Anthropic engineer: "90% of our engineers were already running self-improving loops Now everyone is building agentic graphs" "No more prompting" In just 10 minutes, she builds her complete Claude Code system live from an empty terminal Agents → Loops → Graphs → Self-Improving Systems Prompting is the old workflow... Graph engineering is the new one This 10-minute video is worth more than most $1,000 agent engineering courses Watch it today Then save the full guide below before everyone catches up ↓show more

self improving loops make simple prompts obsolete

fr the prompt becomes a seed instead of the whole system

graph engineering is prompting turned into infrastructure

exactly the prompt became a whole operating layer

This shift to agentic graphs changes the game from passive prompting to active construction. It’s a move from seeking answers to building systems that find them. This focus on execution redefines what it means to be a creator in this space.

fr the shift from asking to building is the whole story

no more prompting sounds dramatic but the workflow shift is real

fr prompts arent gone they just moved backstage

we really renamed workflows “agentic graphs” and called prompting dead lol

fr we renamed the workflow and gave it a graph lol

graph engineering turns prompting into a system that runs itself

exactly the workflow stops waiting and starts running

that's wild. how do you handle the tradeoff between loop efficiency and agent autonomy without stepping on each other?

bounded loops for routine work and human checkpoints when the stakes jump

Graphs make coordination explicit, but they also make recovery the real product. If a node isn't idempotent and resumable, the graph is just a nicer way to rerun side effects.

exactly recovery is where the graph really earns its keep

whatever ai you have writing your posts is utterly botching it 🤡

This video is from May 2025 - that's ancient in AI terms! Showcasing a single agent taking 90 minutes on a localised task is now thoroughly outdated by fast, parallel multi-agent swarms operating in cloud-native sandbox environments.

"No more prompting" is doing a lot of work there. Ran a self-improving loop over 50 marketing skills and two runs on the same export still gave different verdicts, until every threshold carried a number and a unit. Independent graders took it from 6 to 8 out of 10.

wow its cool

Self-improving loops are fun until you hit rate limits or model inconsistency mid-run. If you're building agentic graphs and want a stable backend to test against, come try We support GLM 5.2, DeepSeek V4 Flash-0731 and Kimi K2.6 through one unified API.

no more prompting”有点夸张，提示词只是藏进工作流了