Recently, in the group chat about AI, everyone is more and more obsessed. Several group friends who usually don't touch code now casually say things like "I had AI write a script" or "I put together a little tool." They've practically become half engineers. Looking at these messages, a question suddenly popped up: What about those brothers who really rely on coding for a living? Those who have been doing it for ten or eight years—what are they doing now? Even outsiders can now command AI to type a few lines; for professionals, are they becoming more valuable, or is it the opposite—being pushed out first? This kind of question is useless just spinning in the mind; you have to see a living example. @CreaoAI, this company, lives quite extremely. 25 people, 10 engineers, 99% of production code is written by AI. To what extent? A new feature goes live in the morning. At noon, A/B testing shows the data isn't good. In the afternoon, they cut it directly. By evening, they replace it with a better version and relaunch. The same thing three months ago would have taken them six weeks. Seeing this, I thought: Oh no, those brothers who used to code might really be getting pushed out. But the deeper I dug, the more I felt something was off. Panic is warranted, but the panic is in the wrong direction. → → → What exactly did they do right? First, distinguish two terms; this is the watershed of everything. Most companies use AI like this: ■ Engineers open Cursor to write code. ■ Product managers use ChatGPT to write documents. ■ Testers use AI to patch test cases. The process doesn't change at all; efficiency rises by 10% or 20%, and that's the limit. This is called AI-assisted. Hanging an AI on the old process. CREAO does something else. They dismantle the entire process and reimagine it: AI is the main builder, and humans only provide direction and make judgments. It's not about "how AI helps engineers." It's about "how to rearrange all steps so AI builds and humans oversee." ➤ Here's the harshest example: AI can build a feature in two hours, but product managers still take weeks to write requirements. At that moment, the bottleneck isn't development but the product manager. So they removed that position entirely. Reconstructing the process, when done ruthlessly, even changes the organization. This is completely different from just giving engineers an AI tool. One is addition; the other is multiplication. These two are not the same. → → → So what exactly is the product they make? Here's a particularly counterintuitive point. The truly valuable thing is never the Agent itself. Agents make mistakes, hallucinate, and take paths you don't understand. The more you try to make it smarter, the less controllable it becomes. The real product is the harness outside the Agent. OpenAI gave it a name: harness engineering. >> Meaning: the core work of an engineering team is no longer writing code but building a framework that AI can understand, verify, and constrain to work within. When problems arise, the solution is never to make AI try harder. >> It's about asking: what capability is missing? How to supplement it with rules that AI can read and follow. This idea sounds abstract. On the ground, it’s this thing that impresses me. --- Who backs up the AI? This is the most asked question to CREAO. The answer is not "hire more QA." They don't have a QA team at all. Every morning at 9, an AI automatically scans all service logs and generates a health report. No one needs to ask for it. An hour later, another AI clusters online errors, scores them across nine dimensions, automatically opens tickets, and even directly submits a fix PR. Over 50% of bugs are fixed by AI itself. No QA, no testing environment with a bunch of people clicking through before launch. Instead, there is an "AI jury." Three large models from different vendors score every online response. If the score drops, new code won't go live. AI tests it, AI approves it, and if something goes wrong, AI recalls it. This system can fix itself. —— ➤ At this point, the most interesting part of CREAO emerges. It itself is a platform for making Agents. In other words: they use Agents to build a platform that builds Agents. This sounds like a tongue twister but is the strongest self-proof of the entire methodology. If this harness can truly let AI independently do the work well, then the first product it should deliver is this company itself. Being able to build itself—that’s when it counts. All these efficiency numbers are disclosed by themselves. Whether it can be replicated and how much is exaggerated, you can judge for yourself. But one thing I keep thinking about: Their CTO, a physics PhD and former Meta LLaMA team member, said he didn’t write a single line of code in 2026. He said the most valuable thing for engineers going forward isn’t coding speed but finding loopholes in AI’s plans and judging what truly has value. This is worth every person still staying up late grinding new frameworks to pause and think for three seconds. Note: The above content is purely project analysis and does not constitute any investment advice, ha, DORY

Smoking downstairs this afternoon, I ran into a guy who works as a second-hand real estate agent. He has several big properties in hand. They've been listed for half a year without a single viewing. He's been sighing all the time. Heavy assets are just like that. > If liquidity disappears, having tens of millions in hand looks like a lot. But it's actually just a string of numbers; you can't even cash out enough for a hotpot meal. This is also the easiest pitfall in the crypto world when doing RWA (Real World Assets on-chain). At the start of a project, there's no funding, and the underlying assets can't offer decent interest, so retail investors glance and immediately leave. ---- But today I carefully analyzed Manifest Finance's US real estate token (USH). I found they took a shortcut. The real winner in this game is actually @ethena. ➢ Manifest’s early issued real estate tokens ➢ The underlying assets are actually all Ethena’s USDe. This play of collecting interest first, then delivering physical assets, is quite clever. Retail investors rush in and immediately get to enjoy Ethena’s ready-made arbitrage profits. Once they stabilize the user funds and fatten the pool, They slowly replace the underlying assets with 80% real US real estate. The deadlock of cold start is thus revitalized. This also shows that behind Ethena’s white-label service, there’s actually a significant strategic layout. // Previously, project teams wanting stablecoin yields Had to build their own hedging models and calculate liquidation lines daily. Now Ethena directly made this interest-generating engine into a standardized plugin. Project teams just need to put their own brand on it to start business, outsourcing all the dirty and hard work. Look at Jupiter, Sui, MegaETH, And now add Manifest, the real estate player. Ethena doesn’t care if your frontend is speculating on public chains, DEXs, or real estate. It just focuses on pumping interest-bearing dollars in the backend. Silently becoming the underlying blood pump for all these protocols. At the end of the day, this is a ruthless business strategy. ➤ It doesn’t compete for the spotlight on the front stage; it hides in the back collecting tolls from the entire track. The cigarette is finished. My buddy’s crappy house probably still has to be held onto; in the real world, there’s no one-click arbitrage plugin. Note: The above is purely personal thoughts and not any investment advice. #Ethena #USDe #RWA #DeFi