Tokenmaxxing is for suckers. Try @LucenaCoder, a free, token-efficiency obsessed agentic coding harness. No login req, just add a key! 💻 Mac, Win, Linux.
Joined June 2026
- Tweets522
- Following358
- Followers86
- Likes3.9K
Lucena Coder retweeted
I've regained my zen by switching back to @LucenaCoder with GLM 5.2
It just works. You don't get piles of words for the sake of words, it delivers for absurdly cheap on that model specifically for some reason, and it's right on the first try rather than defaulting to lazy brittle choices.
Also, I realized how much I was fighting audio transcription before (I tried the new grok 4.7/cursor setup) and it's night and day difference. This thing doesn't miss a word, and tells the LLM it's voice transcribed so it doesnt' get myopic and start focusing on mistranscribed things as new directions.
Like my calm level returned after switching, it was massively noticeable. I'd rather pay a la carte for GLM tokens via OpenRouter, it's cheaper because it's done right the first time.
What's interesting is how much differently Gemini Flash 2.8 behaves there. I'll use it, but controlled only for design or copy tasks otherwise it just endlessly wants to "stay quiet and work" and surfaces later with a huge number of tokens used relative to what might have been used by GLM 5.2.
Highly recommended if you're tired of screaming at the mainstream harnesses wanting to sit on a fence and pour piles of word slop at you.
Lucena Coder retweeted
There will always be conflict between those making harnesses and those making models.
The labs want the models to be more intelligent and more proactive and burn tokens on the way generating revenue.
The harnesses want the models to respond to architecture and not brute force through everything while blowing tokens.
Literally every new model release gets a bit more aggressive in how far it will go, which changes how a harness must respond to that model if it cares to keep it contained and not a money pit. The RL training leans on "add" and "solve more" so harnesses are left to fight this behavior as it impacts their interaction quality.
It's weird to be involved in. This weird "what IS best?" time.
This is where MCP really does win. You can let anyone use whatever model they want for the MCP call, but build your own intelligence behind the MCP that doesn't depend on flavor of the month. You may even pin an older model with RL training that does the job perfectly. Specialists, not generalists. They do one thing well, and that's the entire MCP's job. The user's model then can't really negatively impact it.
MCPs should deliver outcomes, not tools.
Lucena Coder retweeted
Normal people do not want to build AI Agents. If I have learned one thing running an AI startup that’s sold and delivered millions in contracts, it’s this.
No one cares. No one cares to learn a new mental model. They just want to talk to a thing and get results.
They don’t care about agents, skills, connectors or any of the other neologisms we in tech fetishize.
They just want the thing to work so they can do something else with their lives.
The mistake I see so many companies—and competitors—making is rushing into an ill thought through human computer interaction paradigm where one human manages teams of purpose built agents.
I hate that future and we are actively building the opposite at AgentPress. Our ideal is the agents and the neologisms and the gizmos fade into the background and our customers experience AI as a singular ongoing conversation with a chief of staff who also happens to be a mentor, a friend, and a genius.
I have always believed the true promise of generative AI is that it radically elevates the quality of our relationships with machines, lowering friction and increasing understanding.
I see very few SaaS companies who seem to understand any of this. It is not enough to put a chatbot in your app. If you are not converting your entire company to selling a conversation between your superhuman AI and your customer, you don’t get it.
—
If this resonates with you and you’re on a sales team of 3-10 people and you’re drowning in follow ups, emails, and CRM updates, we are taking on selected design partners for our AI Chief of Staff for Account Executives.
Lucena Coder retweeted
A harness is everything between a raw AI model and useful work.
The model is the jet engine, and the harness is the cockpit: everything between raw thrust and a safe landing.
I fly planes. You can buy the same turbofan Airbus puts in an A320, but that won't fly 180 people to LA.
The engine is a commodity, but the cockpit is how you harness thrust into a destination.
The failure patterns and recovery logic we encode into harnesses for our clients won't show up in an open-source repo, because that's judgment earned from real deliveries, not documentation.
The argument that smarter models make harnesses obsolete is a bad take.
A more powerful engine needs better avionics, not fewer, and you'll hit orchestration problems only your harness can solve as labs ship smarter models.
The harness is where your alpha compounds.
There is no alpha in building your own agent harness.
The best techniques will get discovered, copied, and eventually incorporated into open-source harnesses like Pi.
A few months after that happens, these techniques will be made obsolete entirely as labs release smarter models + model providers integrate upwards (pulling more of the harness behind the model API endpoint).
Am I wrong? Please let me know 🧐
Lucena Coder retweeted
GLM-5.3 Flash weights are out !!! 🔥🔥🔥
huggingface.co/zai-org/GLM-5…
Lucena Coder retweeted
Ox Alpha was an early version of GLM-5.3-Flash. The official release delivers stronger performance and significantly better stability.
Huge thanks to @opencode and @OpenRouter for making it available to the community. And thank you to everyone who tried it, pushed it to its limits.
Introducing GLM-5.3-Flash
- Leading capabilities at a highly competitive price
- Natively multimodal with a 1M-token context window
- A 320B-A18B model released under the MIT License
- Previously previewed as Ox Alpha, running entirely on Chinese AI chips
Blog: z.ai/blog/glm-5.3-flash
Available now across all official platforms:
Weights: huggingface.co/zai-org/GLM-5…
API: docs.z.ai/guides/llm/glm-5.3…
Coding Plan: z.ai/subscribe
ZCode: zcode.z.ai/en
Chat: chat.z.ai
AutoClaw: autoclaw.z.ai
Lucena Coder retweeted
in the last 24 hours, I have used 43.3M tokens. That cost me $0.28 - why would anyone use frontier models?
Lucena Coder retweeted
Found a good workflow to fix that:
- have it setup a centralised design system in the codebase
- ask it to create mockups using it
- review and iterate on it
- then ask it to implement by extracting html/css from the mockup, to snap values to existing tokens (to avoid implementing drift)
It's been working really well for me
Lucena Coder retweeted
Every person I talk to who isn't heavily into AI doesn't understand the concept of a Codex $200 subscription providing $17K worth of API usage.
Heck, they don't even know what an API is. Everyone has accepted that AI costs just $20, or $200 for heavy usage.
Even worse, only 1–5% of users even pay for AI, so most people think it's just free.
There's going to be a big reckoning once this honeymoon period ends.
Lucena Coder retweeted
AI cannot reliably teach you anything that has significantly changed in the last six months.
