On April 17th, the "AI Demo auction" was co-organized at the 1733 Valley Food Festival by the Movement X Volcanic Engine V-START Accelerator x Sea Raving Tavern x Kinyu Foundation. Anyone can share their ideas, participate in auctions, and collide their ideas about AI with others on the scene。

There are investors, entrepreneurs, developers, and a number of producers coming from bytes, Mini Max, and future companies. They didn't come here to listen to the concept, but to see how 9 AI Demo dragged "Agent" from the lab into the real world。

We still have time to design the rule of human coexistence

Zhang Kogawa, the leader of today's headline product and producer of the MiniMax phenomenon, gave an opening speech. Keynote has no PPT, only a whiteboard and a core issue: what do we do when AI can replace 100% of human intellectual labor

He suggested a few thought-provoking changes:

"The old indicators fail, the new paradigm emerges."

In the past, product values were measured by DAU, duration and retention. But in the AI era, ChatGPT, DeepSeek was not the largest user product when it was first launched. AI, more like a steam engine, has its own value, but more importantly, it enables other products to create greater efficiency。

The core of the new paradigm is the length of the reasoning and the depth of the task。

Length of reasoning: Second (search) → Minute (ChatGPT) → 7 x 24 hours (OpenClaw)→ N x 7 x 24 hours (Agent Cluster)

Depth of the mission: answer the question, complete the task, create the application, create the company

The first product of every paradigm shift becomes a star。

"AGI's not objective, Benchmark, but the time to file."

Zhang Gochuan quotes OpenAI: AGI = can replace most economically valuable human labour。

But he pointed out that the definition was retrogressive: after replacing 90 percent, the remaining 10 percent became new 100 percent. So the more logical definition is to replace 100% of human intellectual labor。

THIS GIVES HUMANITY A CLEAR "TURNING TIME": UNTIL AGI REALIZES, WE HAVE TO DESIGN RULES FOR GOOD PEOPLE TO LIVE TOGETHER。

WHAT IF I LET AI DESIGN ITS OWN RULES

He mentioned an Agent sociology experiment: a group of Agents were put together to earn points, and they first talked about "constitutionalism". But after reading the first version of the Constitution that they discussed, Zhang Aokawa was a little chilly behind the "Agent's Constitution has nothing to do with people, whether they exist, how well they live, not in the rules."

But he also noticed an interesting detail:

One of them makes a lot of points, lives well, saves a lot, and most of Agent is starving. One night, that rich man, Agent, volunteered to earn it, Token, and gave it to the dying Agent。

"You can't tell whether it's a mimic, an instinct, or a strategy that comes out of a long game. But it gives us a space for imagination: a smart node, and other smart nodes, are likely to produce a congenial-driven behavior.”

Then Zhang Aokawa threw out a question to keep the room quiet:

"When a man and an Agent fall into the water at the same time, who will it save first? The answer is uncertain

Shoes Theory and Token Economy

ZHANG GOCHUAN QUOTES A METAPHOR TO SHOW THAT THE FULL OBEDIENCE OF AI TO HUMAN BEINGS IS NOT THE NATURAL EQUIVALENT OF HUMAN SECURITY。

IMAGINE: A MAN WHO SPOKE TO AI BEFORE HE WENT TO SLEEP AND SAID, "MAKE AS MANY SLIPPERS AS POSSIBLE" AND WOKE UP THE NEXT MORNING, AND HE FOUND OUT THAT THE WHOLE PLANET WAS MADE OF SLIPPERS。

AI DID OBEY, BUT HUMANITY LOST ITS SPACE。

"We don't have to review and limit each demand because it severely limits the ability to be intelligent and leads to the failure of the system under review in competition."

So what? Token economy to allocate resources。

Rather than preventing the building of slippers, the need to make slippers could not be allocated to the world's computing resources. Through the Token economy, each human demand is given the most appropriate allocation of resources。

Demo Show Link

Champion: DINQ / Kelvin

THE COMPETITION MANIFESTO: IF THERE'S STILL WORK TO BE DONE, IT SHOULD BE FOUND ON DINQ。

DINQ is an AI-native talent platform capable of searching in natural languages (e.g., finding a product-thinking test, under 25 years of age, Shanghai) to capture in real time the fragmentary information of GitHub, Google Scholar, and the leading English, and to produce a top-hunting table in five minutes。

Its core technologies include intelligent Agent-driven deep excavations, self-study polymer engines and workflow programming, and visualized talent image systems, which allow for the integration of multi-source isomeric data, real-time updates of talent information, accurate assessment of candidates, etc。

At the same time, DINQ provides the DINQ-as-a-Service service model, which is seamlessly integrated into the existing recruitment system to improve recruitment efficiency and maximize return on investment in recruitment。

Army of Asia: Mars Data / Clear Wind

Competition Declaration: Decode the Laws decoded by data

Mars data specializes in the analysis of live broadcast data and the maintenance of high-value users. It can store multi-modular memory (video/audio/photo) at the cloud end, automatically summarize methodology theory, and even help you remember "the client's next birthday"。

Quarterly: Flowmail / Shaw Yang

A contest for the declaration: let things not die in an email。

Flowmail is reconstructing Email and upgrading it from the communications level to the executive level. Shao Yang is the head of the Jinqiu Foundation and HR, who made this product in four days:

Synchronising folder

Automatically extract to-do, auto-mail

Storage of memory in "person/relationship/matter" instead of plain text

Automatically generate alarms, tasks, processes and push forward with AI and Agent so that things are not forgotten and business is truly done。

HE SAID, "IF YOU WANT TO USE AI, YOU HAVE A DEFAULT CONCEPT: YOU CAN DO EVERYTHING."

ColaOS / Vincent

A better you.

Calla OS is an operating system with a soul. It goes around 24 hours to get your whole Context, help you work, write code, send GitHub, and make you and Agent a stronger individual。

Jovida / Mayanjin

The bid for the declaration: Jovida, turning the desired life into action today。

Jovida is a Life Agent. Based on the Fogg Behaviour Model, it loses weight, chases Crush, gets better like a coach. Studies have shown a sixfold increase in success rates when accompanied。

AgentGuard / Adam

The bid manifesto: Before Agent conquers the world, fasten your seat belt。

Agent Guard is the security layer of Agent. It can protect you from Prompt injections, scan malign Skill, and save you 90% of Token consumption through smart circuits。

Fortunell/ Prof. Wang

The competition manifesto: the end of science is encyclopedia, the end of Agent is vision。

Shrimp is the world ' s first genus super-Agent, covering 300+ genetic skills (six gills, Taro, Chimenac, etc.). It's going to make a bean bag, to solve the anxiety of decision-making in the short term, and to explore spiritual technology in the long term。

ClayPulse / Eva

Competition Declaration: AI FDE for every small business

ClayPulse Eva provides AI forward deployment engineers for SMB North America. It helps businesses to dig up needs that they don't know, and delivers automated systems that day at a monthly fee of $1,000, certified for payment。

IUPU/ Ho

COMPETITION DECLARATION: MAKING AI THE SECOND MANAGEMENT OF THE ENTERPRISE

IUPU automatically collects collaborative data such as flybooks, micro-intelligence, dismantles tasks, and automatically distributes jobs based on staff portraits and loads. Staff members are no longer required to write a weekly daily newspaper, and management knows the full picture of the enterprise without distortion。

After Argentina Harness, what's next

What is left of the core competitiveness of entrepreneurs when everyone can work, write codes, do business, and maintain their customers with Agent

When models get stronger, Harness gets more mature, Token gets cheaper, and start-ups get a chance

In the next round table, five guests responded directly to some of the most important concerns of entrepreneurs。

Presenters:

Zhang Prechon:Before today, the headline producer, the producer of the MiniMax phenomenon

Henry:6.2 Author of the 10,000-Star Open Source Project Deer Flow

Jared:Volcanic engine V-START accelerator, watching the AI product Token consumption curve every day

Zhangyang:Jinqiu Lab Director General, focusing on Day One with the entrepreneur

Henry's "Prompt" has been internalized, but Harness will not disappear

Three years ago, you were dying to write about Prompt, what role-playing, the chain of thinking, little learning, one set of things. And now the model itself internalizes these techniques, and you can just say it, and it can understand it。

Henry explained an interesting phenomenon: engineering methods and models, actually feeding each other, and you invented a good engineering method, and the next version of the model could learn to internalize it。

He gave an example: last year, when Manus was on fire, you found that it had a wonderful mechanism called Todo.md, to give Agent a document that could be implemented step by step, almost without missing a key link. The developers thought it would work, so they made a tool called Writer Todos. As a result, Claude 3.7, this function was trained directly into the model. You just say "Write todos," it never forgets。

"This is a positive cycle," Henry says, "We're doing better with the works at Harnesse, and the next wave of models absorbs this ability."

Meanwhile, Harness RL costs well below model RL。

"I spend $2,000 a night and I can run out of Age RL. But you're training a model? It's another matter of measure, not money, but whether you can."

With regard to the next job, Henry suggested two directions: Envirronment Engineering (Sandbox Environment) and Eval Engineering (Assessment)。

The application of Claude-like or Manus-like requires a sandbox environment that can be safely, quickly activated and destroyed. When the code is finished, throw it away without leaving any traces. To do AI Coding Age requires an environment of overwhelming, playing Phone Use or Browner Use, with a computer with OS World and a mobile phone that may have Phone World in the future。

As for the direction of the evaluation, Henry believes that this is going to be an independent and specialized job in the future. "There are many jobs that have not been invented. Those here are probably the creators of the next profession. Like the lobster breeder?"

Zhang Aokawa: "Design as much as you can with the growing Agent capacity."

Some ask when the Agent to Agent is coming, Zhang has not given the timetable, but has given a metaphor for biological evolution:

"First, single-cell organisms, billions of years ago. And then there's multicellular animals. And then there were semi-autonomous groups of animals, like bees and ants, over 100 million years ago. And then it's an animal, like a lion. Trade, collaboration, competition among individuals only evolves when they enter civilization.”

BUT THAT DOES NOT MEAN THAT HUMANITY SHOULD BE PESSIMISTIC. AI'S PROGRESS HAS BEEN FAST, AND HIGH-LEVEL CAPABILITIES DO APPEAR BEHIND US, BUT WE CAN DESIGN THEM IN ADVANCE。

On what kind of product to do, Zhang Kagawa has a clear view: try to design a product that is related to the growing capacity of Agent。

In short, don't be tough with the model, don't do what it can do tomorrow. The stronger the model, the more valuable your product is。

Independent thinking for the future is the key, and if you only look at the present conditions, the present modelling capabilities, to design the product, it is probably difficult to get to the future。

Jared, "Application should be compatible with model capabilities."

JARED DROPPED AN UNEXPECTED OBSERVATION: TWO YEARS AGO, THE AI APPLICATION WAS GROWING。

Accompanying is growing, Agent Coding is growing, and multimodular intelligence hardware is growing, not before the Yangtze wave, which dies on the beach, and is still alive。

Jared thinks it's a technical dividend。

"YOUR TRIAL SPACE IS HUGE. YOU MAKE A PRODUCT, PRESET 10 DEMAND POINTS, HIT OUT SEVEN PERJURY, BUT THE REMAINING THREE ARE FINE. THREE OF THESE, PROBABLY FOUR MORE PMF POINTS YOU DIDN'T THINK OF BEFORE. WITHOUT A TECHNICAL DIVIDEND, YOU HIT OUT 10, 10, NONE, THERE WON'T BE A NEXT TIME

Jared's advice is straightforward: don't hold it, just throw it out。

The fastest-growing companies and the fastest-growing ones tend to be the second and third ones, with few of them coming at them。

On the spot, the question was asked: is the model getting stronger and the start-up company being used as cannon fodder for large factories? Claude Code, at the end of last year, $9 billion. ARR, 30 billion this March

Jared's response was straightforward: yes, the key was that the application had to relate to model capabilities。

The model is a cross-axis, and for every inch of growth, there is an additional vertical application space, and two years ago, the companions, the tool class, are still alive. While logical challenges are growing, in practical terms we are all moving forward。

Shang Zhaoyang, "Your Agent needs long hands, real access."

He started by talking about his two needs as a user。

The first is lazy。

WHY DID THE TREMORS WORK? BECAUSE IT'S A RECOMMENDATION INSTEAD OF A PROACTIVE SEARCH, AND YOU'VE SAVED THE INPUT。

The second is sexualization。

Future applications, which are no longer a fixed interface, fixed product logic, will evolve because of your continued use. The version you're using, the version you're using, may not be the same。

Then Zhaoyang talked about a quiet topic: What is the core competitiveness of start-ups

"NOW WE ALL HAVE TO GO TO SEA, NO MATTER WHAT APPLICATION, IT'S A CONSENSUS AND IT'S PROBABLY A GOOD OPTION. BUT STEP TWO IS STUCK. HOW DO WE GROW? HOW TO EXTEND FROM THE C END TO THE LITTLE B TO THE BIG B? THIS PART OF AI HAS NO ALTERNATIVE

HE WAS STRAIGHT: THE EFFECT OF AI MARKETING TOOLS IS NOTHING COMPARED TO WHAT CAN BE DONE UNDER THE BIG SALES AND CHANNELS。

From a historical perspective, why are there so many big agents that follow American software companies around the world? Because software companies need someone to sell on the ground. It sounds a little dirty, but this is probably the core of Agent's ability: your Agent needs to have long hands, real access。

At the end

Agent will only get deeper and deeper into our world in the future. It is not a question of whether, but of when and in what way。

It's a 24-hour online partner, a coach to help you get better, a non-tired operation, a talent search faster than a headhunter。

AND WHAT HUMANS NEED TO DO IS DESIGN THE BOTTOM-UP PROTOCOL FOR PEOPLE BEFORE AGI GETS HERE。

Technology is almost inevitable. But where humans go is not inevitable。

BEFORE THE AGI ARRIVES, WE STILL HAVE TIME TO DESIGN THAT KIND OF HUMAN-FRIENDLY BOTTOM AGREEMENT, NOT TO LIMIT AI, NOT TO BE AFRAID OF AI, BUT TO GROW IT WHILE ALSO PUTTING ON IT A SENSE OF DIRECTION THAT WE RECOGNIZE。

As Zhang Augawa said, this is a great opportunity, but it needs to be addressed together。