AngelList Replaces Traditional Semantic Layer with AI Agents
adnan_hashmi · x · 2026-09-02
AngelList engineers detail how they maintain a data semantic layer using AI Agents. Instead of traditional MetricFlow YAML or dbt views, they use a corpus of plain markdown files stored in the dbt repo. The agent reads these 'knowledge and skills' to generate correct SQL for 1,400 mart models, while GitHub Actions keep the documentation current. This approach solves inconsistent business logic definitions and avoids the high cost of maintaining a digital semantic layer, enabling their AI analyst Dana to serve the whole company.
More from coding & agent
- Deploying DeepSeek-V4 on Blackwell: Fixing 3 critical SGLang bugs — shrug_hellifino · 2026-09-02
- LangChain fine-tunes Qwen for agent evals, cutting costs by 100x vs GPT-5.5 — LangChain · 2026-09-02
- New Book: Building AI Agents from Design Patterns to Production — JordiRib1 · 2026-09-02
- User praises Grok Bot for eliminating manual data entry — djcows · 2026-09-02
- Claude Code generates Raspberry Pi case in minutes via AgentCad — shekitup · 2026-09-02
- Why Prime Agent Chose RLM Trajectories Over Prompt Tuning — CShorten30 · 2026-09-02