 11 minwithout 2 h 41 min95 developers, 35 per group finished · times of those who finished