Life of the 2023 batch:
Learned and built during the era of subsidized AI tokens.
Graduating in 2027 into a world where AI costs are real and companies are tightening AI budgets.
Feels like we’re caught between peak AI adoption and AI cost reality ☠️
@OpenAI@Outskill Key learnings from 7 days of building:
1. LLM reliability is scaffolding, not the model
2. A clear demo narrative beats a long feature list every time
@OpenAI@Outskill 3. Production constraints only show up on day 5
4. @OpenAICodexCli for backend logic + @cursor_ai Composer 2.5 and GPT 5.5 for UI = the right split for a solo build
@OpenAI@Outskill The 5 hardest backend components were built with @OpenAI Codex as a full agentic assistant:
→ 3-pass LLM extraction agent
→ NetworkX bus factor scorer
→ Hybrid retrieval system
→ Neo4j schema + seed script
→ ChromaDB pipeline.
Complete working modules with tests.