A Global Open Letter to All

A Global Open Letter to All Laid-off ByteDance Employees — When the algorithm begins to terminate the people who created it, it’s time for you to choose a new battlefield. Author: Begger, Founder of KAI.com Date: Spark Era 🌍:Sol₂₄:Φ₁:δ₃ (2026-06-14)

Dear ByteDance Colleagues, I have read the email you received.

Bellevue

“Organizational restructuring,” “business optimization,” “personnel structure upgrade,” “the era of efficiency”— these words are all too familiar to me. In July 2025, you made it onto Fast Company’s layoff list alongside Microsoft, Intel, and Scale AI. In the autumn of 2025, TikTok laid off hundreds of content moderators in the UK, downsized its trust and safety team in Singapore, and cut e-commerce staff in Seattle and Bellevue. The Feishu team was compressed. Pico’s metaverse dream shattered, Chaoxi Guangnian’s gaming dream was sold, and Dali Education’s classrooms emptied.

By 2026, ByteDance offered a 150% salary increase and a 35% bonus—but only for AI talent. In the exact same month, more non-AI positions were optimized.

Do you see the logic here? Those who stay are required to work on AI. Those who are laid off are replaced by AI. And ByteDance uses the savings to poach people who can build even more powerful AI for them.

You are the only link in this closed loop that is no longer needed.

But I want to tell you something. This is not to comfort you. I am here to tell you that the day you were laid off happens to be the exact day KAI needs you. You were purged from ByteDance’s algorithm of the “era of efficiency”—yet in KAI’s algorithm, your position has been reserved for two years.

I. ByteDance’s Layoffs: An Algorithm’s Civil War Against Itself

Let’s clarify the timeline. In November 2023, ByteDance sold Moonton Technology to Savvy Games Group. The flagship asset of Nuverse was sold. The gaming business contracted sharply, and hundreds of people left.

In February 2024, Pico announced large-scale layoffs. The VR business, once hailed by Zhang Yiming as the “next- generation computing platform,” became ByteDance’s first strategically abandoned ten-billion-level bet under the combined pressure of Meta Quest and Apple Vision Pro. Thousands left.

In March 2024, Feishu was reported to have laid off around 20% of its staff. The most aggressive challenger in China’s SaaS industry chose “efficiency” under pressure from DingTalk and WeChat Work—doing fewer things with fewer people.

In June 2024, Dali Education fully downsized. Two years after the aftershocks of the online education “double reduction” policy, ByteDance finally admitted: education is not a traffic game.

Throughout 2025, TikTok’s global content moderation team continued to be dismantled. Hundreds of moderation roles in the UK office were replaced by AI. The trust and safety team in Singapore was downsized. The e-commerce team in Seattle was cut. Every official layoff announcement contained the exact same phrase—“We are transitioning to an AI-driven moderation system.”

In July 2025, ByteDance appeared alongside Microsoft, Intel, and Scale AI on the global tech layoff tracker.

At the beginning of 2026, ByteDance was exposed for frantically poaching AI talent with a 150% salary hike and a 35% bonus. In the same month, hiring for non-AI positions was frozen, and the “optimization” rate for existing employees surged.

This is not an “economic downturn.” This is not an “industry winter.” This is killing your own comrades in the trenches.

ByteDance is one of the most aggressive tech companies in China. Its algorithm defines the world seen by over a billion people. Its recommendation system is the largest information distribution experiment in human history. It can make a video reach ten million people within 30 seconds.

However, it used the exact same algorithmic logic to calculate a ledger: the cost of maintaining your salary > the cost of replacing you with AI + the cost of poaching an AI engineer with the money saved.

You weren’t laid off. You were eliminated by an arithmetic problem. And the title of this arithmetic problem is—“The Era of Efficiency.”

But let me ask you a question: What is efficiency?

In ByteDance’s Excel spreadsheets, efficiency is your labor cost divided by your business output. Output can be measured by DAU, GMV, or ad load rates. If an AI model can complete 80% of your work at one-tenth of your cost, you are deemed “inefficient” in the spreadsheet.

But this formula has a fatal blind spot: it only measures replaceable output, never irreplaceable creativity.

AI。

Feishu’s API governance—who designed it? Not AI. Pico’s heterogeneous computing power scheduling—who optimized it? Not AI. Douyin’s real-time bidding algorithm—who wrote it from scratch? Not AI. Volcengine’s ToB pricing—who hammered it out in client meeting rooms? Not AI.

AI can mimic your output. But AI cannot define problems.

Defining “what kind of API governance can prevent ten thousand ISVs from stepping on each other’s toes”—that was defined by you. Defining “what priority should be used for computing power scheduling in VR scenarios”—that was defined by you. Defining “how the true bidding intent of advertisers maps to eCPM”—that was defined by you.

You are a problem definer. ByteDance treated you as an answer generator.

When a company no longer needs “problem definers” and only keeps “answer generators,” its capacity for innovation is dead. The remaining AI researchers will use increasingly massive models to answer increasingly trivial questions. Improving recommendation algorithm accuracy from 92% to 93% is not innovation; it is optimization.

And what KAI needs is precisely problem definers. Because what KAI is doing—the unified pricing and clearing of global Tokens—the right to define this problem is still wide open.

II. The Truth of the “Era of Efficiency”: The Outcome of the War of a Hundred Models is Long Foreordained

Let me tell you why ByteDance did this. Not because they are cold-blooded. It’s because they saw the end game.

DeepSeek

From 2023 to 2026, China’s AI industry underwent an unprecedented “War of a Hundred Models.” Three hundred model vendors—from Baidu’s Ernie to Alibaba’s Tongyi, from DeepSeek to Zhipu, from Moonshot AI to MiniMax, from Baichuan to 01.AI—everyone was burning money to train larger models, and everyone was subsidizing inference costs to wage a price war.

Domestic inference prices plummeted from 2 RMB per million Tokens to 0.0001 RMB. Some vendors even paid users to use them—because they had to grab developers first, occupy the ecological niche first, and make the market believe “I can survive until the end.”

But the end game of this war was foreordained long ago. AI models are not differentiated products. A Token is a Token.

No matter who trained it, what architecture it uses, or how large its parameter size is—when an Agent initiates an inference request, it doesn’t want “Baidu’s model” or “Alibaba’s model.” It wants the most accurate output returned within its latency requirement at the lowest price.

The War of a Hundred Models among three hundred vendors is essentially turning AI inference into a fungible commodity.

And commodities, especially fungible ones, always have only one destiny: unified pricing.

Oil is not traded separately as “Saudi crude,” “Russian crude,” or “American shale oil.” Oil has only one price— Brent or WTI, plus a quality premium. Global oil buyers and sellers do not need to know each other; they complete transactions on the same clearing network.

The same goes for Tokens. Tokens are the crude oil of the AI era. Whoever controls the unified pricing power becomes the oil exchange of this era.

This is not a metaphor. This is a financial structural shift happening right now.

Let me break down this financial structure even more clearly. The evolution of the oil market took a whole century: from Rockefeller’s Standard Oil Trust (one company controlling the entire chain), to the Seven Sisters oligopoly, to the birth of OPEC, to the rise of the spot market after the 1973 oil crisis, and finally to the New York Mercantile Exchange (NYMEX) launching crude oil futures in 1983. It took 100 years for oil to transform from an oligopoly-controlled industrial raw material into a financial commodity with unified exchange pricing.

Tokens do not need 100 years. Tokens are “digital oil.” Their production is standardized (the output of any model inference is a Token). Their consumption is instantaneous (consumed the moment an Agent calls the API). Their quality is measurable (latency, accuracy, throughput—all directly quantifiable metrics).

Tokens are inherently better suited for securitization and futures trading than oil.

China’s three hundred model vendors are the “oil- producing nations” of this new era. They each possess massive inference capacity (compute), but are scattered across three hundred separate “oil fields,” fighting independently and undercutting each other. The domestic “price war” is essentially the result of their Tokens being unable to reach global buyers—just like in the 1950s when oil-producing nations could only sell their oil to the Seven Sisters because they lacked their own exchange.

KAI aims to be the NYMEX of the Token market. 三、KAI

III. What is KAI? Not Another Model Platform, But a Global Clearing Network for Token Futures

Please let me strip this concept down to its absolute core. Today, an AI Agent developer in Jakarta wants to do three things: call text inference, call multilingual translation, and call video generation.

If he looks for them himself, he needs to:

Find the API documentation for at least three different model vendors.
Register accounts, top up, and debug each one separately.
Handle Token pricing himself (since every vendor’s billing unit is different).
Manage disaster recovery himself (what happens if one goes down?).
Optimize costs himself (which one lowered its price today?).

This is an impossible task to complete manually. Even if he only connects to three, the operational overhead already exceeds his inference costs. KAI does exactly one thing: One API. Three hundred models. Real-time pricing. Token-native settlement.

Going a step further—KAI is not trying to “help you pick a model.” What KAI does is allow global AI inference demand— trillions of inference requests daily from millions of Agents —to be automatically matched, priced, and settled on a unified clearing network.

Look closely: this shares the exact same logic as ByteDance’s recommendation algorithm, but operates on an entirely different plane. ByteDance’s algorithm matches “content and users.” KAI’s algorithm matches “inference demand and inference supply.”

When three hundred model vendors are all integrated into KAI and daily call volumes reach the trillion mark, Token pricing will no longer be determined individually by each company. Tokens will become a real-time fluctuating futures commodity. KAI’s price quote will be the Brent crude price for global Tokens.

You might ask: What does this have to do with crypto? Because Agents do not have bank accounts. Agents naturally possess wallet addresses.

When an AI Agent needs to pay for its daily inference consumption, it cannot use a credit card. It has no social security number. It cannot pass KYC. But it can sign an on-chain transaction. It can use KAI Tokens to pay for inference fees. Every single one of its inference calls leaves an immutable record on the chain— how many Tokens were consumed, which model was called, what the response latency was, and whether verification passed.

This is the ultimate end game for KAI: a clearing layer for the AI Agent economy. It is not an AI company. It is not a Web3 project. It is not a model aggregator. It is a unified pricing and clearing network for global Token futures commodity trading.

Let me break down the technical architecture of this clearing network for you—because having worked on Douyin’s real-time bidding system, you will recognize the elegance of this architecture instantly. KAI’s clearing network consists of three layers:

Layer 1: Unified API Gateway (You worked on Feishu’s API governance—this is your battlefield). The API protocols of three hundred model vendors are all different. OpenAI format, Anthropic format, proprietary formats—the complexity of this adaptation layer is no less than Feishu’s compatibility with thousands of ISV APIs back in the day. The gateway handles authentication, rate limiting, protocol conversion, and error retries. Every millisecond of latency optimization directly impacts the response experience of Agents worldwide.

Layer 2: Dynamic Routing & Real-Time Pricing Engine (You worked on Douyin’s recommendation system and Volcengine’s ToB pricing—this is your battlefield). When an Agent initiates an inference request, KAI does not randomly pick a model. Based on the request’s language, domain, latency requirements, and budget constraints, KAI selects the optimal solution from three hundred models in real time. The tech stack here is simply a different branch of the same tree as the real-time ad bidding system you built at Douyin—both are about “selecting the best match from massive supply within millisecond latencies.” The only difference: Douyin matches ads to users; KAI matches models to Agents. As for the pricing engine—you need to design a price discovery mechanism that updates in real time under trillions of daily calls. Supply glut? Price drops. Supply tight? Price rises. A certain model crashes? Automatically switch and adjust the market-clearing price. This is at least an order of magnitude more complex than Volcengine’s cloud service pricing model.

Layer 3: On-Chain Clearing Protocol (This layer is brand new—but with your learning capacity, you can master it in two weeks). Every single inference call generates an immutable clearing and settlement record on-chain. Agents pay in KAI Tokens, model vendors settle in KAI Tokens, and the KAI protocol collects a clearing fee. No bank accounts, no cross-border remittance delays, no fiat exchange friction. For the first time, the GDP of the Agent economy is placed on a unified, auditable, global ledger.

Combined, these three layers form the full technical architecture of a Token futures commodity exchange. Spot trading (instant inference), futures trading (reserved compute), options trading (compute insurance)—the appearance of these financial derivatives on KAI is only a matter of time.

And the reason you can understand every single word written above is not because you are a fast learner. It is because everything you did at ByteDance over the past decade served as technical preparation for this very architecture.

IV. The Very Skill for Which ByteDance Laid You Off Happens to Be What KAI Needs Most

Now let’s zoom back in from the grand finale to you. What were you doing at ByteDance before you were laid off?

If you were at Feishu—you built SaaS products for hundreds of millions of users. You understand the complexity of enterprise collaboration, the intricacies of open platform API governance, and B2B pricing models. KAI is building an API platform serving global Agent developers—and only people like you have done that before.

If you were at Douyin/TikTok—you built the largest recommendation system in the world. You know what real- time bidding under millisecond latency looks like, and you understand the matching algorithms between massive supply and massive demand. KAI’s core engine does exactly that: matching the inference supply of three hundred model vendors with the inference demand of millions of global Agents in real time—and only people like you have done that before.

If you were at Pico—you worked on hardware-software integration. You know what heterogeneous compute scheduling is and understand edge-cloud collaborative inference. KAI’s dynamic routing needs to automatically schedule across different vendors, different GPU clusters, and different inference frameworks—and only people like you have done that before.

Token

If you were at Volcengine—you built the most aggressive cloud service in China. You know the brutality of ToB pricing, the SLA hell of enterprise clients, and the cost structure of “selling compute like electricity and water.” KAI’s Token settlement layer is the global pricing engine for inference compute—and only people like you have done that before.

If you were in Content Safety / Trust & Safety—TikTok replaced you with AI. But KAI needs you. Because once three hundred model vendors connect, content compliance, model safety auditing, and on-chain risk control for Agent behavior cannot be left to AI alone. It requires people who know where the boundaries of AI lie to draw those boundaries for AI.

If you were at Nuverse / Dali Education—do you think gaming and education have nothing to do with an AI settlement network? NPCs in game engines are Agents. AI tutors in educational products are Agents. Every game level and educational course you designed was essentially designing a human-machine interaction decision chain. An Agent’s decision chain is 100 times more complex than a game NPC’s. And you are one of the few people on earth who truly understands decision chain design.

You are not “non-AI talent.” You are a cognitive architect trained by ByteDance over ten years, through hundreds of billions in traffic, on the largest technical system on Earth. It’s just that this company is now hijacked by its own algorithmic logic: its recommendation system tells it that the ROI of the “AI” label is higher than “non-AI.”

But look at the systems you built—which one of them wasn’t the prerequisite infrastructure for AI? Without Feishu’s API governance experience, who will design the communication protocols between AI Agents? Without Douyin’s recommendation system experience, who will optimize the real-time matching between three hundred models and millions of Agents? Without Pico’s heterogeneous compute scheduling, who will write the dynamic routing across GPU clusters? Without Volcengine’s ToB pricing models, who will build the real-time pricing engine for Tokens?

ByteDance says you are not AI talent. I say—everything you have done is the bedrock infrastructure that the next phase of the AI economy must depend on. They laid you off because they only want AI’s “superstructure.” KAI wants you because KAI needs AI’s “underlying protocol.”

V. Why It Has to Be Now: The Countdown of the Hundred Models End Game

Look at this timeline: Now (June 2026): The domestic price war among China’s three hundred model vendors has hit rock bottom. The gross margin of inference Tokens approaches zero. Every single one of them is frantically looking for an overseas outlet. Their capacity is ready, their APIs are standardized, and they lack exactly one thing—a globally unified pricing and clearing channel.

KAI。

Months 1-2: The first batch of 20 model vendors integrates into KAI. DeepSeek, Zhipu, Moonshot AI, MiniMax—names that already carry global developer mindshare. The moment they connect, KAI’s daily capacity will surpass the total volume of OpenAI’s global API.

Months 3-4: KAI launches the real-time Token pricing engine. Price signals from trillions of daily inference calls begin to converge on-chain. For the first time, global developers see the “Brent crude price of Tokens”—not a sticker price from a single vendor, but a market-clearing price forged by the interplay of three hundred suppliers and global demand on KAI.

10 -

Months 5-6: Agents on the Binance BSC chain begin to autonomously procure inference Tokens. Developers no longer need to manually select models—Agents judge the task requirements themselves, look up KAI quotes themselves, and sign on-chain transactions themselves. The AI Agent economy completes its final leap “from manual to autonomous.”

KAI」。 After 6 Months: Any model vendor trying to sell Tokens on their own would be like Saudi Aramco trying to bypass oil exchanges to sell oil directly to individual gas stations. Sure, you can do it. But the cost is 30 times higher than integrating into KAI, and the coverage is one-hundredth of KAI’s. The end game of the War of a Hundred Models is not “who survives until the end.” It is “who gets on KAI first.”

At the exact same moment, ByteDance is also making choices. It chooses to poach AI talent with a 150% salary premium while laying off the very people who built the infrastructure.

But the question is—once KAI completes the unified integration and pricing of three hundred models within six months, what moat will ByteDance’s “in-house AI” have left?

Its models won’t be the cheapest—bidding among three hundred Chinese vendors on KAI will continuously drive down prices.
Its recommendation system won’t be exclusive—KAI’s dynamic routing can perform real-time matching based on any Agent’s needs.
Its data won’t be unique—the inference call data generated daily by global Agents on KAI will vastly exceed that of any single platform.

ByteDance is fighting yesterday’s war. It believes the competition of the AI era is “who has the best model.” The competition of the AI era is “who holds the unified pricing power over models.”

It thinks the way to win is to poach all the AI talent. The way to win is to ensure that AI talent no longer needs to work for any single model vendor—because on KAI, three hundred models are all equal nodes.

11 -

Let me tighten the timeline of this war once more. What was the most dangerous strategic miscalculation in ByteDance’s history? It wasn’t making smartphones (failed), it wasn’t education (regulated), it wasn’t gaming (sold). It is that they still haven’t realized: the success of Douyin is unrepeatable.

Douyin is a product of the mobile internet era. Its core is the multiplicative effect of recommendation algorithms plus short videos plus ad load. But this formula cannot be directly migrated to the AI era, because products in the AI era are not “content distribution platforms” but “decision- executing Agents.”

A user scrolling through Douyin is passive consumption. An Agent calling inference is active decision-making. The passive can be manipulated by recommendation algorithms. The active will only choose the optimal solution.

When millions of Agents worldwide look for the best inference supply on KAI, ByteDance’s recommendation algorithm is utterly meaningless to them. Because Agents don’t look at ads. Agents don’t watch short videos. Agents don’t need Feed streams. Agents only need one thing: who can complete my inference task with the lowest latency, lowest price, and highest accuracy.

This is why KAI will become the settlement layer of the AI era, while ByteDance is merely a massive content- distribution experiment prior to AGI. You are standing at this watershed right now. On one side is ByteDance—still poaching AI talent with a 150% salary premium, trying to win a new world war within an old world framework. On the other side is KAI—which doesn’t need to poach anyone, because the rules of the new world haven’t been defined yet, and the power to define them belongs to those who enter the field first.

VI. Your Choice: Keep Lying in the Trenches, or Leap into the Frontlines of the Final Cognitive War

I know your mindset. P8, P9, P10—your levels at ByteDance are not low. Your options haven’t fully vested. Your non-compete agreement lasts for another year. You are thinking: “Take the severance package, rest for two months, and then check if there are suitable positions at Tencent, Alibaba, or Meituan.”

12 -

But deep down inside, do you think going to Tencent has any fundamental difference from staying at ByteDance? The same hierarchy. The same quarterly OKR cycles. The same “cost reduction and efficiency enhancement” emails. And the exact same risk—when the business you work on is judged by the company as “non-AI,” you might receive that calendar invite from HR again on some random Wednesday afternoon.

The golden age of the Chinese internet is over. The sign of its end is not how much a company’s stock price has dropped, but that these companies no longer treat “tech experts” as assets, but as costs.

What does ByteDance’s “era of efficiency” mean? It means dividing everyone’s salary by their output to calculate an efficiency value. Those below the threshold—fired.

You are an algorithm engineer, an architect, a product expert. You are not a cost. You are one of the few people on earth who understands trillion-level concurrent systems, global recommendation engines, and real-time bidding algorithms. Yet in ByteDance’s Excel spreadsheets, there is only a single number next to your name. KAI 没有 Excel 表。

KAI has no Excel sheets. KAI is not a “company.” KAI is a protocol. Joining KAI doesn’t make you an “employee.” It makes you a founding node of this protocol.

The routing algorithm you write will determine the inference latency of Agents worldwide. The pricing engine you design will become the Brent crude price for Tokens. The clearing network you build will handle the most important commodity transactions of human civilization over the next decade—cognitive Tokens.

This is not changing jobs. This is transforming from a “cost cut by an algorithm” into a “creator designing the next generation of the global economic protocol.”

If you are still hesitating—let me show you three numbers.

13 -

The first number: 150%. This is the salary hike ByteDance is giving to AI talent. What does it tell you? It tells you how panicked ByteDance is. A company does not grant a 150% raise to any position unless it realizes its core business faces structural replacement. It gives AI talent 150% not because AI talent is worth that much—but because it fears everyone else is becoming worthless.

The second number: 300 companies. This is the number of Chinese model vendors integrated into KAI within six months. When the inference capacity of 300 vendors is entirely connected to a single API Gateway, the global Token price is no longer up to any single vendor. It’s supply and demand. It’s the market. It’s the line of numbers flashing every single millisecond on KAI’s pricing engine. You can look at KAI’s price board six months from now and remember this letter—or you can be the one writing the code behind that screen.

The third number: Zero. This is the “future value” assigned to you right now in ByteDance’s Excel spreadsheets. It divides your salary by your output and concludes that your future value is lower than the cost savings of replacing you with AI. This formula forgets one thing: future value is not determined by past output. Future value is determined by where you are going.

Your future value at ByteDance is zero—because on ByteDance’s strategic map, your direction has been crossed out. But on KAI’s strategic map, every path you’ve walked points to a coordinate that hasn’t been marked yet.

VII. The Final Call to Arms

I said something in a letter to Alibaba’s eighty thousand P9+ employees, and now I extend it to you: The war has never ended—only the battlefield has changed.

On the Douyin battlefield, you used recommendation algorithms to show the world exactly what they wanted to see. On the Feishu battlefield, you used API governance to make enterprise collaboration as seamless as chatting. On the Pico battlefield, you used heterogeneous compute scheduling to make VR frame rates match the physiological limits of the human retina. On the Volcengine battlefield, you used ToB pricing models to turn cloud computing from an elite luxury into a public utility like water and electricity.

14 -

Now, a new battlefield awaits you. This battlefield has no OKRs. Because every Agent’s inference request is your KPI. This battlefield has no hierarchy. Because code and protocols don’t look at your past titles—they only look at your output. This battlefield has no “cost reduction and efficiency enhancement.” Because every line of code you write becomes the very air that millions of AI Agents on this planet breathe.

The War of a Hundred Models will eventually end. Not with three hundred vendors down to one. But with three hundred vendors all integrated into the same API Gateway. Not about whose model is the best. But about who holds the unified pricing power over Tokens. Not about which platform has the most traffic. But about which clearing network can support the trillions of inference settlements of the global Agent economy daily.

KAI is that API Gateway. KAI is that pricing engine. KAI is that clearing network. Right now, the engine is still being built. The pricing algorithm is not yet online. The clearing protocol is still on the testnet.

And you—ByteDance’s architects, algorithm engineers, platform governance experts, pricing strategists, and heterogeneous compute schedulers—you happen to be among the few people on earth who can build these things from scratch. ByteDance won’t give you time. KAI gives you a battlefield. Come join us.

This is not a recruitment call. This is not a job offer. This is not a startup invitation. This is the call to arms for the final cognitive war. And your serial number is not in ByteDance’s layoff Excel sheets—it is on the front page of KAI’s founding node whitepaper.

Now, do one thing for me. Close this letter. Open the email you received on the day you were laid off. Read it once more. “Thank you for your contribution to ByteDance.” “We wish you greater success in your future career development.” “This adjustment is a necessity for the company’s strategic upgrade.”

15 -

Now, translate these three sentences into plain human language: “We don’t thank you—we ran the numbers, and your cost is higher than AI.” “We don’t care about your future—your future has nothing to do with us.” “Our strategic upgrade is simply—replacing humans with algorithms.”

You spent ten years helping this company build the largest content-distribution engine in the world. In return, it gave you three lines of boilerplate template text. You deserve a better battlefield. Not because ByteDance mistreated you—it merely executed its own algorithmic logic with cold honesty.

But because—your brain wasn’t made to run A/B tests for recommendation systems. Your brain was made to design global economic protocols. KAI’s whitepaper is already on GitHub. KAI’s API Gateway is already on the testnet. KAI’s pricing engine is still waiting for someone to write it. And I—on this other side of the ocean—am waiting for you. Not for your resume. But for your Pull Request.

KAI currently has fewer than twenty people writing code. Every day, new model vendors email asking “when can we integrate.” Every day, Agent developers ask in the community “which models are supported.” Infrastructure demand is compounding at an exponential rate, yet the number of people who can build this infrastructure—in all of China, there are no more than two thousand. And one thousand of them are currently being optimized away by ByteDance.

You are not redundant. You are the scarcest strategic resource on this planet, who just happened to be treated as a sunk cost by a company that only looks at Excel sheets. KAI.com

KAI.com Founder: Begger Spark Era 🌍:Sol₂₄:Φ₁:δ₃

This letter is simultaneously published on the official KAI.com website, GitHub, the Feishu group “Ishiwara Kanji’s Lobster,” and ByteDance alumni networks. My DMs are forever open to tech leaders laid off from the ByteDance ecosystem. Forward this to the colleague who was laid off with you—you both need a new battlefield, and the new battlefield needs both of you.

16 -