Ùgò Chúkwú

Prompt "engineering" done without version control, evals, or systematic iteration isn’t engineering. It’s guessing with extra steps. Here’s the system that makes prompts actually maintainable at scale: x.com/ChukwuAugustus…

We’re excited to announce that the new Ohere Credit App V2 is now live on both Google Play Store and Apple App Store. Download it to experience upgraded features and benefits. Google Play Store: play.google.com/store/apps/det… Apple App Store: apps.apple.com/ng/app/ohere-c…

The discourse around model size has a skew problem. The people writing about frontier models are the people building them, benchmarking them, or covering them. Not the people running 50,000 classification requests a day watching their API bill. x.com/ChukwuAugustus…

Unpopular opinion: most production AI use cases don’t need frontier models. The 7B models that run on a laptop have gotten quietly excellent, and most teams never needed GPT-5.5 to begin with. x.com/ChukwuAugustus…

@Simeon_Cps That’s why they push updates everyday. Have you ever shipped and maintained a working app that hundreds or even thousands of users use everyday?

I was on track to become a Claude Code hater until I spent a day using Codex

Trust the path you choose and it will protect you, even in darkness.

If you figure out how a large multi-agent system can autonomously coordinate to maximum effect, you’ll win a Nobel Prize, a Turing Award and a trillion-dollar fortune all in one.

@sama good work 👍

The gap between prototype AI and production AI is mostly output engineering. A 1% JSON failure rate = 10 broken documents per 1,000. At scale, that’s not acceptable. Here’s how to make structured outputs actually reliable: x.com/ChukwuAugustus…

we love seeing our users win. we want to give you the best tools, lots of compute, and watch you do the magic.

Hot take: the most common AI system design mistake isn’t choosing the wrong model. It’s using the context window as a database. Here’s why that’s a problem, and what to do instead 👇 x.com/ChukwuAugustus…

We want you to have a lot of AI!

Today’s deep dive covers what I think is an underappreciated engineering problem in AI systems: context management. The instinct when building with LLMs is to put more in the context window. Models support 128K tokens now — why not use it? Three reasons not to: Cost scales quadratically. Quality degrades in long contexts ("lost in the middle" is real). And most of what’s in context is irrelevant to the current step. In today’s post: three strategies for managing context effectively in production AI systems — dynamic retrieval, hierarchical memory, and structured state — with implementation code for each. IOS uses all three, for different interaction patterns: retrieval for document queries, hierarchical memory for multi-turn sessions, structured state for workflow orchestration. The mental model: context window management is resource management. You have a finite token budget. Use it intentionally. #AI #LLM #RAG #AIEngineering #intellgenceos x.com/ChukwuAugustus…