deepseek finished the work faster. for my workflow — figure out how something should work, give the bot a list of instructions how the software should work and how to create it — deepseek gave better results, faster. claude sonnet could finish too, but it used more tokens and the result was usually more buggy.

and i can hear you screaming in the comments — why sonnet and not opus or even fable? the reason is simple, token budget. but my point stands: the smaller model, at least from my day to day work, outperforms the bigger one. but only when im providing the instructions, i think. this is just how i work — i have a vision how software should work and i want my agent to execute it.

lets get technical

and the main deal for me is that this model can run locally on consumer hardware. im not saying you should cancel your claude sub. this works especially for my workflow.

Real world tests

so lets rewrite my old software — twitch viewer alert. it works like this: it watches the viewer count, and when the number changes, it fires a notification. theres some settings too, but i dont remember much more. the task for the agents is to convert this into a chrome extension. originally it was just a script you paste into the browser console — dont judge me, i was 19 when i wrote it.

each model was tested once in autonomous mode. the metrics are: token usage, time, and how many times i had to step in to keep things moving.

Prompt

hi i have this older project https://github.com/wallach-game/twitch-viewer-alerter and i wanna convert it into chrome addon, btw do not push to this repo, keep it as local cloned repo

DeepSeek V4 Flash

time: 4 minutes
tokens: 46,796
cost: $0.000004
interventions: 0
notes: added 7 new features and an icon

features added:

• viewer count tracking: every 5s → every 3s
• alert box: replaced with browser notifications
• notifications with buttons (open stream / mute for 30m)
• on/off switches in popup
• sound alert
• cooldown (10s, configurable)
• spa navigation support via mutationobserver
• offline stream detection
• viewer count badge on toolbar
• robust selectors with 3 strategies and fallbacks
• purple play triangle icon

ui:

notification working:

Sonnet 5

time: 2 minutes
tokens: ~562,910 (mostly cache)
cost: $0.40
interventions: 0
notes: the ui has less stuff than the original extension

verdict: imo it just yanked the old code and removed unnecessary stuff. but it figured out that browser notifications are not possible anymore — good catch.

ui:

notification:

Opus 4.8

time: 1:37
tokens: ~148,500
cost: $0.30
interventions: 0
notes: this version doesnt have a settings ui. kinda odd. and the notification doesnt even work — it created some localstorage nonsense and literally told me at the end it might not work. also, it never figured out that browser notifications dont work without https in chrome extensions.

Verdict

so heres the thing. deepseek did it in 4 minutes, added 11 new features, and cost basically nothing. sonnet did it faster (2 minutes) but the output was worse — stripped features, yanked old code. opus was the fastest at 1:37 but the result literally didnt work.

the numbers speak for themselves. deepseek v4 flash cost $0.000004, sonnet cost $0.40, opus cost $0.30. for my workflow — give clear instructions and let the model execute — the cheap model wins.

but in the end sonnet did what the prompt said — convert into an extension. and it did it. next time ill try a better prompt, something like “copy the original behavior first, then add new features and polish”.

am i saying you should cancel your sub? no. this is what works for me. but if you work like i do — plan first, then let the ai build it — give deepseek v4 flash a shot. you might be surprised.

and also deepseek screwed me over a few times, especially paired with opencodes auto exec command. its idea of fixing a running server app was deleting it without backup. but honestly, this can be blamed on my prompting.

Final words

im aware deepseek pricing is probably not sustainable at this level, but imo the same can be said about claude subscription. but i dont wanna get political here. the fact is that deepseek outperformed claude with a model that can be run locally on consumer hardware. thats the part that matters.

for f in *.txt; do
  wc -l "$f"
done

Or this?

find . -name '*.sh' -exec wc -l {} +

Or maybe you pipe into xargs and pray your filenames don’t have spaces:

find . -name '*.log' | xargs rm

These all work, but each has its own syntax, its own flags, its own quirks. I wanted something simpler — one consistent way to iterate over anything.

Enter bashumerate

enumerate -f '*.sh' -- 'wc -l {}'

That’s it. -f for files, {} is the placeholder for each item, -- separates source from command.

It supports four source types:

If you omit the command, items are printed to stdout — pipe them anywhere:

enumerate -f '*.log' | xargs rm
cat urls.txt | enumerate -l - | while read url; do curl -s "$url"; done

Why not just use a for loop?

You can. But enumerate gives you:

• Consistent syntax — same {} placeholder for files, lines, ranges, or lists
• Template mode — single-quoted commands work as shell templates: enumerate -f '*' -- 'cat {} | head -5'
• Filters — --include and --exclude with glob patterns
• NUL-safe — -0 flag for NUL-delimited output
• Extensible — drop a file in lib/enumerators/ to add custom sources

Adding your own source

Create lib/enumerators/docker.sh:

enumerate_source_docker() {
  docker ps --format '' | while read -r name; do
    printf '%s\0' "$name"
  done
}

Now you can do:

enumerate docker -- 'docker restart {}'

That’s it. No registration, no config — just name the function enumerate_source_<name> and it’s available.

Under the hood

The core is about 140 lines of bash. Sources are NUL-delimited internally (safe for spaces and special chars), and the {} placeholder is replaced at the shell level — no sed, no eval tricks.

enumerate -f '*.md' -- 'wc -l {}'
# → wc -l README.md
# → wc -l docs/guide.md

Install

basher install wallach-game/bashumerate

# or just clone and symlink
git clone https://github.com/wallach-game/bashumerate.git
ln -s "$PWD/bashumerate/bin/enumerate" ~/.local/bin/

The name

“bash” + “enumerate” — a bash enumerator. Not creative, but it does what it says.

Check it out: github.com/wallach-game/bashumerate

Acronyms used:
ACA → Azure Container App
ACI → Azure Container Instances

When I got tasked with dockerizing a service, I thought, “Easy. Done it before.” A couple of weeks later, I had a functional, tested container, after porting a dusty .NET 4.3 app. No explosions. Great start.

The team got together to test the container talking to the VM over the internet. It wasn’t smooth, but after some light “encouragement” (aka debugging), they started communicating. We jotted down a to-do list, grabbed some coffee, sent something vaguely optimistic to the CTO, and went home.

Then reality hit: How do I make the VM talk to the container over an internal network?

Private Networking: Where Hope Goes to Die

The service was deployed as Azure Container Apps (ACA). Switching the ACA environment to private networking slapped me with the ominous label (Preview). What could go wrong? Turns out, everything.

Enabling private networking auto-created a new resource group that I wasn’t part of. Quick message to the sysadmin and I can continue.

After some tweaks, I got a faint glimmer of life with tcpping on the service port. Curl? Dead. Tried changing ports. Now even tcpping gave up on me. Switched them back. Still nothing.

At this point, I was asking existential questions.

Googling for Sanity

YouTube to the rescue? Nope. The first tutorial I found was two years old. Eventually, I stumbled upon an official Microsoft guide, that is 1 month old. Hallelujah! Except the guide was incomplete, so I had to cross-reference the CLI version to make it “work”.

I finally got the container’s internal URL automatically registered in a private DNS zone. Success? Nope. The service still didn’t work. Oh, and the container started crashing because following Microsoft’s official steps broke its communication with Key Vault. Key Vault! The thing the entire service depends on. Lovely.

Static IPs and Moving Targets

Desperate, I tried Azure Container Instances (ACI). They have very little feature set, no automatic updates, http-scaling etc. These get an IP from the subnet. Sounds perfect, right? Wrong. Static IPs? Forget it. The IP changes every time the container restarts. Because, of course, nothing screams “enterprise-ready” like random IP roulette.

Here’s the best part: you need the container’s local URL before the service launches. And the only way to get it? Launch the container, then query it using the Azure CLI. Sure, that makes sense.

So, I wrote a quick bash script with jq to scrape the IP and auto-update my config. Was it clever? Yes. Did it feel like duct-taping my solution together? Also yes.

Final Thoughts

As I sit here, beer in hand, reflecting on the past few days, here’s my verdict:

Yes, it’s technically possible to make this work. But honestly? It’s not worth it. Save your time, your sanity, and your liver. Either go full-on containers or use Dapr. Anything else is asking for trouble.

Cheers to surviving the cloud! 🍻

Zkratky použité:
ACA → Azure Container App
ACI → Azure Container Instances

Když mi zadali dockerizovat službu, řekl jsem si: „Snadné. Už jsem to dělal.” O pár týdnů později jsem měl funkční a otestovaný kontejner po portaci staré aplikace .NET 4.3. Žádné exploze. Skvělý začátek.

Tým se sešel, aby otestoval komunikaci kontejneru s VM přes internet. Nebylo to hladké, ale po lehkém „popostrčení” (tj. debugování) začali komunikovat. Zapsali jsme si seznam úkolů, dali si kávu, poslali něco vlažně optimistického CTO a šli domů.

Pak přišla realita: Jak přimět VM, aby komunikovalo s kontejnerem přes interní síť?

Privátní komunikace: Peklo na zemi

Služba byla nasazena jako Azure Container Apps (ACA). Přepnutí prostředí ACA na privátní síťování mi zobrazilo label (Preview). Co by se mohlo pokazit? Ukázalo se, že všechno.

Povolení privátního síťování automaticky vytvořilo novou skupinu zdrojů, ke které jsem neměl přístup. Rychlá zpráva správci systému a mohl jsem pokračovat.

Po pár úpravách jsem viděl slabou jiskru života s tcpping na portu služby. Curl? Mrtvý. Změnil jsem porty. Teď už ani tcpping nefungoval. Vrátil jsem je zpět. Už nic.

V tu chvíli jsem si kladl existenciální otázky.

Googlení pro zachování rozumu

YouTube na záchranu? Ne. Jediný tutoriál, který jsem našel, byl dva roky starý. Nakonec jsem narazil na oficiální Microsoft návod, který byl měsíc starý. Haleluja! Až na to, že byl neúplný, takže jsem musel cross-referencovat CLI verzi, aby to začalo „fungovat”.

Nakonec se automaticky zaregistrovala interní URL kontejneru v privátní DNS zóně. Úspěch? Ne. Služba stále nefungovala. A kontejner začal padat, protože postupování podle oficiálních kroků Microsoftu rozbilo komunikaci s Key Vaultem. Key Vault! Základ celé služby. Skvělé.

“Statické” IP adresy a ACI

Zoufalý, vyzkoušel jsem Azure Container Instances (ACI). Ty mají velmi málo funkcí, žádné automatické aktualizace, škálování dle požadavků atd. Tyto dostávají IP ze subnetu. Perfektní, ne? NE. Statické IP? Zapomeňte. IP se mění při každém restartu kontejneru.

Nejlepší část? Potřebuji lokální URL/IP kontejneru před spuštěním služby. A jediný způsob, jak ho získat? Spustit kontejner a poté se zeptat Azure CLI na jeho IP. Skvělé.

Tak jsem napsal rychlý bash skript s jq, který sebere IP a automaticky aktualizuje konfigurační soubor. Bylo to chytré? Ano. Připadal jsem si, jako bych svůj projekt obalil lepicí páskou? Taky ano.

Závěrečné myšlenky

Jak tu sedím s pivem v ruce a přemýšlím o posledních dnech, zde je můj verdikt:

Ano, technicky je možné to zprovoznit. Ale upřímně? Nestojí to za to. Ušetřete si čas, rozum a játra. Buď přejděte kompletně na kontejnery, nebo použijte Dapr. Cokoliv jiného dopadne takto.

Na zdraví pro přežití cloudu! 🍻