Jun 4, 2026·2 MIN READ
Using AI Without Losing Your Humanity
Everyone has the same AI tools now. The thing that sets work apart isn't who generates the most, it's who can still make someone feel something, and that's where the premium is moving.
Jeff Liu
·AI Engineer · Multi-Agent SystemsField notes from building, shipping, and hardening production AI agent systems.
How I BuildBuild Log
11 EntriesJun 4, 2026·2 MIN READ
AI Native, Human First
You can outsource thinking and execution to AI. What you can't outsource is understanding, and that's the difference between the teams keeping up with it and the ones getting moved by it.
Apr 16, 2026·3 MIN READ
The Accountability Layer I Built for My AI Agents
When AI is wrong, it is wrong with the same confidence it has when it is right. The accountability standard and observability I built around my agents so that confidence stops costing me time, and the template I now reuse on every new build.
Apr 15, 2026·1 MIN READ
Exploring Higgsfield: Creative Video Generation Examples
A hands-on look at Higgsfield for AI video generation. Real prompts, real outputs, and why speed of iteration across models is the thing that actually decides where it fits.
Apr 15, 2026·3 MIN READ
The Real Skill Behind Getting Good Work Out of AI
Prompting isn't the gap with AI, judgment is. The pattern recognition you only get from real reps is what lets you catch the subtle, confident errors models make, and it's the leverage the tool can't hand you.
Apr 11, 2026·3 MIN READ
The AGENT Rule Refactor: Cutting 45 Files Halved My Context Load
I had 45 rule files telling my AI how to behave, and the output was getting worse. Then I measured 170KB of context loaded per session. Here's what happened when I cut half of them, and the context-budgeting discipline I now reuse on every build.
Apr 1, 2026·3 MIN READ
Why I Let AI Fight With Itself Before I Ship Anything
I run two models against each other before I ship anything. One builds the plan, a different one interrogates it. Why the second read is the closest thing to a real quality gate, and a default step I now bake into every build.
Apr 1, 2026·2 MIN READ
How I Keep AI Output Sharp Across Long Sessions
AI output degrades as a session runs, not because the model forgot but because the signal-to-noise ratio collapses. The context-fatigue meter and token-budget setup I built to keep it sharp, and now reuse across every build.
Mar 31, 2026·1 MIN READ
Routing Across Models When a Provider Hits Capacity
A single-provider AI setup is a single point of failure. Why I treat model routing as infrastructure you build in from the start, so a rate limit slows you down instead of stopping you.
Mar 28, 2026·2 MIN READ
Why Polished AI Output Still Hides Structural Bugs
AI output can look brilliant and still be spaghetti underneath. The fix is documenting your anti-patterns, the traps and bad shortcuts, so the model stops repeating them. Portable guardrails I now load into every new build.
Mar 22, 2026·2 MIN READ
What a Decade in Adtech Taught Me About the AI Adoption Cycle
Banners, mobile, programmatic. Every cycle felt too fast from the inside, and every time you adjust. AI is the same, with one difference: the gap between what it makes possible and what people have time to understand is wider than ever.