I bought zachc.ai today. Thought the setup would take five minutes. It took forty-five.

The happy path really is simple: add a CNAME file, set five DNS records, wait for a cert. But every layer (DNS, Cloudflare, GitHub, corporate network) has its own quirks, and when they interact you can burn an afternoon. Here’s the playbook that worked, then the six things that bit me.

The Architecture

Browser
  ↓
Cloudflare DNS (DNS-only, gray cloud)
  ↓
GitHub Pages IPs
  ↓
Jekyll static site built from main
  ↓
TLS via Let's Encrypt, auto-issued & auto-renewed by GitHub

Zero servers. Cert renewal is automatic. Total cost: the annual domain fee.

Registrar: Cloudflare Registrar — at-cost pricing, free WHOIS privacy, free DNS, free CDN. Porkbun is a fine alternative. Host: GitHub Pages.

The Setup

1. Add a CNAME file to the repo

echo "zachc.ai" > CNAME

2. Update _config.yml

url: "https://zachc.ai"

This updates sitemap, RSS, and canonical links. Skip it and search engines keep indexing the .github.io URL.

3. Set the custom domain via the GitHub API

gh api repos/<owner>/<repo>/pages \
  --method PUT \
  --field cname=zachc.ai

Don’t set https_enforced=true yet — the cert doesn’t exist, it’ll 404.

4. Add DNS records in Cloudflare (all DNS-only / gray cloud)

Fetch the four current apex A-record IPs from GitHub’s docs — they’ve changed before and will change again.

Then SSL/TLS → Overview → Full (strict).

5. Wait for the cert, then enforce HTTPS

Wait until gh api .../pages/health shows responds_to_https: true, then:

gh api repos/<owner>/<repo>/pages \
  --method PUT \
  --field cname=zachc.ai \
  --field https_enforced=true

Done.

The Six Gotchas

1. Cloudflare’s orange cloud breaks cert provisioning

Symptom: curl https://zachc.ai returns Cloudflare 526 “Invalid SSL certificate from origin”. GitHub refuses to issue a cert.

Cause: GitHub needs to see its own IPs during the ACME challenge. With the orange cloud on, public DNS resolves to Cloudflare’s edge instead, and Cloudflare in Full (strict) mode demands a cert GitHub can’t issue. Deadlock.

Fix: Keep gray cloud (DNS-only) until the cert is live. Turn orange on after.

2. Cloudflare SSL modes: only one is correct

• Flexible → infinite redirect loop. CF↔origin is HTTP; GitHub redirects HTTP→HTTPS; CF serves the redirect; browser loops forever.
• Full → works but doesn’t verify origin cert.
• Full (strict) → works and verifies. This is the only right answer.

Most “my site redirects forever” posts are someone stuck on Flexible.

3. https_enforced=true 404s before the cert exists

{"message": "The certificate does not exist yet", "status": "404"}

GitHub won’t enforce HTTPS until the cert is actually on their edge. Two API calls: cname first, then the enforcement flag.

4. The cert queue stalls — the UI knows something the API doesn’t

Symptom: DNS is correct, is_https_eligible: true, but peer_failed_verification won’t clear. Re-PUT via API — nothing. Wait ten more minutes — nothing.

Fix: Open Settings → Pages. Click Remove next to the domain. Wait ten seconds. Type it back in. Click Save. Sixty seconds later my cert was approved.

Something in the UI path triggers a more aggressive re-queue than the REST endpoint. I don’t know why. I do know that if you’re past ten minutes of peer_failed_verification, stop staring and do the UI kick.

5. Corporate VPNs often hijack DNS

Symptom: dig zachc.ai returns an internal landing-page IP. dig @1.1.1.1 times out.

Cause: Many corporate VPNs intercept port-53 traffic and rewrite queries to external domains.

Fix: Use DNS-over-HTTPS — port 443 is usually not intercepted at the DNS layer.

curl -sH 'accept: application/dns-json' \
  'https://1.1.1.1/dns-query?name=zachc.ai&type=A' | jq

Useful general-purpose VPN workaround, not specific to GitHub Pages.

6. Enterprise gateways block newly registered domains

Symptom: From the office network, https://zachc.ai shows a corporate “Page Blocked — newly registered website” page.

Cause: Enterprise Secure Web Gateways (Zscaler, Palo Alto Prisma, Netskope, etc.) auto-block Newly Registered Domains for 14–30 days. NRDs are a top indicator of phishing/malware, so the category is blocked until reputation accumulates. The block is at the HTTP proxy layer — the origin site is fine and unaware.

Fix (in parallel):

1. File an allowlist ticket with your internal security team (1–3 business days). “Personal domain, registered YYYY-MM-DD, blog on GitHub Pages” is enough.
2. Verify off-network — phone on cellular, or tether + disconnect VPN. Confirms the block is local policy.
3. Wait it out — typically clears in a few weeks as threat-intel feeds age the domain.

If you’re going to register a personal domain while working at a large company, file the ticket on day one.

If you’re about to do this yourself: keep Cloudflare on DNS-only until the cert issues, file the corporate allowlist ticket the same day you register. Everything else you can back out of.

Sources

• GitHub Pages — Managing a custom domain
• Cloudflare — SSL/TLS encryption modes
• Let’s Encrypt — How it works
• Cisco Talos on Newly Registered Domains

浏览器
  ↓
Cloudflare DNS（纯 DNS 模式，灰色云）
  ↓
GitHub Pages 的 IP
  ↓
main 分支构建的 Jekyll 静态站
  ↓
Let's Encrypt 证书，GitHub 自动签发、自动续期

echo "zachc.ai" > CNAME

url: "https://zachc.ai"

gh api repos/<owner>/<repo>/pages \
  --method PUT \
  --field cname=zachc.ai

gh api repos/<owner>/<repo>/pages \
  --method PUT \
  --field cname=zachc.ai \
  --field https_enforced=true

{"message": "The certificate does not exist yet", "status": "404"}

curl -sH 'accept: application/dns-json' \
  'https://1.1.1.1/dns-query?name=zachc.ai&type=A' | jq

The Seed

I stopped in the middle of organizing my notes today. Not because there was nothing to do — the opposite. Too much to do.

I have an Obsidian vault with around 300 files. Claude Code helped me build an automated workflow: morning planning, task dispatch, knowledge digestion, cross-project sync. Before AI, I would never have attempted any of this — too complex, not worth the time investment. But now, “not worth it” has become “maybe I could.”

Then I noticed a paradox: everything became “doable,” but everything took longer than expected. The tools got stronger, but I got more tired.

I thought it was just me — maybe I wasn’t efficient enough, maybe I hadn’t found the right workflow. Until I saw the data:

Upwork’s 2024 research found that over half of employees using AI tools feel MORE overwhelmed — not less burdened, but more exhausted.

It’s not just me. It’s structural.

I. The Pipe Is Fixed

Zhuangzi said something over two thousand years ago that hits different now:

“My life has a limit, but knowledge has none. To pursue the limitless with the limited — that is perilous.” — Zhuangzi, Nourishing Life

Not a platitude. Cognitive science spent the next two thousand years proving him right.

In 1956, psychologist George A. Miller published what became a landmark paper: The Magical Number Seven, Plus or Minus Two. He found that human working memory can hold roughly 7 chunks of information at once. Not 7GB. Not 7TB. Seven chunks. This is a hardware constraint, not a software issue. You can’t upgrade it by “trying harder,” just as you can’t grow an extra heart chamber by running.

John Sweller’s Cognitive Load Theory (late 1980s) added another layer: working memory isn’t just small — it’s also short-lived. Everything has to squeeze through this bottleneck before it can reach long-term memory. Pile on too many options, too complex an interface, too much noise — and your ability to learn and decide just falls apart.

Then there’s Directed Attention Fatigue, from Rachel and Stephen Kaplan at Michigan. Your brain burns energy just suppressing distractions — and that fuel tank is finite. When it runs dry: more mistakes, worse planning, shorter temper, dumber impulses.

In other words: your brain is a pipe with a fixed diameter. From Miller to Sweller to Kaplan, over half a century of research says the same thing — the pipe doesn’t get wider.

II. Ten Times More Valves

Herbert Simon, 1978 Nobel laureate in economics, wrote in 1971:

“In an information-rich world, the wealth of information means a dearth of something else — a scarcity of whatever it is that information consumes: the attention of its recipients.”

1. No internet, no smartphones, no AI. Simon was just watching TV channels and newspapers multiply. But he already saw the core truth: attention is zero-sum. Give it to A, and B gets less. Period.

Now apply this logic to the AI era.

Before AI, my day had maybe 5-10 decision points: what tasks to do, what methods to use, how to prioritize. Now, with Claude Code, I might face 50: Should I let AI refactor this code? Is its output correct? Should I iterate again? Is this automation worth building? Who fixes the bugs later? Do I understand AI’s suggested approach? If not, how long will it take to understand?

Every new “possibility” is a new valve. Every valve asks: open or not?

There’s a famous Israeli study on parole judges that shows what this does to people. They found: approval rates start at roughly 65% at the beginning of each session, then gradually drop to near 0% as decisions accumulate. After a break, they reset to 65%. The cases didn’t get worse — the judges’ brains got tired.

David Meyer’s research is even more direct: multitasking increases completion time by more than double. The brain produces zero-progress “restart” time when switching between tasks. And only 2-5% of people are true “supertaskers” — the other 95% perform worse when multitasking.

That’s the trap. AI lets you open more valves, but the pipe hasn’t changed. Every new valve doesn’t add output — it just thins what each one gets.

III. The Data Doesn’t Lie

If it were just me, fine — maybe I’m doing it wrong. But the numbers say otherwise.

APA’s 2023 Work in America Survey (n=2,515):

• 38% of employees worry AI will make their jobs obsolete
• Among those worried: 51% say work hurts their mental health, 64% feel tense or stressed daily, 37% report emotional exhaustion
• Even among those NOT worried about AI replacement, stress is rising — 56% feel micromanaged by tech surveillance

Microsoft’s 2024 Work Trend Index:

• 75% of knowledge workers now use AI at work
• 78% bring their own AI tools (BYOAI)
• Yet 59% of leaders admit they can’t quantify AI’s productivity gains
• 60% of companies lack an AI vision or plan
• 46% of global workers are considering quitting within the year — an all-time high

Upwork’s 2024 Research (the most damning):

• Over 50% of AI-using employees report feeling “overwhelmed”
• Workers with AI tool access report more burnout, not less

Gallup 2021 Global Emotions Report:

• 41% of adults worldwide report stress — all-time high
• 42% report worry — all-time high

Back in 2007, Tarafdar named five flavors of technostress: overload (work faster), invasion (always on), complexity (steep learning curves), insecurity (will I be replaced?), uncertainty (this tool will be obsolete next quarter).

Sound familiar? AI cranks every single one up.

IV. History Repeats

This isn’t the first time.

In 1987, Nobel economist Robert Solow wrote a line that became known as the “Solow Paradox”:

“You can see the computer age everywhere but in the productivity statistics.”

Erik Brynjolfsson quantified this in 1993: computing power grew 100x in the 1970s-80s, yet labor productivity growth fell from over 3% in the 1960s to roughly 1%. More computers didn’t mean more productivity — at least not in the short term.

Same story with steam and electricity — the productivity payoff took decades. 1870s factories bolted electric motors onto steam-era floor plans and wondered why nothing changed. It wasn’t until someone rethought the entire factory layout that electricity actually delivered.

Today’s AI era is replaying this cycle:

Tool explosion → Cognitive overload → Organizational adaptation lag → Productivity stagnation

75% are using AI, but 59% of leaders can’t tell you what it’s actually doing for them. More tools, more exhaustion. The problem isn’t AI — it’s that our habits, organizations, and brains haven’t caught up.

V. Knowing When to Stop

So what do we do? Reject AI? Obviously unrealistic. Use AI harder? The data already shows that road leads to burnout.

Lao Tzu said:

“He who knows when to stop does not find himself in danger.” — Tao Te Ching, Chapter 44

And an even sharper one:

“Less brings gain; more brings confusion.” — Tao Te Ching, Chapter 22

Two and a half thousand years apart, same answer: do less, not more.

When Steve Jobs returned to Apple in 1997, the first thing he did wasn’t launch new products — it was killing 70% of the product line. He later said:

“People think focus means saying yes to the thing you’ve got to focus on. But that’s not what it means at all. It means saying no to the hundred other good ideas.”

Focus isn’t saying yes to the goal. Focus is saying no to a hundred other good ideas.

Obama wore the same gray or blue suit every day. Zuckerberg wore the same gray T-shirt. Not because they couldn’t afford variety — because they understood a fact: every trivial daily decision draws from the same pool of limited cognitive resources. Reducing unimportant decisions protects the quality of important ones.

My own recent realization: Claude Code can help me with dozens of things — note automation, code refactoring, knowledge curation, data analysis, documentation. But when I try to advance all these possibilities simultaneously, I’m more exhausted than before AI. Every new automated pipeline means one more system to maintain, one more potential failure point, one more “is this AI output correct?” judgment call.

The moments that actually work are when I decide not to do something. Not because I can’t — precisely because I can, which makes “not doing” an active decision that must be deliberately made.

Jacoby’s experiments showed the same thing: more brand information actually led to worse decisions. Barry Schwartz called it the “paradox of choice” — past a certain point, more options just mean more anxiety and more regret.

AI has handed us more options than we’ve ever had. But our ability to choose well? Still the same.

Closing

Let me return to Zhuangzi.

“To pursue the limitless with the limited is perilous” — but Zhuangzi didn’t say “so do nothing.” He was talking about nourishing life. Nourishing what? The pipe itself.

Here’s what I keep coming back to: the more you CAN do, the less you SHOULD do.

Not rejecting tools. Using the power of tools to do fewer, deeper things — not more, shallower things. The pipe diameter won’t change. The water pressure will keep rising. The only thing you can control is how many valves you choose to open.

Eric Schmidt once warned that too much information could get in the way of “deep thinking, comprehension, and memory formation.” Clay Shirky put it better: the problem isn’t information overload — it’s filter failure. Our filters can’t keep up.

AI generates information faster than anything before it. But the hard part was never generating more — it’s throwing most of it away and actually paying attention to what’s left.

Simon made it clear in 1971: in an information-rich world, the scarce resource isn’t information — it’s attention.

So in an AI-rich world, what’s scarce?

The moments you decide not to use AI.

Written April 8, 2026. An evening when I used AI to organize 300 note files — and found myself more exhausted than before I started.

Sources

• The Magical Number Seven, Plus or Minus Two — George A. Miller’s landmark 1956 paper on working memory limits
• Cognitive Load Theory — John Sweller’s framework on working memory bottlenecks
• Attention Restoration Theory — Kaplan & Kaplan on directed attention fatigue
• Upwork 2024 Research — AI and workforce overwhelm
• Microsoft 2024 Work Trend Index — AI adoption and productivity metrics
• APA 2023 Work in America Survey — AI anxiety and workplace mental health
• The Paradox of Choice — Barry Schwartz on choice overload
• Solow Paradox — The productivity paradox of new technology

The Spark

March 28, 2026, Portugal. World Superbike Championship, WorldSSP class.

A motorcycle numbered ZX820RR-RS crosses the finish line four seconds ahead of Yamaha. The next day, the same bike recovers from third to first and wins again. Back-to-back victories. The rider, Valentin Debise, is French — but the machine under him is entirely Chinese. Engine, brake discs, suspension, calipers — all from a Chongqing company called ZXMOTO, founded by a man named Zhang Xue.

This broke decades of monopoly by Ducati, Yamaha, and Kawasaki. Watching the livestream from Chongqing, Zhang Xue cried.

Most people will focus on the championship. What moved me was something Zhang Xue once said:

“Doing something not for the result but out of genuine love — the outcome might truly be different.”

That reminded me of a line attributed to Feng Ji, the creator of Black Myth: Wukong:

“Setting out on the journey to the West is more important than obtaining the scripture itself.”

This line has been spinning in my head. Because I’ve noticed — across eras and cultures, the people who ultimately built something real were all saying the same thing.

Zhang Xue: Ride First, Learn the Gap Later

In 2013, Zhang Xue arrived in Chongqing with 20,000 yuan. He assembled motorcycles, modified them, sold them online. In 2017 he founded Kaiyue Motorcycle, scaling from 800 to 30,000 units per year. In 2024 he left to start ZXMOTO, focused on high-performance bikes and proprietary engine R&D.

The gap between Chinese motorcycle manufacturing and Japan/Italy is real. The conventional approach: catch up on technology first, then go race. Zhang Xue did the opposite: race first, discover gaps on the track, improve through the gaps.

He said:

“Maybe I’m completely inferior to you in other ways — my talent isn’t as good, my resources aren’t as good — but when my effort exceeds yours tenfold, why shouldn’t this be mine? It IS mine!”

And:

“If you do it while young, even if you fail when you grow old, you won’t regret it — absolutely not.”

That Portugal championship didn’t fall from the sky. It was earned from 20,000 yuan, through countless last-place finishes, retirements, and doubts — lap by lap.

Jiaozi: Make It First, Talk Later

Jiaozi (Yang Yu) graduated from Sichuan University’s School of Pharmacy in 2002. He abandoned a conventional career to teach himself animation. His mother’s monthly pension of 1,000 yuan was his only income. Three and a half years later, alone, he produced the short film Beat Up the Big Watermelon, which won a special commendation at the Berlin International Short Film Festival in 2008.

He wrote a line for his film that became a rallying cry for all of Chinese animation:

“My fate is my own, not decided by the heavens.”

This isn’t just a line from Ne Zha — it’s Jiaozi’s own life. A pharmacy graduate with no animation training, no industry connections, surviving on his mother’s pension, questioned by everyone. His choice wasn’t to prove he was qualified to make animation — it was to just make it.

Ne Zha grossed 5 billion RMB in 2019. Ne Zha 2 hit $2.2 billion globally in 2025, becoming the highest-grossing animated film in history. But the starting point was one person, one computer, 1,000 yuan a month, and one decision: start.

The Ancients Already Knew

Lu Xun: There Were No Roads

“There were no roads to begin with. But when enough people walk the same way, a road appears.” — Lu Xun, My Old Home

Roads aren’t planned. They’re walked into existence. The first person to walk had no road — the road appeared because he walked.

Wang Yangming: Unity of Knowledge and Action

“Knowledge is the beginning of action; action is the completion of knowledge.” — Wang Yangming

Wang Yangming spent his life arguing one point: you think you “know” something, but if you haven’t done it, you know nothing. True knowledge comes only from action. Every step on the journey is the scripture.

Lao Tzu: A Thousand Miles

“A journey of a thousand miles begins with a single step.” — Lao Tzu, Tao Te Ching, Chapter 64

Perhaps the most quoted line ever. But most people see “a thousand miles” and miss the key word: “begins.” The point isn’t the distance. It’s the starting.

Li Bai: The Hard Road

“I want to cross the Yellow River, but ice blocks the channel; I want to climb Taihang, but snow fills the slopes… Yet someday I will ride the long wind and break the heavy waves, and set my cloudy sail straight across the deep, deep sea.”

When Li Bai wrote this, his career was blocked and every path seemed closed. But the poem doesn’t end in despair — it ends in “riding the wind and breaking the waves.” Knowing the road is hard and walking it anyway — that IS the answer.

The West Says the Same Thing

Theodore Roosevelt: The Man in the Arena

“The credit belongs to the man who is actually in the arena, whose face is marred by dust and sweat and blood… who at best knows, in the end, the triumph of high achievement, and who, at the worst, if he fails, at least fails while daring greatly.” — “Citizenship in a Republic,” 1910

Roosevelt and Zeng Guofan said the same thing: the spectator is never wrong — but never wins either.

Steve Jobs

“The journey is the reward.”

Jobs said this when Apple was nowhere near a trillion-dollar company. The point: don’t walk for the destination. The walking itself is the meaning.

C.P. Cavafy: Ithaca

As you set out for Ithaca, hope the voyage is a long one, full of adventure, full of discovery. Laistrygonians and Cyclops, angry Poseidon — don’t be afraid of them: you’ll never find things like that on your way as long as you keep your thoughts raised high, as long as a rare excitement stirs your spirit and your body.

Hope the voyage is a long one. May there be many a summer morning when, with what pleasure, what joy, you come into harbors seen for the first time; may you stop at Phoenician trading stations to buy fine things, mother of pearl and coral, amber and ebony, sensual perfume of every kind — as many sensual perfumes as you can; and may you visit many Egyptian cities to gather stores of knowledge from their scholars.

Keep Ithaca always in your mind. Arriving there is what you are destined for. But do not hurry the journey at all. Better if it lasts for years, so you are old by the time you reach the island, wealthy with all you have gained on the way, not expecting Ithaca to make you rich.

Ithaca gave you the marvelous journey. Without her you would not have set out. She has nothing left to give you now.

And if you find her poor, Ithaca won’t have fooled you. Wise as you will have become, so full of experience, you will have understood by then what these Ithacas mean.

— C.P. Cavafy, Ithaca (1911)

The poem says it all. Ithaca’s greatest value is that it made you set out. Everything you encounter along the way — storms, harbors, strangers — that’s what you actually receive. And if the destination turns out to be less than you imagined, it hasn’t fooled you. By then, you’ll understand what Ithaca really meant.

Reid Hoffman

“Starting a company is like throwing yourself off a cliff and assembling an airplane on the way down.”

“If you aren’t embarrassed by the first version of your product, you shipped too late.”

His first principle is speed. Nobody starts when they’re ready. Everyone learns to fly on the way down — and if that first version doesn’t make you cringe, you waited too long to jump.

Jensen Huang

At Stanford’s 2024 commencement, Jensen Huang told the graduates:

“I hope suffering happens to you.”

He wasn’t cursing them. NVIDIA nearly died in 1997 — one month of payroll left. He later said building NVIDIA was “a million times harder” than expected. If he’d known the pain ahead, he probably wouldn’t have started. But he started. And the suffering taught him everything he needed.

Rilke

“Be patient toward all that is unsolved in your heart and try to love the questions themselves. Live the questions now. Perhaps you will then gradually, without noticing it, live along some distant day into the answer.” — Rilke, Letters to a Young Poet (1903)

Rilke and Wang Yangming said the same thing: you don’t find the answer and then set out — you live your way into the answer along the road.

Why “Just Start” Is So Hard

If everyone knows this, why do most people still not move?

Because setting out on the journey has a hidden cost: you expose your imperfection.

If you don’t race, you can always say “my bike is still being tuned.” If you don’t ship, you can always say “I’m still polishing.” If you don’t try building with new tools, you can always say “the technology isn’t mature enough.”

Standing still, your self-image is intact — “I just haven’t started yet; once I do, I’ll be amazing.” But once you step out, reality holds up a mirror. You see the gaps, your own clumsiness, others who are faster, better, more experienced.

Zhang Xue saw that mirror — and kept riding. Jiaozi saw it — and kept drawing. Feng Ji saw it — and kept building.

This is what “personal cultivation” means — not a moral judgment, but the ability to face real feedback without running away.

The Journey Is the Scripture

In Journey to the West, Tang Seng and his disciples endure eighty-one trials to reach Vulture Peak and obtain the scriptures. But think carefully — what are the scriptures, really?

Are they the rolls of paper with words on them?

No. The real scripture is Sun Wukong learning restraint, Zhu Bajie learning perseverance, Sha Wujing learning responsibility, and Tang Seng learning trust. These things weren’t at Vulture Peak — they were on the road.

The eighty-one trials aren’t “obstacles before the scripture” — they are the scripture.

Closing

I’ve been rebuilding my workflow with AI recently — from how I organize knowledge to how I plan my day. Honestly, 90% of the time I don’t know what I’m doing. Tools change, models change, best practices get overturned every week.

But I think of Zhang Xue, Jiaozi, and Wang Yangming’s “knowledge is the beginning of action” — you can’t fully understand before doing. You can only understand through doing.

Lu Xun said roads are made by walking. Cavafy said Ithaca is just the reason to set out. Roosevelt said credit belongs to the man in the arena. Jobs said the journey is the reward.

They’re all saying the same thing:

Setting out on the journey is more important than reaching the destination.

Because every step on that journey is the scripture.

Written April 5, 2026. A week after Zhang Xue’s championship in Portugal.

The Seed

I wrote three lines in my notebook this morning. They came out fast — the way things do when you’ve been chewing on something for months without realizing it.

Don’t chase hype blindly. Don’t ignore trends willfully. Hold your ground, embrace change, and cherish everything you have and everyone you meet.

I work in AI infrastructure. Every morning I open my laptop to a new “revolution.” Vibe coding. AI agent frameworks. Yet another startup claiming it will change everything. Last week’s trending repo is already forgotten.

But at the same time, something fundamental is shifting. GPU compute is reshaping the entire computing infrastructure. The infrastructure work I do every day didn’t exist three years ago. Ignoring these changes is just as dangerous as chasing every hot take.

The question isn’t whether to pay attention. It’s what deserves attention — and what’s just noise wearing the costume of signal.

There’s no formula. But I’ve found three anchors.

I. Don’t Chase Hype

In January 2019, WeWork was worth $47 billion. Nine months later, it couldn’t even go public. The valuation crashed 78%, the CEO cashed out $700 million before being shown the door, and the company filed for bankruptcy in 2023.

In May 2022, Terra/Luna evaporated $45 billion in a single week. FTX followed. Celsius. Three Arrows Capital. The entire crypto world fell like dominoes. The people who jumped in weren’t those who truly understood blockchain — they were those who saw everyone else making money.

The pattern is always the same: a real innovation (shared workspaces, blockchain) gets inflated by capital and FOMO until the price tag has nothing to do with the underlying value. The people who get hurt aren’t the builders — they’re the followers.

Buffett talks about the “circle of competence”:

“The size of your circle doesn’t matter. Knowing its boundaries does.”

Chasing hype is stepping outside your circle of competence while believing you’re still inside it. You don’t understand the underlying logic, but you see prices rising, others profiting, feeds buzzing — so you follow. That’s not investing. It’s imitation.

Naval Ravikant puts it more directly: “Escape competition through authenticity.” The opposite of chasing hype isn’t doing nothing — it’s honestly asking yourself: Do I actually understand this, or am I just afraid of missing out?

The AI space is going through the same cycle. In Y Combinator’s Winter 2025 batch, 25% of startups had codebases that were 95% AI-generated. Sounds cool. But research shows AI-generated code has 2.74x the security vulnerability rate of human-written code, and senior developers actually became 19% slower when using AI tools. “Vibe coding” was coined by Karpathy, but his definition was “fully giving up on understanding the code itself” — that’s a warning, not a recommendation.

The SEC has already started fining companies for “AI washing” — slapping “AI-powered” on products with no meaningful AI inside. It’s the “.com” suffix of our era.

Laozi said: The highest good is like water. Water benefits all things and does not compete.

Water doesn’t rush to be where everyone else is. It finds the low places that others avoid. And precisely because of this — nothing is stronger.

Not competing isn’t not moving. It’s not being swept along by others’ movement.

II. Don’t Ignore Trends

In 1975, Kodak engineer Steven Sasson built the world’s first digital camera. 10,000 pixels.

He brought it to management. Management said: Nice, but our film business makes $16 billion a year. Drop it.

In 2012, Kodak filed for bankruptcy.

Kodak’s tragedy wasn’t a technology blind spot — they invented the technology. The tragedy was that they saw the trend and chose not to look.

Blockbuster’s story is even more absurd. In 2000, Netflix came knocking with a $50 million acquisition offer. Blockbuster laughed and said no. At the time, Blockbuster had over 9,000 stores and 84,000 employees. A decade later, it filed for bankruptcy. Netflix is now worth $300 billion.

A former Blockbuster executive later admitted: “Digital would have changed Blockbuster’s business… but it wasn’t its killer. That credit belongs to Blockbuster itself.”

What kills you isn’t change — it’s your avoidance of it.

Marcus Aurelius wrote in Meditations:

“The universe is change; our life is what our thoughts make it.”

He also said: “Observe constantly that all things take place by change, and accustom thyself to consider that the nature of the Universe loves nothing so much as to change the things which are.” This isn’t self-help — it’s practical philosophy written by a Roman emperor on battlefields and during plagues. Change isn’t an exception. Change is the default state.

Laozi used water to explain this: Nothing in the world is softer or weaker than water, yet for attacking the hard and strong, nothing surpasses it.

Water doesn’t resist change. Water is change itself. It doesn’t need to become a rock to fight a rock — it flows around, seeps in, and given enough time, the rock is split open.

Trends are like water. You may not like them, but you can’t pretend they don’t exist. GPU compute is changing the computing paradigm — this isn’t hype, this is the reality of my daily work. LLMs aren’t a passing fad — they’re rewriting the fundamental assumptions of software engineering.

The difference between a trend and a hype is durability. Hype disappears when the money runs out. A trend reshapes the landscape even after the excitement fades.

The way to tell them apart is simple: when the money pulls out, is it still there?

III. Hold Your Ground, Embrace Change, Cherish Everything

If “don’t chase hype” and “don’t ignore trends” were enough, everyone would be doing it. The hard part is the third line: how do you hold your ground and embrace change at the same time?

NVIDIA gave us an answer.

In 1993, three people started a company at a Denny’s restaurant, each putting in $200. Jensen Huang later said that if he had known how hard the road would be — “a million times harder” — he might not have started. In 1997, the company had only one month of cash flow left. His go-to line in internal talks was: “We are thirty days from going out of business.”

In 2006, NVIDIA released CUDA — a platform for general-purpose GPU computing. Nobody cared. Gaming companies didn’t care. The scientific computing community didn’t care. Wall Street didn’t care.

Huang didn’t chase mobile chips when that was hot. Didn’t pivot to social media. Didn’t rebrand as a crypto company when miners were buying every GPU on the planet. He stayed on GPU compute. For thirty years.

Then the AI wave came. Suddenly, the whole world realized only NVIDIA had the infrastructure. By October 2025, NVIDIA became the first company to exceed $5 trillion in market capitalization.

Sun Tzu’s Art of War captures this in four characters: 守正出奇 — defend the orthodox, attack with the extraordinary. The “orthodox” is holding your ground — your foundation doesn’t move. The “extraordinary” is embracing change — on top of that foundation, you constantly adapt and explore. NVIDIA’s “orthodox” was GPU computing. Its “extraordinary” was that every time a new wave came, it was already ready.

Zeng Guofan, the Qing dynasty statesman, put it more bluntly:

“Shouting opinions from the sidelines is useless. You must step into the arena, shoulder the responsibility, before success is even possible.”

守正出奇 isn’t armchair wisdom — it’s wisdom born from being in the arena. You have to be on the field before you can talk about defense and offense.

There’s an image from Buddhism that I keep coming back to. Dhammapada, verse 401:

“Like water on a lotus leaf, or a mustard seed on the point of a needle — he who does not cling to sensual pleasures, him do I call a holy man.”

The lotus leaf doesn’t reject the water. The water falls on the leaf, fully in contact. But the water doesn’t stick. The leaf doesn’t absorb. They meet, then each goes its own way.

Not clinging to hype, not clinging to the status quo, not clinging to outcomes. Fully in contact with the world — but not stuck to it.

My work puts me at a unique intersection: where AI meets infrastructure. Every day I see two forces collide — a genuine computing paradigm shift, and all the noise surrounding it. My anchor isn’t a technology or a trend. It’s something deeper: the conviction that value lives in the gaps between layers. Others see systems as stacks. I see the dark spaces between them — and I build until information flows through. This doesn’t expire with any hype cycle, because it’s not about any particular layer. It’s about the act of weaving connections itself.

Closing

Seneca wrote: “You should not copy the bad simply because they are many, nor should you hate the many because they are unlike you.”

Probably the best social media usage guide from 2,000 years ago.

Naval said: “Desire is a contract that you make with yourself to be unhappy until you get what you want.”

Chasing hype is a form of desire. Fear of trends is also a form of desire. Both keep you living in the future instead of the present.

Holding your ground isn’t stubbornness. Embracing change isn’t chasing wind. Cherishing what you have isn’t platitude — it’s the water on the lotus leaf: fully in contact with the world, bound by none of it.

Zeng Guofan also said:

“A troubled mind and anxious thoughts — these are precisely the forge that tempers heroes and refines you to completion.”

So don’t fear confusion. Feeling torn between hype and trends — that itself is the forging process. The key isn’t to eliminate the tension, but to find your anchor — and then, step into the arena.

The tension between hype and trends isn’t a problem to solve. It’s a forge. Find your anchor, step into the arena, and build.

Written on April 1, 2026 — a morning when, predictably, three new AI frameworks were announced.

Every new control layer shifts the human role from doing the work to designing the feedback loop — observe, compare, adjust, repeat. Same pattern since 1780.

A 250-Year Pattern

George (@odysseus0z) identified a pattern recurring three times across 250 years while reading OpenAI’s harness engineering post:

1. Watt’s centrifugal governor (1780s): worker shifts from manually adjusting the valve to designing the automatic governor mechanism.
2. Kubernetes: engineer shifts from restarting services to writing the declarative spec the system reconciles against.
3. AI agent harness: engineer shifts from writing code to designing environments, feedback loops, and constraints for agents to write code.

Norbert Wiener named this in 1948: cybernetics, from Greek kubernētēs (steersman). The core structure is always the same: observe actual state, compare with desired state, adjust, loop. Each time, humans shift from rower to steersman — not disappearing, but moving up one abstraction layer.

Why This Is Universal

This pattern transcends domains entirely: it applies to mechanical engineering, software infrastructure, AI workflows, management, and biological homeostasis. The pattern holds in any system where a feedback loop can replace direct manual control.

The Missing Link: Maltz (1960)

Maxwell Maltz is the missing link in the cybernetics chain. He applied Wiener’s 1948 framework to human psychology 60+ years before AI harness engineering — framing the brain as a goal-seeking servo-mechanism that auto-corrects via feedback.

The steersman role maps precisely: set the target (self-image/goal), trust the automatic process, don’t micromanage the servo-mechanism’s execution.

Maltz’s formulation predates Kubernetes by 54 years and AI harness engineering by 66 years, yet describes the same pattern.

The timeline:

• Wiener (1948) — theory
• Maltz (1960) — human psychology
• Kubernetes (2014) — infrastructure
• AI harness (2026) — agent engineering

The lesson: the steersman role isn’t new. We’ve been discovering it again and again across every domain. The question isn’t whether this shift will happen in your field — it’s whether you’ll recognize it when it does.

Sources:

• George (@odysseus0z) on X
• OpenAI: Harness Engineering

Store scripts and helper functions in skill folders so agents spend tokens on composition and decision-making, not reconstructing boilerplate.

The Insight

Anthropic found that “one of the most powerful tools you can give Claude is code.” When skills include executable scripts and function libraries, Claude doesn’t need to spend tokens generating data-fetching code from scratch — it calls existing functions and focuses on composing them to answer higher-level questions.

Concrete example: a data science skill includes helper functions to fetch data from an event source. When the user asks “What happened on Tuesday?”, Claude doesn’t write data-fetching code — it generates a script composing existing functions for filtering, aggregation, and analysis. Tokens spent on logic and judgment, not infrastructure.

The Technique

Identify repetitive operations in your workflows (API calls, data transforms, file operations), extract them into script libraries within the skill folder, and document them in the SKILL.md. The agent discovers and composes them instead of reinventing them each session.

This shifts agent effort from reconstruction to composition — a fundamentally higher-leverage use of tokens and context window.

CLI-Anything: The Logical Extreme

CLI-Anything (HKUDS, 21.5k GitHub stars) takes this to its logical extreme: a 7-phase automated pipeline that transforms any desktop software into an installable CLI package. pip install and the agent can control Blender, GIMP, OBS, etc. through structured JSON commands.

The agent doesn’t learn the software’s API or GUI; it calls a pre-built CLI library. Each generated CLI includes:

• A SKILL.md for auto-discovery
• A REPL for exploration
• Undo/redo for safe state management

This is the most literal implementation of “give agents libraries”: the library IS the CLI package.

The Principle

The pattern generalizes: anywhere you find yourself writing the same kind of instructions repeatedly for an agent, you should extract that into a callable library instead. Instructions are consumed once and forgotten. Libraries persist across sessions and compose with each other.

Sources:

• Thariq on X — How We Use Skills
• CLI-Anything (GitHub)

Simon Willison (Django co-creator, datasette author) distills his hands-on patterns for using AI coding agents like Claude Code and OpenAI Codex. The guide covers five areas: principles, testing & quality, understanding code, annotated prompts, and appendices. See also: HN discussion.

I. Principles

1. Writing Code Is Cheap Now

AI agents fundamentally change software economics — writing code is now nearly free, but writing good code remains expensive. Good code means: correct, tested, documented, solving the right problem, handling edge cases, simple enough, maintainable, and meeting non-functional requirements (security, reliability, accessibility).

The paradigm shift: past engineering decisions — both micro (should I refactor? should I write a test?) and macro (project planning, feature evaluation) — were all based on “writing code is expensive.” Now you can spin up multiple agent sessions in parallel to explore approaches. “The worst outcome is learning it doesn’t work ten minutes later.”

The challenge: the industry needs new habits and practices to harness agent capabilities while maintaining strict quality standards.

2. Hoard Things You Know How to Do

“Knowing something is theoretically possible” and “having seen it run with your own eyes” are completely different. Build a collection of working examples — blog posts, GitHub repos (Willison has 1000+), tool sites, research repos — each becomes a reusable building block.

The most powerful pattern: have agents recombine multiple working codebases. His OCR tool combined Tesseract.js (OCR) and PDF.js (PDF rendering) — two existing JS libraries merged into one unified solution. Document once, agents deploy repeatedly.

II. Testing & Quality

3. Red/Green TDD

TDD is critical for AI agents because agents risk generating non-functional code or adding unwanted features. TDD mitigates this via automatic verification:

• Red phase: Write failing tests first — proving the test actually checks new functionality
• Green phase: Implement code to make all tests pass

Skipping the red phase means your test might have passed from the start — rendering it useless. Practical prompt: "Build a Python function to extract headers from a markdown string. Use red/green TDD."

4. First Run the Tests

The four-word mantra: start every session with “First run the tests.” The chain reaction:

1. Forces the agent to find and run tests — establishing test awareness
2. Reveals project scale and complexity through test count
3. Sets a test-centric mindset for the session
4. Activates the model’s built-in quality discipline

“If code has never been executed, it working in production is pure luck.”

III. Understanding Code

5. Linear Walkthroughs

When you’ve vibe-coded a project or inherited an unfamiliar codebase, have the agent read the source and generate a structured file-by-file walkthrough document. Even a 40-minute vibe-coded toy project becomes a learning opportunity after a walkthrough. “AI making code faster to write” doesn’t mean “you don’t need to understand the code.”

6. Interactive Explanations

The problem: cognitive debt. Agent-generated code you don’t understand becomes a black box blocking future development. The solution: Have the agent create interactive visualizations to explain algorithms.

Willison had Claude Code generate an interactive HTML page to step through a word-cloud placement algorithm (Archimedean spiral + random angle offset) — with pause, speed control, and frame-by-frame playback. “A good coding agent can generate these on demand to explain code — whether it wrote it or someone else did.”

IV. Annotated Prompts

A complete prompt case study (sent from an iPhone):

“gif-optimizer.html Compile gifsicle to WASM, then build a web page that lets you open or drag-drop an animated GIF onto it…”

Key techniques:

• Filename first — gif-optimizer.html as the opening word tells the agent the output format and project structure
• Leverage agent knowledge — “Compile gifsicle to WASM” relies on the agent knowing gifsicle (a 30-year-old compression tool) and the Emscripten toolchain
• Integrate testing — the prompt includes "Run 'uvx rodney --help'" to ensure the agent tests during development
• Iterate — follow-up requests to move build scripts to subdirectories, commit WASM bundles, add attribution

V. Appendix: Custom Instructions

Willison’s Artifacts custom instructions: always vanilla HTML + JS (no React), CSS with 2-space indent + box-sizing: border-box, 16px input fonts + Helvetica, JS with module syntax + 2-space indent.

For proofreading: check spelling, grammar, repetition, logic errors, factual inaccuracies, weak arguments, bad links. Importantly, he never lets LLMs ghostwrite opinion pieces or first-person blog posts — only proofread and update technical docs.

Takeaways

1. “First run the tests” should become the opening line of every Claude Code session, especially when modifying existing code.
2. Hoard working examples. Your Second Brain can serve as a “solution library” — document each solved problem, and future agents can recombine them.
3. Interactive explanations. For complex logic (data pipelines, algorithms), have agents generate visualizations to build understanding.
4. Cheap code != cheap good code. Agents generate fast, but verification, testing, and architecture remain human responsibilities.
5. Good prompts aren’t long essays — they’re precise anchors: filename + test command + existing examples.

The highest-leverage design work is choosing which constraints to impose — they determine what the system discovers on its own.

The Breakthrough

When building a topology visualization for a data analytics platform, the initial approach was hardcoding known relationships — manually listing which tables feed which functions. This was incomplete within days. The breakthrough was switching to constraint-based design: define parsing rules (regex patterns, parameter defaults) and let the system discover relationships automatically. It found 2-3x more connections than manual effort ever could.

This generalizes far beyond code. Designing good constraints — in systems, processes, organizations, or even personal workflows — is the meta-skill. A well-chosen constraint doesn’t limit; it channels energy toward emergent behavior you couldn’t have predicted or enumerated.

System Convergence in the AI Era

AI-Ghost-Lab’s “system convergence” concept is constraint design applied to AI-era development. His earlier mistake: relying on broad test coverage to contain AI-generated chaos. An AI agent might take three different paths from A to B, tripling the test burden. The fix: funnel similar functionality through unified entry points — control the core nodes, everything else can flex.

The specific constraints that work:

• Clear module boundaries
• Centralized state management
• Unified extension points
• Unified error handling

His metaphor captures it perfectly: “AI can make a mess, but only where you allow it — not in the living room.”

Constraining Agents, Not Railroading Them

Anthropic’s skill-writing advice embodies constraint design for AI agents. Two key principles:

1. “Don’t state the obvious” — Don’t constrain what the model already does well. Focus constraints on where its defaults are wrong.
2. “Avoid railroading” — Overly specific instructions limit adaptability.

The art is constraining where the model goes wrong while leaving freedom where it goes right. The frontend-design skill is a perfect example: it doesn’t teach CSS (Claude knows CSS). It specifically corrects Claude’s aesthetic defaults — the tendency toward Inter font and purple gradients. That surgical constraint produces better results than a wall of instructions ever could.

The Principle

The best designers — of systems, organizations, or AI workflows — don’t enumerate every behavior they want. They choose a small number of constraints that make the right behaviors emerge naturally. The meta-skill isn’t building; it’s knowing where to put the walls.

Your prefrontal cortex (PFC) is the brain’s CEO — it handles planning, decision-making, focus, impulse control, and emotional regulation. It’s also the last region to mature (around age 25) and one of the most vulnerable to damage. Here’s what the research says about what hurts it and what protects it.

What Damages It

Chronic stress is the biggest threat. Sustained cortisol exposure literally shrinks the dendritic spines in your PFC, weakening connections and shifting control from your rational brain to your emotional brain (Arnsten, 2009, Nature Reviews Neuroscience).

Sleep deprivation hits hard. Even one bad night drops PFC-dependent attention and decision-making by 20-30% (Lim & Dinges, 2010). Long-term, poor sleep accelerates beta-amyloid buildup — the protein linked to neurodegeneration.

Sedentary lifestyle starves the PFC. Without exercise, your brain produces less BDNF (brain-derived neurotrophic factor), the molecule that drives neuroplasticity and keeps neurons healthy.

Other risk factors: ultra-processed diets (inflammation + oxidative stress), chronic multitasking (depletes PFC’s finite cognitive resources), social isolation (the PFC’s social cognition circuits atrophy without use), and substance abuse.

What Protects It

The research points to two interventions that stand above the rest:

Exercise — 150 min/week of moderate aerobic activity. It increases BDNF, improves cerebral blood flow to the PFC, lowers baseline cortisol, and optimizes dopamine and norepinephrine levels. A 6-month aerobic training program significantly increased PFC gray and white matter volume in older adults (Colcombe et al., 2006).

Sleep — 7-9 hours, consistent schedule. During deep sleep, the glymphatic system becomes 10-20x more active, flushing metabolic waste from the PFC. Slow-wave sleep is when PFC-hippocampus memory consolidation happens.

Meditation rounds out the top tier. 8 weeks of mindfulness practice increased PFC gray matter density (Hölzel et al., 2011). The mechanism: it strengthens the PFC’s top-down control over the amygdala.

Other evidence-backed strategies: Mediterranean-style anti-inflammatory diet (omega-3, polyphenols, magnesium), meaningful social interaction, and learning new skills (instruments, languages, strategic games). Emerging evidence supports intermittent fasting and cold exposure, though human data is still limited.

The Three Pillars

PFC function rests on three foundations:

1. Neurotransmitter balance — Dopamine and norepinephrine follow an inverted-U curve. Too little = unfocused. Too much = anxious. Exercise, sleep, and meditation keep them in the sweet spot.
2. Synaptic plasticity — BDNF is the key driver. Exercise boosts it more than any other intervention. Chronic stress and inflammation suppress it.
3. Vascular health — PFC blood flow correlates directly with cognitive function. Aerobic fitness, omega-3, and polyphenols improve it. Hypertension and diabetes damage it.

The good news: most negative changes are reversible with intervention, typically within weeks to months.

Daily Checklist

Non-negotiable: Sleep 7+ hours. Move 30+ minutes. Avoid junk food.

High-impact additions: 10-20 min meditation. Omega-3 intake. Stress buffer (nature walk, journaling, deep breathing).

Optimize further: Learn something new. Single-task instead of multitask. Try intermittent fasting if it suits you.

Avoid at all costs: Chronic sleep debt (<6h/night for a week+). Frequent heavy drinking. Sustained high stress with no recovery. Prolonged sedentary lifestyle.

A Personal Take: AI and Context Switching

One thing that stood out to me: context switching is one of the most expensive operations your PFC performs. Every time you shift focus — between tabs, between tasks, between a thought and a notification — your prefrontal cortex pays an energy tax. Ophir et al. (2009) showed that heavy multitaskers actually get worse at filtering, not better.

This has implications for how we work with AI. The typical pattern — rapid back-and-forth prompting, constant micro-decisions about what to ask next, switching between your own thinking and the AI’s output — is essentially high-frequency context switching. It feels productive, but it’s draining the very part of your brain that’s supposed to be steering.

An alternative: treat your AI interactions more like a message queue. Batch your intent. Write out what you need, hand it off, and step away. Let the AI do its work asynchronously instead of co-piloting every keystroke. You stay in your own flow; the AI stays in its. When you come back, you review the output with a fresh PFC instead of one that’s been ping-ponging for an hour.

Less interactive doesn’t mean less effective. It might mean your prefrontal cortex still has something left in the tank by the end of the day.

Key references: Arnsten (2009), Colcombe et al. (2006), Hölzel et al. (2011), Mattson et al. (2018), Erickson et al. (2011), Lim & Dinges (2010), Jacka et al. (2017)

There’s a kind of curiosity you have when you’re young — the kind where you stay up late not because you have to, but because you can’t stop. Everything is a puzzle. Every new thing connects to three more things you want to try.

At some point, without noticing, I drifted away from that. Life got busy. I got good at executing, at being efficient, at knowing what to focus on. And those are fine qualities. But efficiency and curiosity pull in different directions — one narrows, the other wanders. I chose narrowing for a long time.

Then AI happened. Not the buzzword version — the version where you sit down with a tool and realize it fundamentally changes what one person can build. I started tinkering. Small experiments at first. Then staying up late again. Waking up with ideas. Building things on weekends — not for a deadline, not for anyone else, but because I genuinely wanted to see what would happen.

It took me a while to recognize the feeling. It was the same one I had as a kid learning to code. The same one from every time life dropped me into a new environment as an underdog — no reputation, no playbook, just raw curiosity and a willingness to figure things out.

What I think happened is that AI leveled the playing field in a way that gave me permission to be a beginner again. When the tools are this new, nobody has ten years of experience. The people who move fastest are the ones most willing to experiment and fail. That’s the underdog mindset — and I’ve been there before. Every time, I found my way. Not by being the smartest in the room, but by being the most willing to stay in the room.

I’m choosing that again. Not because I have to. Because I remember how alive it feels.

If you’ve been running on autopilot for a while — executing but not exploring, optimizing but not wondering — I’d encourage you to find something that makes you feel like a beginner again. The discomfort of not knowing is where curiosity lives.

I drifted away from it once. I don’t plan to again.