Picture this: 2 a.m., cold pizza, unresolved merge conflicts. You’re a junior dev trying to ship a single “hello world” API.
You run:
npx create-express-app-
npm installa logging lib - an ORM “for later”
- some rate-limiting middleware “just in case”
…and boom — you’re live.
Six months later that same endpoint is a 4 GB container that costs more per month than your rent. Your CTO opens the cloud bill and quietly reconsiders his life choices.
This isn’t a story about one dumb decision.
This is the entire autobiography of modern software:
Death by 1,000 Defaults.
The Bloat That Ate Silicon Valley
We never voted for 800 dependencies in a todo app.
We never signed up to ship the entire build toolchain to production.
It just… happened. One harmless default at a time:
- Grab the official Node image → 900 MB of Ubuntu you’ll never use
-
npm install lodashfor one line of code → 127 transitive deps - Split into microservices “so teams can move fast” → 400 ms added latency and a $40k/month network bill
- “Log everything, we’ll filter later” → Datadog becomes your biggest line item
- Autoscale with the “safe” instance type → 3× overspend forever
Each choice is locally rational.
Together they’re a horror movie in slow motion.
Why Defaults Win (And Why Cloud Providers Send Quiet Thank-You Cards)
Defaults are the fast food of engineering: engineered for instant gratification, optimized for addiction.
They prioritize speed, safety, and zero-thought adoption — never long-term leanness.
And the cloud vendors? They don’t have to push bloat.
They just sit back and let your unexamined defaults turn into their recurring revenue.
Every idle gigabyte is a margarita on some account manager’s yacht.
The Rebels Who Refused to Play Along
There are still a few projects that looked at the default parade and said:
“Hard pass.”
TigerBeetle
TigerBeetle is the loudest middle finger to bloat I’ve seen in years.
A financial transaction database that:
- Runs at single-digit megabytes
- Handles millions of transactions per second on a single box
- Was designed from day one for correctness and speed
- Ships a binary so small you can email it as an attachment (and people actually do)
They didn’t win by being 10× smarter.
They won by making a cultural decision most companies forgot existed:
We will not accept bloat as inevitable.
Stle
Stle is the new kid doing the same thing at the language level.
A Zig-inspired systems language that’s basically:
“C, but without the 50 years of footguns and without the 800-page standard.”
Tiny compiler. Tiny runtime. Zero-cost abstractions.
Still early — but the vibe is unmistakable:
Taste > convenience.
Both projects expose the uncomfortable truth:
We didn’t lose performance because the problems got harder.
We lost it because we stopped fighting the defaults.
The Seven Deadly Defaults (Your Post-Mortem Lineup)
Image Armageddon
Shipping dev tools to prod because multi-stage builds are “hard”Dependency Diarrhea
200 packages to parse JSONConcurrency Chaos
Unbounded goroutines/threads per requestMicroservice Mayhem
Network tax disguised as architectureLogging Landfill
“Verbose by default” never gets turned downScaling Sloth
“Just use m5.xlarge, it’s fine”
(narrator: it wasn’t)Temporary Hacks Forever
That debug endpoint still running in prod two years later
Individually forgivable.
Collectively fatal.
The Real Cost Isn’t Money — It’s Entropy
Sure, the bill hurts.
But the deeper wound is the fog:
- onboarding takes months
- deploys are scary
- nobody knows why latency spiked last Thursday
- your best engineers spend half their week doing digital archaeology
Fat stacks don’t just cost dollars.
They cost clarity.
And clarity is the actual throughput limiter in 99% of companies.
Your No-Zealot-Required Escape Kit
You don’t need to rewrite everything in Zig tomorrow.
Just recover your taste:
1) Make efficiency a build breaker
- Container > 100 MB? CI fails.
- RSS > 512 MB per request? CI fails.
No heroics. Just guardrails.
2) Start stupidly small, scale only when metrics scream
t3.micro first.
m6g.16xlarge never.
(unless you can prove it)
3) Bound literally everything
Max workers.
Max fan-out.
Max queue depth.
Max log volume.
Unbounded systems metastasize.
4) Fewer, sharper services
Microservices are a tool, not a personality trait.
5) Track cost like a product metric
Cost per endpoint.
Per customer segment.
Per feature flag.
If you can’t attribute it, you can’t fix it.
Epilogue: From Bloat to Boss-Level
Death by 1,000 defaults is real.
But so is the antidote:
Stop sleepwalking through your stack.
Go look at what TigerBeetle and Stle are doing.
Steal their attitude, if not their entire codebase.
Notice the “normal” that crept in. Name the compound cost. Re-choose your tools with intention.
Because the opposite of a fat stack isn’t hair-shirt minimalism.
It’s engineering on purpose.
Your move, builder.
What’s the first default you’re going to murder this week?
(And if you need inspiration, go stare at https://tigerbeetle.com for five minutes. Warning: may cause sudden urge to delete node_modules.)
Top comments (0)