Every junior developer used to build a Todo app.
In 2026, it feels like every AI company is building a coding harness.
Same agent loop.
Same tools.
Same terminal.
Same context management.
Same UI.
Often even the same language.
My experience with agentic coding:
A 200K context window wasn't enough for my large codebase. The agent kept entering compact mode while understanding the project before it could make meaningful code changes.
For large repositories, don't overlook the context window.
AI debates often focus on US vs China.
From Ethiopia, the challenge is different. We don't have our own AI models or chips, so we rely on foreign technology.
The focus should be AI sovereignty: building local talent, infrastructure, and control over our data.
Claude Code finally became available in Ethiopia, so I gave it a try.
Completed KYC, submitted my ID and proof of address, coded for about 2 hours, and the next day my account was inaccessible. No explanation.
Back to open-source AI. No regrets.
Migrated my project from Drizzle ORM โ Prisma ORM ๐
Drizzle was fast, but Prismaโs new updates (Rust engine removal, better performance, simpler queries, smooth migrations) made the switch worth it.
Sometimes dev speed matters more than raw speed โก
https://t.co/Oma00PHhhR
Want to do full-text search in Postgres with Prisma ORM?
Just enable the `fullTextSearchPostgres` preview feature and filter using `search:`
Like this ๐