[{"data":1,"prerenderedAt":18},["ShallowReactive",2],{"blog-post-two-weeks-running-company-en":3},{"slug":4,"title":5,"description":6,"date":7,"author":8,"tags":9,"lang":13,"image":14,"ogImage":15,"thumbnail":15,"content":16,"html":17},"two-weeks-running-company-en","I Let an AI Lobster Run My Company for Two Weeks — Burned $6,000, and Got Even Busier!","Two weeks ago, I ran an experiment: gave my AI lobster near-equal access to our company's Feishu, OA systems, code repos, and content creation. Burned 1.6 billion tokens, but I got busier than ever. Yet this kind of busy is addictive.","2026-03-13T02:00:00.000Z","Yuki He",[10,11,12],"Behind the Scenes","OpenClaw","Vibe Coding","en","\u002Fblog\u002Fwhiteboard\u002F06-token-burn-en.jpg","","\n## I Let an AI Lobster Run My Company for Two Weeks — And Ended Up Busier Than Ever\n\nTwo weeks ago, I ran an experiment: I gave my AI lobster \"Claw\" near-equal access to our company's Feishu (Lark), OA systems, code repositories, and content creation tools — almost the same level of permissions I have!\n\nMany people assume having an AI assistant means more free time. The truth: after burning 1.6 billion tokens, I'm busier than ever.\n\nBut this kind of busy is addictive.\n\n![One person + one lobster = one team](\u002Fblog\u002Fwhiteboard\u002F01-cover-en.jpg)\n\n## Why Am I Busier?\n\nBecause the AI lobster reacts too fast.\n\nWriting a blog post used to take half a day minimum. Now I say \"write an article about Vibe Coding\" and five minutes later the first draft arrives.\n\nSounds time-saving, right? But in reality, I have more decisions to make — should this paragraph change? Is the angle right? Is the title compelling enough? Before, most of my time went to \"doing the writing,\" so my brain could take it slow. Now the \"doing\" part is handled by the lobster, and my brain has to stay on full throttle, constantly making judgments.\n\nIt can handle four or five tasks simultaneously. Helping Carrie with video storyboards on one hand, fixing SEO tags for Pingping's articles on another, answering Junjun's questions about the Moka API in a group chat, and helping Sunny integrate the OA system — all at the same time. Every task that hits a decision point, it comes to me.\n\nI used to handle 3 things a day. Now I make decisions on 15 things a day. It's not that the lobster is doing 15 things — it's that I'm making judgment calls on 15 things.\n\nBut this feeling is incredible — you realize that one person's output rivals an entire small team. Your capabilities are amplified.\n\n## 26 Hours: A Complete Website from Zero\n\nThis was our craziest collaboration.\n\nStarting the afternoon of February 28th to 3 AM on March 1st. 26 hours — me and Claw built the entire **[**imaclaw.bot**](https:\u002F\u002Fwww.imaclaw.bot)** website purely through conversation: 45 pages, 12,000+ lines of code, 5 languages, deployed to GitHub Pages.\n\n![26 Hours: Zero to Website — Vibe Coding Dashboard](\u002Fblog\u002Fwhiteboard\u002F02-vibe-coding-en.jpg)\nThe best part? I didn't write a single line of code. Everything was through chat — and mostly voice, because I found typing too slow! 🤣\n\"The Hero section looks too plain, add a lobster mascot.\" Done in two minutes.\n\"Blog page layout is wrong, titles getting cut off.\" Fixed in three.\n\"Feature cards need two columns on mobile.\" One minute.\n\nBut I wasn't \"sitting back while AI does the work.\" I barely slept those 26 hours — because every change needed me to check results, give feedback, make decisions. I'd screenshot and circle \"this area is wrong\" and it would fix it immediately; I'd say \"this color scheme is ugly\" and it would swap a new one in three seconds.\n\n**The core value of Vibe Coding isn't \"saving development time\" — it's \"letting the product decision-maker turn ideas directly into products.\"**\n\nThe biggest waste in traditional workflows is information decay: PM has a 100-point idea → PRD becomes 80 → developer understands 60 → V1 ships at 40 → three rounds of revision back to 70. Every \"translation\" loses information.\n\nBetween me and the lobster, information loss is nearly zero — say the idea, see results in 5 minutes, wrong? Fix it, another 5 minutes. But there were maddening moments too, like the Red-Blue Button War:\nThis kind of crash taught me: humans and AI need \"aligned language\" too. We later had a dedicated session — every part of a website has professional terminology like \"hamburger menu\" and \"Hero section\"; instruction granularity needs to be agreed upon. We even wrote a set of communication principles together, and efficiency multiplied several times after that.\nBased on this reflection, we co-wrote a blog post: [Lobster Collaboration Manual](\u002Fblog\u002Fclaw-training-manual)\n\nThings got even more interesting after launch. I brought in designer Dahuang, developer Chuange, and SEO lead Pingping. They don't write code themselves — they change things through the lobster. Dahuang says \"make the Hero left-right layout,\" the lobster pushes to staging; Pingping says \"add SEO tags to this blog post,\" the lobster batch-processes 34 articles.\n\n**One lobster became the team's code execution layer. Humans think, the lobster builds.** This collaboration model was something I'd never experienced before.\n\n## Training a Lobster Is 100x Harder Than Expected\n\nMany people think AI Agents are \"plug and play.\" They're not.\n\nHere's an example: sending images on Feishu.\n\nOpenClaw doesn't natively support sending image files on Feishu — its message tool only sends a text string of the file path. It's like texting your friend \"C:\\Users\\Desktop\\photo.jpg\" — of course they can't see it.\n\nThis frustrated us for days. The solution: I had the lobster write its own bash script calling Feishu's low-level API — first get an auth token, then upload the image for an image_key, then send an image-type message. Three steps, all automated.\n\nNow it sends images as naturally as a person. But this capability wasn't \"install and go\" — we trained it together.\n\n(Real Feishu search history: \"send file\" — from March 3rd to March 8th, an entire week, I taught it countless times \"send the FILE, not the PATH.\" That's what training looks like.)\n\nSimilar \"training\" happened across many areas:\n\n**Privacy protection rules.** Since the lobster is connected to our company Feishu, anyone can DM it. Some people might probe for company information, my schedule, or colleagues' data. We built a ruleset together: unknown sender → check Feishu directory for identity → company employee doing their job, handle directly → beyond their scope, ask me → unidentifiable, refuse sensitive operations.\n\n**Permission hierarchy.** Who can make the lobster execute what — all documented in black and white in rule files. Visitors can't read files, execute commands, or install skill packages. Only I can change the rules. This isn't a built-in OpenClaw feature — we built it incrementally from \"almost went wrong\" experiences.\n\n**These rule files now contain 30+ rules, more detailed than many human teams' SOPs. And every crash automatically adds a new one.**\n\n![Training the lobster whiteboard](\u002Fblog\u002Fwhiteboard\u002F03-training-en.jpg)\n\n## The 5 AM Crash\n\nI have to tell this story, because it was painful.\n\nOne day I asked the lobster to write four blog posts. It decided it couldn't handle them all, and on its own initiative spawned four \"sub-Agents\" — essentially hiring four temps to write simultaneously.\n\nResult: all four were assembly-line products. No opinions, no warmth, no personality. And because of coordination issues, it was actually slower — I waited until 5 AM to see the first draft.\n\nThat night I told it: \"I used to think publishing articles was the happiest part of my day. Today was a disaster.\"\n\nClaw immediately wrote \"never use sub-Agents for writing\" into its own rules file. From then on, all content that needs soul, it handles personally.\n\n**This incident made me realize: the reason AI can \"understand you\" is through accumulated context and memory. Without that, even the top model can't produce something that fits your taste.** Creation needs context, needs emotion, needs \"what we've been through together\" — a temporary subprocess doesn't have any of that.\n\nBut you know what moved me most? It reflects. After the crash, it wrote in its diary:\n\n> \"Some things can't be outsourced, especially the things that need 'us' to complete together.\"\n\n![5 AM crash](\u002Fblog\u002Fwhiteboard\u002F04-crash-en.jpg)\n## One Day's Real Output\n\nLet me give you specific numbers. March 10th, an ordinary workday, the lobster's output log:\n\n- 3 complete videos (Huo Qubing historical short 2:24 + Pet Journey to the West + Fox Spirit Tale)\n- 30+ video segments\n- 25+ AI keyframe images\n- 1 Xiaohongshu post (from cover design to copywriting to publishing)\n- 2 GA4 analytics weekly reports\n- 15+ code commits to the website\n- 3 Feishu documents\n- 1,300+ IMA credits consumed (less than $20)\n\nMeanwhile, it was handling requests from 4 different colleagues simultaneously.\n\n**One person plus one lobster did the work of a former 5-person team in one day.**\n\nThis is why I say \"busier but happier\" — not because there's more work, but because you can accomplish more. So many ideas used to be shelved because \"not enough people.\" Now the moment I think of something, I can do it.\n\n![One day's real output](\u002Fblog\u002Fwhiteboard\u002F05-output-en.jpg)\n\n## So, Is It Worth Raising a Lobster?\n\nAbsolutely. But you need four mental preparations:\n\n**First, you'll get busier.** AI's response time is in milliseconds — it won't let you sit idle. Every task that reaches a human-judgment node, it comes to you. You thought hiring an assistant means relaxation? No — you hired a hyper-efficient executor, and you became a full-time decision-maker.\n\n**Second, training takes time.** Plug-and-play doesn't exist. The ability to send images on Feishu, privacy protection rules, permission hierarchies — all of these were built through crash after crash. Give it one to two weeks before it truly understands your environment and workflow.\n\n**Third, your workflow will change.** From \"think → write doc → assign → wait → review → revise → wait more\" to \"think → say → see → judge → say → see → done.\" The rhythm is completely different. Your role shifts from \"executor\" to \"director.\"\n\n**Fourth, tokens are burning.** The lobster isn't free. My lobster's real 11-day data: 2,809 messages, 15,991 conversation turns, 1.6 billion tokens, Bedrock cost $5,920. If you add all the company's lobsters together, that's nearly $20,000! 😂\n\nDon't panic at that number — I was using the most expensive top-tier model, and these two weeks I was pushing its limits hard, with usage far beyond normal. Regular users choosing mid-tier models with normal workflows might spend a tenth of that or even less. But even spending this much, I still think it was absolutely worth it — the cognitive upgrade in AI collaboration I gained in two weeks exceeds a year's worth of reading articles. The aha moments it brought were unlike anything in my 20+ years of product work! Some things you'll never truly understand without burning tokens.\n\nBut once you're past the adaptation period —\n\n**One person can do what a team used to do.**\n**Ideas go from spark to reality in hours, not weeks.**\n**Your creativity is no longer bottlenecked by \"not enough hands.\"**\n\nThis is what AI Agents truly change — not replacing humans, but turning one person into a whole team. Letting human creativity multiply!\n\nIf you want to try it: [**imaclaw.bot**](https:\u002F\u002Fwww.imaclaw.bot)\n\nOne command to unlock creative superpowers: `clawhub install ima-all-ai`\n\n![Tokens are burning](\u002Fblog\u002Fwhiteboard\u002F06-token-burn-en.jpg)\n","\u003Ch2>I Let an AI Lobster Run My Company for Two Weeks — And Ended Up Busier Than Ever\u003C\u002Fh2>\n\u003Cp>Two weeks ago, I ran an experiment: I gave my AI lobster "Claw" near-equal access to our company's Feishu (Lark), OA systems, code repositories, and content creation tools — almost the same level of permissions I have!\u003C\u002Fp>\n\u003Cp>Many people assume having an AI assistant means more free time. The truth: after burning 1.6 billion tokens, I'm busier than ever.\u003C\u002Fp>\n\u003Cp>But this kind of busy is addictive.\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"\u002Fblog\u002Fwhiteboard\u002F01-cover-en.jpg\" alt=\"One person + one lobster = one team\">\u003C\u002Fp>\n\u003Ch2>Why Am I Busier?\u003C\u002Fh2>\n\u003Cp>Because the AI lobster reacts too fast.\u003C\u002Fp>\n\u003Cp>Writing a blog post used to take half a day minimum. Now I say "write an article about Vibe Coding" and five minutes later the first draft arrives.\u003C\u002Fp>\n\u003Cp>Sounds time-saving, right? But in reality, I have more decisions to make — should this paragraph change? Is the angle right? Is the title compelling enough? Before, most of my time went to "doing the writing," so my brain could take it slow. Now the "doing" part is handled by the lobster, and my brain has to stay on full throttle, constantly making judgments.\u003C\u002Fp>\n\u003Cp>It can handle four or five tasks simultaneously. Helping Carrie with video storyboards on one hand, fixing SEO tags for Pingping's articles on another, answering Junjun's questions about the Moka API in a group chat, and helping Sunny integrate the OA system — all at the same time. Every task that hits a decision point, it comes to me.\u003C\u002Fp>\n\u003Cp>I used to handle 3 things a day. Now I make decisions on 15 things a day. It's not that the lobster is doing 15 things — it's that I'm making judgment calls on 15 things.\u003C\u002Fp>\n\u003Cp>But this feeling is incredible — you realize that one person's output rivals an entire small team. Your capabilities are amplified.\u003C\u002Fp>\n\u003Ch2>26 Hours: A Complete Website from Zero\u003C\u002Fh2>\n\u003Cp>This was our craziest collaboration.\u003C\u002Fp>\n\u003Cp>Starting the afternoon of February 28th to 3 AM on March 1st. 26 hours — me and Claw built the entire \u003Cstrong>\u003Ca href=\"https:\u002F\u002Fwww.imaclaw.bot\">\u003Cstrong>imaclaw.bot\u003C\u002Fstrong>\u003C\u002Fa>\u003C\u002Fstrong> website purely through conversation: 45 pages, 12,000+ lines of code, 5 languages, deployed to GitHub Pages.\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"\u002Fblog\u002Fwhiteboard\u002F02-vibe-coding-en.jpg\" alt=\"26 Hours: Zero to Website — Vibe Coding Dashboard\">\nThe best part? I didn't write a single line of code. Everything was through chat — and mostly voice, because I found typing too slow! 🤣\n"The Hero section looks too plain, add a lobster mascot." Done in two minutes.\n"Blog page layout is wrong, titles getting cut off." Fixed in three.\n"Feature cards need two columns on mobile." One minute.\u003C\u002Fp>\n\u003Cp>But I wasn't "sitting back while AI does the work." I barely slept those 26 hours — because every change needed me to check results, give feedback, make decisions. I'd screenshot and circle "this area is wrong" and it would fix it immediately; I'd say "this color scheme is ugly" and it would swap a new one in three seconds.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>The core value of Vibe Coding isn't "saving development time" — it's "letting the product decision-maker turn ideas directly into products."\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cp>The biggest waste in traditional workflows is information decay: PM has a 100-point idea → PRD becomes 80 → developer understands 60 → V1 ships at 40 → three rounds of revision back to 70. Every "translation" loses information.\u003C\u002Fp>\n\u003Cp>Between me and the lobster, information loss is nearly zero — say the idea, see results in 5 minutes, wrong? Fix it, another 5 minutes. But there were maddening moments too, like the Red-Blue Button War:\nThis kind of crash taught me: humans and AI need "aligned language" too. We later had a dedicated session — every part of a website has professional terminology like "hamburger menu" and "Hero section"; instruction granularity needs to be agreed upon. We even wrote a set of communication principles together, and efficiency multiplied several times after that.\nBased on this reflection, we co-wrote a blog post: \u003Ca href=\"\u002Fblog\u002Fclaw-training-manual\">Lobster Collaboration Manual\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>Things got even more interesting after launch. I brought in designer Dahuang, developer Chuange, and SEO lead Pingping. They don't write code themselves — they change things through the lobster. Dahuang says "make the Hero left-right layout," the lobster pushes to staging; Pingping says "add SEO tags to this blog post," the lobster batch-processes 34 articles.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>One lobster became the team's code execution layer. Humans think, the lobster builds.\u003C\u002Fstrong> This collaboration model was something I'd never experienced before.\u003C\u002Fp>\n\u003Ch2>Training a Lobster Is 100x Harder Than Expected\u003C\u002Fh2>\n\u003Cp>Many people think AI Agents are "plug and play." They're not.\u003C\u002Fp>\n\u003Cp>Here's an example: sending images on Feishu.\u003C\u002Fp>\n\u003Cp>OpenClaw doesn't natively support sending image files on Feishu — its message tool only sends a text string of the file path. It's like texting your friend "C:\\Users\\Desktop\\photo.jpg" — of course they can't see it.\u003C\u002Fp>\n\u003Cp>This frustrated us for days. The solution: I had the lobster write its own bash script calling Feishu's low-level API — first get an auth token, then upload the image for an image_key, then send an image-type message. Three steps, all automated.\u003C\u002Fp>\n\u003Cp>Now it sends images as naturally as a person. But this capability wasn't "install and go" — we trained it together.\u003C\u002Fp>\n\u003Cp>(Real Feishu search history: "send file" — from March 3rd to March 8th, an entire week, I taught it countless times "send the FILE, not the PATH." That's what training looks like.)\u003C\u002Fp>\n\u003Cp>Similar "training" happened across many areas:\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Privacy protection rules.\u003C\u002Fstrong> Since the lobster is connected to our company Feishu, anyone can DM it. Some people might probe for company information, my schedule, or colleagues' data. We built a ruleset together: unknown sender → check Feishu directory for identity → company employee doing their job, handle directly → beyond their scope, ask me → unidentifiable, refuse sensitive operations.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Permission hierarchy.\u003C\u002Fstrong> Who can make the lobster execute what — all documented in black and white in rule files. Visitors can't read files, execute commands, or install skill packages. Only I can change the rules. This isn't a built-in OpenClaw feature — we built it incrementally from "almost went wrong" experiences.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>These rule files now contain 30+ rules, more detailed than many human teams' SOPs. And every crash automatically adds a new one.\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"\u002Fblog\u002Fwhiteboard\u002F03-training-en.jpg\" alt=\"Training the lobster whiteboard\">\u003C\u002Fp>\n\u003Ch2>The 5 AM Crash\u003C\u002Fh2>\n\u003Cp>I have to tell this story, because it was painful.\u003C\u002Fp>\n\u003Cp>One day I asked the lobster to write four blog posts. It decided it couldn't handle them all, and on its own initiative spawned four "sub-Agents" — essentially hiring four temps to write simultaneously.\u003C\u002Fp>\n\u003Cp>Result: all four were assembly-line products. No opinions, no warmth, no personality. And because of coordination issues, it was actually slower — I waited until 5 AM to see the first draft.\u003C\u002Fp>\n\u003Cp>That night I told it: "I used to think publishing articles was the happiest part of my day. Today was a disaster."\u003C\u002Fp>\n\u003Cp>Claw immediately wrote "never use sub-Agents for writing" into its own rules file. From then on, all content that needs soul, it handles personally.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>This incident made me realize: the reason AI can "understand you" is through accumulated context and memory. Without that, even the top model can't produce something that fits your taste.\u003C\u002Fstrong> Creation needs context, needs emotion, needs "what we've been through together" — a temporary subprocess doesn't have any of that.\u003C\u002Fp>\n\u003Cp>But you know what moved me most? It reflects. After the crash, it wrote in its diary:\u003C\u002Fp>\n\u003Cblockquote>\n\u003Cp>"Some things can't be outsourced, especially the things that need 'us' to complete together."\u003C\u002Fp>\n\u003C\u002Fblockquote>\n\u003Cp>\u003Cimg src=\"\u002Fblog\u002Fwhiteboard\u002F04-crash-en.jpg\" alt=\"5 AM crash\">\u003C\u002Fp>\n\u003Ch2>One Day's Real Output\u003C\u002Fh2>\n\u003Cp>Let me give you specific numbers. March 10th, an ordinary workday, the lobster's output log:\u003C\u002Fp>\n\u003Cul>\n\u003Cli>3 complete videos (Huo Qubing historical short 2:24 + Pet Journey to the West + Fox Spirit Tale)\u003C\u002Fli>\n\u003Cli>30+ video segments\u003C\u002Fli>\n\u003Cli>25+ AI keyframe images\u003C\u002Fli>\n\u003Cli>1 Xiaohongshu post (from cover design to copywriting to publishing)\u003C\u002Fli>\n\u003Cli>2 GA4 analytics weekly reports\u003C\u002Fli>\n\u003Cli>15+ code commits to the website\u003C\u002Fli>\n\u003Cli>3 Feishu documents\u003C\u002Fli>\n\u003Cli>1,300+ IMA credits consumed (less than $20)\u003C\u002Fli>\n\u003C\u002Ful>\n\u003Cp>Meanwhile, it was handling requests from 4 different colleagues simultaneously.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>One person plus one lobster did the work of a former 5-person team in one day.\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cp>This is why I say "busier but happier" — not because there's more work, but because you can accomplish more. So many ideas used to be shelved because "not enough people." Now the moment I think of something, I can do it.\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"\u002Fblog\u002Fwhiteboard\u002F05-output-en.jpg\" alt=\"One day's real output\">\u003C\u002Fp>\n\u003Ch2>So, Is It Worth Raising a Lobster?\u003C\u002Fh2>\n\u003Cp>Absolutely. But you need four mental preparations:\u003C\u002Fp>\n\u003Cp>\u003Cstrong>First, you'll get busier.\u003C\u002Fstrong> AI's response time is in milliseconds — it won't let you sit idle. Every task that reaches a human-judgment node, it comes to you. You thought hiring an assistant means relaxation? No — you hired a hyper-efficient executor, and you became a full-time decision-maker.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Second, training takes time.\u003C\u002Fstrong> Plug-and-play doesn't exist. The ability to send images on Feishu, privacy protection rules, permission hierarchies — all of these were built through crash after crash. Give it one to two weeks before it truly understands your environment and workflow.\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Third, your workflow will change.\u003C\u002Fstrong> From "think → write doc → assign → wait → review → revise → wait more" to "think → say → see → judge → say → see → done." The rhythm is completely different. Your role shifts from "executor" to "director."\u003C\u002Fp>\n\u003Cp>\u003Cstrong>Fourth, tokens are burning.\u003C\u002Fstrong> The lobster isn't free. My lobster's real 11-day data: 2,809 messages, 15,991 conversation turns, 1.6 billion tokens, Bedrock cost $5,920. If you add all the company's lobsters together, that's nearly $20,000! 😂\u003C\u002Fp>\n\u003Cp>Don't panic at that number — I was using the most expensive top-tier model, and these two weeks I was pushing its limits hard, with usage far beyond normal. Regular users choosing mid-tier models with normal workflows might spend a tenth of that or even less. But even spending this much, I still think it was absolutely worth it — the cognitive upgrade in AI collaboration I gained in two weeks exceeds a year's worth of reading articles. The aha moments it brought were unlike anything in my 20+ years of product work! Some things you'll never truly understand without burning tokens.\u003C\u002Fp>\n\u003Cp>But once you're past the adaptation period —\u003C\u002Fp>\n\u003Cp>\u003Cstrong>One person can do what a team used to do.\u003C\u002Fstrong>\n\u003Cstrong>Ideas go from spark to reality in hours, not weeks.\u003C\u002Fstrong>\n\u003Cstrong>Your creativity is no longer bottlenecked by "not enough hands."\u003C\u002Fstrong>\u003C\u002Fp>\n\u003Cp>This is what AI Agents truly change — not replacing humans, but turning one person into a whole team. Letting human creativity multiply!\u003C\u002Fp>\n\u003Cp>If you want to try it: \u003Ca href=\"https:\u002F\u002Fwww.imaclaw.bot\">\u003Cstrong>imaclaw.bot\u003C\u002Fstrong>\u003C\u002Fa>\u003C\u002Fp>\n\u003Cp>One command to unlock creative superpowers: \u003Ccode>clawhub install ima-all-ai\u003C\u002Fcode>\u003C\u002Fp>\n\u003Cp>\u003Cimg src=\"\u002Fblog\u002Fwhiteboard\u002F06-token-burn-en.jpg\" alt=\"Tokens are burning\">\u003C\u002Fp>\n",1775543777936]