Here's what I've been working on nearly 3 years. This is a deterministic coding and refactoring engine. What coding should be, and the backbone AI needs to become provably safe. You can learn more about what it unlocks (and read things in dark mode) at evok.dev

Same task, same model. All successful. LucenaCoder spent 62k tokens and 6.5 cents. Pi spent 110k tokens and 5.7 cents. OpenCode spent 216k and 8.7 cents. Kilo Code spent 316k and 12.9 cents. Copilot spent 393k tokens at and 13.3 cents. Codex spent 430k tokens and 14.5 cents. Architecture matters. #TokenEfficiencyBench

Ooh baby. We're moving now... first benchmark results for token-efficiency. Lowest token spend by far. Notice the caching difference? Less caching, but fewer tokens through more intelligent architecture. lucena.one/blog/almedra-agen…

lol $10 per million INPUT. Only 2x as expensive. Those tokenmaxxing suckers are gonna be hurting...

Some new themes we've added in prep for our desktop app...

We added 2 new features to the totally free LucenaCoder.com agentic coding harness. 1) DoubleCheck. This is part of our new V2 convo threading architecture and is an optional setting to have a second agent who is briefed on your task double check the first agent's work. Turn it on and off as desired (Off by default to keep max token-efficiency) 2) Audio playback. Activate it in the header, and you can highlight text right click in the most recent response and have it read aloud, or choose to have it auto-read messages as they arrive. You can select from OpenRouter voice models, but fair warning, they aren't cheap! Off by default. Welcome to all our new users in the last 2 days especially... we must give special thanks #Copilot for raising your rates, we appreciate your outflow! Evōk x LucenaCoder implementations should be active for Codebase.Observer beta users next week. Your agents will automatically receive your architectural blueprints, further reducing token spend. This is the one I'm most excited about, because in early testing it's like having an oracle that just gets it while using cheaper models.

rofl, guy told me most of his traffic comes from the midwest. He's huge in Council Bluffs.

In 2025 I had to remove Deepseek from even being AVAILABLE in a product for 2 specific small corporate buyers. In 2026, I'm being asked what Chinese models it supports, and all the security questions of last year have vanished around "is our code persisted?"

Stop being retarded, start Retardmaxing Being retarded means you’re ruminating and overthinking your decision Retardmaxing means you’re making decisions based on your gut instinct — zero rumination, zero regrets Stop being retarded, start #Retardmaxing

I want to scream when someone calls brittle string parsing deterministic. The best is when it's keyword guessing based on English. Wow. What a system.

Can anyone give me a warm intro to the guy who just blasted this to the entire public timeline?

If you want to know the difference between a junior dev and a senior dev, you can get it from this entire interaction with AI: "You’re right to stop me here. I patched the wording around a broken invariant instead of eliminating the invariant breach. I’m going to read the local architectural standard and then reframe this as “the agent cannot enter an invalid-anchor state,” not “the agent gets a nicer error.” Junior devs code defensively. Senior devs code in ways that make invalid states impossible to ever exist.

I'm gonna give up on my main round to raise a smaller one and just fix social messaging before I go nuts. Communicating thru walled platforms is the worst. We need monogram, and monogram only needs a little love back to break this dependency. Missed a DM from someone I actually wanted to hear from because their newish account gets trapped in Hidden. They think they're being ignored. I didn't even know. Both of us waste the only asset that matters as a result, time. I can fix this.

I do not respect tokenmaxxing because the metric doesn’t make sense to me. Like imagine uber drivers getting judged based on how much fuel they use.

Tokenmaxxing is for suckers.

Tokenmaxxing is like being proud of your truck that gets 1 mile per gallon on premium.

Types are sugar. Intent. Not reality. They bloat tokens and help nothing. The future is typeless because you don't need to declare, reality declares for you. It works or it doesn't. Binary outcomes where codebases exist in glass boxes touched only by math. Types are a trope.

white noise is gaussian blur for the ears

Love seeing this. Like if people just decided to split $10 over a 5 OSS projects per month it really matters to those devs.

I saw this in my feed and thought we were gonna get to see a codebase... if (seat.recline()) { throw new CriticalError("Feature disabled."); } // TODO from corporate: Monetize this