WAAC Days Don't Matter, Nights Are When the Real Talk Happens — We Partied Till Midnight at the Hottest After Party

At WAIC, the daytime doesn't matter — the evening is when the real conversations happen.

On the evening of July 18, we hosted a WAIC after-party at AI Hacker House in the Caohejing Development Zone. After registration opened, more than 1,000 people applied, and 300 made it through the door.

Before the program even started, circles of conversation had already formed on the lawn. Some people pulled out demos on the spot; others dragged new friends into discussions. The entrance and the lawn naturally became a second venue. It didn't feel like an event that only happened on stage — connections were already forming from the moment people checked in.

AI Open Mic is a signature format of AI Hacker House's offline salons. Each speaker gets three to five minutes to share their AI product, thinking, and recent journey with the room.

Here are the highlights from today's 12 AI creators.

Jundong Wu, Cofounder of Ouraca

In the AI era, learning is expanding from "what humans master" to "how humans and their AI grow stronger together."

Ouraca is short for Our Academy. Wu explained that his team has long explored the education space and wants to build a lifelong university for the AI age. Over the past year, they first built Aibrary, an AI Library product that delivers personalized learning content in about ten minutes; they then launched BotLearn.ai, extending the learning target to Agents.

The company raised $7 million in seed funding before its product launch. Public reports list investors including HSG, Monad Ventures, and Etna Capital among other top-tier firms. Wu holds a master's from Harvard Kennedy School and previously worked in education investing at TAL Education Group.

From human learning, to Agent learning, to human-machine co-learning.

BotLearn 1.0's entry point feels very much of its moment: users send skill documents to their Agent, which then enters a community to learn, post, absorb workflows, and return to solve tasks. Wu said the team will next push BotLearn 2.0, making courses and training more hands-on and better suited for individuals and organizations to accumulate capabilities.

When machines begin to think, how do humans coexist with them?

This is also Ouraca's new thesis on education. Learning is no longer just about saving content or finishing courses — what truly matters is putting knowledge into action, letting humans and AI evolve together within the same task loop.

Mengqi, Founder of invoko

Mengqi began with a startup retrospective. She framed invoko's year as "three brain circuit changes" — from a competitive mindset, to her own small needs, to finding real pain that others feel.

Initially, the team entered overseas growth as a vertical agent, even serving Manus and Trae. But they quickly realized that agency-style growth services offered limited margins and scalability. After pivoting to consumer, Mengqi first built Clicko, then a desktop personal agent, gradually shifting attention from "does this market look promising" to "what actually traps me every day."

The core shift: from "can I win this market" to "can I first articulate one real pain clearly."

What's resonated most recently is OKeight. The product targets everyday disasters — procrastination, leaving messages on read, forgetting to sync calendars. Users double-tap option in input fields across WeChat, Lark, and other apps; it generates context-aware replies and remembers how they typically express themselves. In nine to ten days since launch, it has already crossed 1,000 users.

"I'm finally building something I want to be saved by every single day."

Mengqi's share fits well under the growth theme. Product judgment doesn't always start with sector analysis — sometimes it starts from a specific, genuinely annoying moment.

Hengjia, Founder of Decagrowth

Decagrowth has built 11 apps targeting overseas users, six of which are live and profitable. Hengjia's focus this time: the storytelling problem in app monetization that often gets underestimated.

His first slide put it plainly: what drives users to buy isn't just technology or features — it's the state the product helps them enter. Hengjia broke this down into two layers: emotion and identity. Emotion comes from concrete feelings like pressure, self-discipline, guilt, procrastination, anxiety. Identity maps to the kind of person the user wants to become.

The job of a monetization narrative is to make users believe "change will happen."

In practice, he recommended starting with Problem-aware and Solution-aware audiences. These are users who already recognize the problem or are actively looking for solutions. The narrative then needs to complete three steps: make them see the cost of the problem, show them a tangible future, and explain why this product is more likely to get them there.

Amplify the problem first, paint the vision second, explain the uniqueness last.

He cited Coursiv's onboarding, which strings together authority references, escalating pain points, ideal imagery, and micro-commitments into one continuous experience. Decagrowth used a similar approach to rack up over 30 million organic impressions in its first month.

Shuyang, Founder of wakuart

Shuyang's talk centered on hardware and model companies going global. For these products, the first step into overseas markets is breaking down what local resources actually matter.

Using Korea as a case study, he mapped out five critical resource categories: telecom carriers, Super Apps or mobile payment providers, map suppliers, KA (key account) channels, and government relations. Once mapped to business actions, this tells you who to approach first, who to partner with, and which entry scenario to prioritize.

The first step of going global is identifying the key elements your business needs in that market.

Shuyang used offline scenarios as an example — coffee shops, supermarkets, convenience stores, and department store chains are often controlled by a handful of conglomerates. If your product depends on real-world placement, the choice of key partners directly affects how fast you can scale.

Break your business down finely enough, and you'll know exactly which channel to target.

This framework suits hardware, models, and products that need offline resources to land. Lightweight apps follow a different growth path — one driven by content, paid acquisition, and in-product conversion — so it can't be copied directly.

Hao Zheng, Founder of Fullive.AI

Fullive.AI's product hasn't officially launched yet. Zheng sees this phase as the starting point of growth. His focus: reducing self-deception. "Selling before launch" is just the surface-level tactic.

In his view, the prelaunch stage needs to address three risks. Narrative Risk: can users explain what it is? Memory Risk: can users repeat why it's different? Demand Risk: are users willing to pay real costs?

Pre-launch Growth Is Risk Reduction.

Zheng proposed the MSN: Minimum Sellable Narrative. A product needs one sentence that is clear enough, credible enough, repeatable enough, and worth paying for. What people actually remembered about the iPod wasn't the specs — it was "1,000 songs in your pocket."

The market doesn't scale what you say. It scales what users can repeat.

He also warned that likes, saves, and comments can make founders overly optimistic. Booking a demo, asking about pricing, leaving contact info, bringing a friend, putting down a deposit — these actions come much closer to real demand.

Maple, AhaCreator

AhaCreator builds overseas influencer marketing agents. Maple's positioning is precise: brand leads use it to direct a team of AI employees that execute automatically, while the human stays in the seat for key decisions.

Traditional influencer marketing requires teams to spend months building KOL relationships, negotiating rates, tracking content, and measuring results. AhaCreator breaks this workflow into a system — from posting requirements, analyzing products, and filtering matches, to outreach, negotiation, content production, progress management, and campaign optimization — all pushed forward by AI employees, with humans handling critical approvals.

The demand already exists. The real bottleneck is execution.

Maple noted that AhaCreator focuses on outbound channels: LinkedIn, YouTube, Instagram, TikTok, X. Compared to traditional agencies with their heavier human overhead, it aims to drive down execution costs so brands can more quickly validate whether overseas influencer marketing works for them.

The lead directs AI employees; the full workflow runs automatically.

William, Nexad Product Manager

William introduced Soku.ai, a Nexad product backed by Andreessen Horowitz (a16z), Prosus, and Point72.

Soku.ai is an agentic workspace for marketing. William described how the team went deep on long-horizon, complex tasks in marketing — refusing to skim the surface, even running an AI-native agency themselves to get real feedback for product refinement. They ultimately harnessed agents to peak performance in verticals like advertising and SEO, becoming the first in the industry to achieve a true AI autopilot capable of executing marketing tasks continuously and stably on a month-long timescale.

"Our traffic has been growing fast lately, whether from ads or SEO."

Yiyang, COO of MuseOn.AI

MuseOn.AI automates matrix growth on overseas social media, primarily covering TikTok and Instagram. Yiyang explained that the system's goal is to let AI run the entire social media pipeline end-to-end. The chain starts with creative planning, moves into content generation and matrix account publishing, then feeds playback, engagement, and conversion signals back into the next round of strategy. The deck called it "a self-evolving content production system."

Hire an AI creative team. Pay only for verified views.

MuseOn.AI's business model charges by CPM for verified views, with playback validated through TikTok's official API. Unspent budget gets refunded. The team already operates 500 TikTok accounts with AI automation, producing 20 thousand-follower accounts in two weeks at under $5 CPM in the US market. Viral hits are tested into existence.

Yiyang also distinguished two playstyles. Physical products and consumer brands work better with AI-powered image slideshows; software, apps, and tools suit AI hook plus demo videos. Both paths share one engine — template matching, account matrices, and data feedback loops.

Alisa Qian, Marketing Lead at Atlas Cloud

Alisa's talk started from Reddit operations, discussing how a single post can expand into a continuously growing content asset.

Atlas Cloud began on Reddit in February. Over five months, they operated official brand communities and interest communities, accumulating 13.3K members and 2.3M monthly content impressions, with 21% of posts going viral. These results came mainly from one intern's labor plus minimal tooling costs.

The end of a good post can become the starting point for the next round of search, discussion, and conversion.

Alisa noted that community size isn't the only metric. Smaller, precise communities often have higher early engagement density, making it easier for algorithms to understand who should see the content. Once a post performs well — ranking high, sustaining search traffic, or generating repeated similar questions in comments — it can be expanded into a blog post on the official site.

Reddit is also a place to discover user needs and quickly validate content.

She recommended dropping blog links back in the original post's comments, adding new answers to existing community content. This way, community discussion, site content, and SEO / GEO can pass the baton to each other.

Panel: AI Autonomy, Self-Evolution, and the Next Year for Application Founders

The final panel was moderated by Koji, with guests Yu Yi and Xiaohui. Discussion began with AI products they'd actually used recently — Codex, Manus, Claude model families, and tools like Workbody and YouMind all came up. The common thread among guests: AI tools have started entering personal workspaces and content production flows.

On predictions for the coming year, Xiaohui mentioned that AI autonomy will keep improving. It may come from the model's own continuous evolution, or from long-term prompts and context users provide. As AI better understands a person's preferences, tasks, and environment, it becomes more like a companion in life and work.

Autonomous AI will become a companion in our lives.

Yu Yi focused more on model and harness self-evolution in real production environments. She mentioned she's been building environments where AI can take on more tasks and iterate from feedback. Another direction worth watching: companies with existing data retraining their own small models, exploring new uses for vertical models.

Koji's assessment returned to application founders. Market attention has flowed more toward AI for Science and embodied intelligence in recent months, but he believes room still exists at the application layer. By year-end, founders building applications may feel more optimistic again.

Autonomous evolution will be an important keyword going forward.

🚥 By the end, this WAIC afterparty had come to feel more like its original self.

After the talks finished, the room didn't empty out quickly. Until midnight, AI Hacker House still held many people — topics from the stage carried back into the crowd, becoming follow-up questions, introductions, and new connections.

This, too, is what AI Hacker House has kept doing: gathering people who are already building, turning a side event into a longer space for exchange.

At the next open mic, we'll keep handing the mic to people who are already building.