17/03/2026 11:50 am
Topic starter
Notifications
Clear all
0
What are the actual must-have technical skills for building AI agents because I'm honestly so lost right now. I spent the last week reading about LangGraph and CrewAI but then some senior devs on Twitter are saying those are just 'wrappers' and I should focus on raw API calls and memory management. I'm trying to build a simple scheduling agent for my uncle's plumbing shop in Chicago on a tight 3 week timeline and I'm stressed as hell. Do I really need to master vector DBs like Pinecone or is that overkill for basic tasks? I just don't wanna sink time into the wrong tools when I'm already struggling with the logic loops...
4 Answers
12
17/03/2026 11:50 am
ngl, I had issues with those frameworks too. Unfortunately they're not as good as expected for quick projects. Stay budget friendly:
- Master raw API calls with OpenAI gpt-3.5-turbo 4k context to save money
- Use Chroma DB Open Source Vector Store because it is free
- Get good at JSON parsing Dont sink money into Pinecone, it is overkill for this.
10
17/03/2026 12:15 pm
saw this earlier but just getting back to ya. for a three week sprint, you should prioritize function calling and deterministic state management. skip the complex frameworks for now.
- use OpenAI GPT-4o-mini API for fast, low-cost reasoning
- implement Supabase Vector Database for managed postgres and pgvector basically focus on prompt engineering and strict schema validation. dont let the twitter hype distract you from the core logic.
3
29/03/2026 5:51 am
Yep, this is the way
1
17/03/2026 12:55 pm
Over the years, Ive found state management is what actually breaks agents. Use Redis Cloud Managed Database with Anthropic Claude 3 Haiku LLM API for a cheap, reliable build.
