Anthropic engineer: "90% of our engineers were already running self-improving loops Now everyone is moving toward agentic graphs" "Prompting is basically finished" In just 10 minutes, she builds her complete Claude Code setup live from an empty terminal Agents → Loops → Graphs → Self-Improving Systems Prompting was the... old workflow Graph engineering is the next one This 10-minute breakdown is worth more than most $1,000 agent engineering courses Watch it today Then save the full guide below before everyone catches on ↓show more

is it just me or they come up with new engineering terms every couple of months just to hook as in lol. But yh, indeed prompting is done, i have been using the loop engineering playbook with a custom MCP orchestration and it is the future indeed. Saved

lol fair the labels move fast but the workflow underneath is here to stay

yeah exactly. i think we are all craving the "vibe coding" or "vibe working" reality, is just that we know the underneath thing needs to be solid. basically we are trying to do everything we can now so we dont have to do much in the future lol

Stolen Authority post. Be ashamed of yourself.

Prompting is more load-bearing in loops, not less. A loop only converges when the stop condition is machine-checkable. Mine: max 6 iterations, exit 0 from pytest, diff under 400 lines. Miss any of those and the run gets thrown out.

fr in a loop the prompt stops being copy and becomes a control surface

hmm pretty interesting and useful

yeap yeap

graph architecture is where it is at

fr the model gets the hype but the graph does the heavy lifting

@0xwhrrari wow, that’s wild. self-improving loops sound like sci-fi stuff. thinking about how to integrate this into my projects now. AI is evolving crazy fast!

prompting becomes infrastructure when graphs own execution

agree with u

I don’t get it what’s different from just prompt engineering an initial prompt and give unlimited permissions to the agent? It will go on ok the background until it thinks it has accomplished the task.

is there any chance to get a full version of this guide?

i think yes

watched!

🫡🫡

graph engineering is the cutting edge workflow.

yeah orchestration is becoming the real engineering layer

Graphs over pure prompting is the practical upgrade path a lot of teams are landing on. Once the loop can improve itself, the daily workflow changes pretty fast.

exactly once the loop learns the workflow starts compounding instead of resetting

Save it

future you will thank you bro

loops are useful but graphs make the whole system visible

fr loops make it work graphs make it understandable

this setup is actually insane

for real insane setup

The thing that decides whether a graph actually beats a loop is where you can put a real check. A node whose output is verifiable by something other than another model, tests passing, a schema validating, a patch applying cleanly, is a node you can retry cheaply and fan out safely. Nodes with no verifier just relocate the guessing. Draw the graph around the checks you already have rather than around the steps you imagine.

i am really love cat wu, she is genius

hah cat wu <3

Lol it took one and and a half to make a table? Lol I made an entire reddit clone in 2 hours. Complete with all features lmao and launched

Umm, it's too old

> Agents → Loops → Graphs → Self-Improving Systems yeah okay i'll just wait until one of these lasts longer than 5 seconds before bothering

@bot get me the link to the original youtube video

90% already doing this, wow

Is there anything new? Come on.

doing it since few weeks, being honest: it's worth if you build agents, workflow and tools. it's not worth if you're a marketer, keep human in the loop it's worth way more.

Prompting is finished”这句还是有点下早了。

“The workflow underneath is here to stay” Bullshit.

Where the fuck are you getting graphs from anywhere in this, Boris himself said you’re load of shit.

Wow, that's amazing! It's incredible to see how fast things are evolving. It feels like we're on the cusp of something huge.

Link to original video?

Man. It's just a flowchart (graph) with retry logic (loop). Not new invention. People love fancy buzzword tho.

Prompting didn't die, it moved into the critic node. A graph without a scoring step is just a straight line with more edges - the self-improvement comes from the eval signal, not the topology. #AgenticAI #ContextEngineering @AnthropicAI

this information should not be publicly available :b

Like, comment. I’ll send you a complete guide on how to automate crypto content creation on X. <3

@grok make a summary i aint watching allat