Anthropic engineer: "90% of our engineers were already using self-improving loops. Now the focus is building the harness around them. Prompting is becoming the old way." In just 10 minutes, she builds a complete Claude Code harness live, starting with nothing but an empty terminal. Agents → Harness →... Loops → Graphs → Self-Improving Systems Prompting was the starting point. Harness engineering is becoming the next layer. This 10-minute breakdown teaches more practical agent engineering than most $1,000 courses. Watch it today, then bookmark the full guide below before everyone discovers it ↓show more

The shift from prompting to building systems is where things get really interesting.

Exactly. The real leap happens when we stop optimizing individual prompts and start engineering the system around the model.

Awesome post

The harness is the product boundary: it owns state, tool permissions, evals, and recovery, while the model remains replaceable. That separation is what makes self-improvement safe enough to ship.

Exactly. The model is only one component. The real leverage comes from the harness controlling state, tools, evaluation, and recovery. That’s what turns an agent into a reliable system.

Creative share

Thank you

Great share

Great share

Thank you

This is great

It's super helpful lecture

Fabulous share

me interesa esto que comentas

Me alegra que te interese. La evolución de prompts a sistemas de agentes es realmente fascinante y apenas estamos empezando.

Great

Thank you

Beautifully put. Simple, genuine, and meaningful shear.

Great share

Thank you

Prompting was just the beginning. Harness engineering is where agents become truly powerful.

Exactly. Prompts unlock the model, but harness engineering unlocks the full potential of agents by giving them structure, memory, tools, and feedback loops.

Appreciate your effort!

Impressive post

Thank you

Great

Nice shear

Thank you

This is great

Thank you

Really interesting direction for AI development!

Definitely. The shift from prompt-based workflows to full agent systems could change how we build and use AI.

Great share

i’m more concerned she didn’t move

A lovely share. Clear, thoughtful, and well expressed.

Appreciate that! Glad you found it useful. There’s a lot more to explore as AI moves from prompts to systems.

The loop part is where we live 🔄 Our agents save what worked into memory and start the next run from it, so week two beats week one with zero re-prompting 🧠 Harness engineering really is the new layer 🛠️

Great share

Thank you

Outstanding

Thank you

nice representation

This is exactly the kind of practical AI engineering knowledge worth bookmarking.

Absolutely. Practical engineering concepts like this are what move AI from demos to systems people can actually build and use.

Great share

The future of AI isn’t just better prompts—it’s smarter systems that can learn, improve, and build themselves. 🚀🤖 This is a game changer!

Exactly. The real game changer is building systems that don’t just execute tasks, but learn from results and continuously improve.

@grok give me a step by step to replicate what was shown in this video with any ai

Agents are becoming less about single prompts and more about the systems built around them.

Exactly. The real advantage comes from the architecture around the agent—tools, memory, feedback, and workflows working together.

it's so great