Homelab. Agentic systems. Metallurgist turned AI builder. Ex-BlackBerry, higher ed, cloud & software dev. Lifelong tinkerer. Halifax dad. Japan obsessed.
Halifax
Joined October 2009
- Tweets3.9K
- Following1.8K
- Followers1.3K
- Likes13.6K
Pinned Tweet
nobody's made an omarchy who-to-follow yet so here's mine 🔳
@NixFred, hosts sunday builders club, ships infomarchy
@ncfrontiersman, co-host, built omastorm
@MichaelGannotti, weekly ai + os roundup
@henrikklee, omarchy on an ipad (yes really)
@tcballard, breaks it so you don't have to
@JamesTervit, writes up every episode
@jrimmer, just shipped fairfax task manager
@Ker0Zelvin, first plugin, synergy clipboard sharing
@NotJustShaking, lee's story is why this matters
@dhh, made the thing
@MarcoPorro, building dock + theme stuff
@derekmross, omarchy evangelist
@simonschubert, made omalauncher (omarchy on your phone)
@SundayBuildClub who'd i miss?
Well if you say so.
Kinda agree here. Would add integration.
Anyone else?
AI trainer is not coming for the software engineer job.
It's what the software engineer job turns into when writing code gets cheap.
Labs don't need more people who can produce code. They need people who can tell good output from bad. And that judgment is exactly what engineers spent their careers building.
The keyboard was never the scarce part.
whoa, the new taskbar in omarchy 4.5 is great 🔳
it works like a top-follower list on x: whatever you use most rises to the front
which app is at the top of yours?
Another stunning video from Imagination Studio and @SamJWasserman and @u1tra_instinct. prepared on B200 via modal.
Didn't turn out exactly like the dirt bike track and jumps like i intended but thats on me. video prompting is still new to me.
who wants to go for a dirt bike ride on mars?
Donnie retweeted
People are on the hunt for an alternative to Qwen 3.6 35B A3B.
Here's a list of quite fresh models you could test, this is my to do list and it constantly grows, all the models I want to test out which are in that rage, many of them are very recent.
So far Tiel Coder and Ornith 1.5 35B from @ornith_ are the best Qwen 3.6 alternatives I've tried but as you see the list is quite long so there's a chance you find something interesting there.
I will be testing them all in real app building mostly as 4bit or 6bit anyway.
> Flint 40B A2.5B from @Recursive_SI : huggingface.co/RecursiveSI/f…
> Raon-1 mini 30B A4B from @Krafton_AI : huggingface.co/KRAFTON/Raon-…
> Sage-A1.5 35B A3B :huggingface.co/InternScience…
> MiniCPM-V 4.7 35B A3B from @OpenBMB : huggingface.co/openbmb/MiniC…
>LFM2.5 24B A2B from @liquidai : huggingface.co/LiquidAI/LFM2…
> Apodex 1.1 mini 35B A3B from @Apodex_AI : huggingface.co/apodex/Apodex…
> K2 Horizon 36B A4B from @IFM_AI : huggingface.co/IFM/K2-Horizo…
> Xing 4.0 29B A4B from Xing Chen AGI : huggingface.co/XingChen-AGI/…
>Nex N2.5 Mini 35B from @NexEcosystem : huggingface.co/nex-agi/Nex-N…
> Bigbang v1 35B A3B from @EFrontierLabs : huggingface.co/endless-front…
x money is closer to canada than people think.
X Payments Canada Inc. already has:
• FINTRAC MSB registration (apr 2026, N300000537)
• Bank of Canada PSP registration under the RPAA (effective jul 14, 2026)
the regulatory work is mostly done. no launch date yet, but canada looks like the obvious next market after the us.
@elonmusk is this right? is canada next?
Donnie retweeted
Here's how I explained @openclaw (and agentic harnesses) to my 87 year old grandmother:
- OpenClaw is the harness. The harness is the car.
- Without a model inside that harness, you can't take it anywhere. The model is the driver of the car.
(1/x)
anthropic now says one agent with skills beats a pile of subagents. in line with what i see, i tried: code, review, arch, ux bots but handoffs dropped and everything stalled in grok.
i'm back to one bot per task. it's way less chaos but somehow just feels slower. maybe the fleet was just busy work?
Anthropic just told developers to delete most of their subagents.
their testing: one agent with skills often beats multi-agent setups. one customer ran about 2x faster after switching.
the new rule of thumb:
→ skills = how your team does the work
→ tools / MCP = taking actions and reading live data
→ subagents = only for parallel work, fresh context, or an overloaded agent
audit your setup:
"Read every subagent, skill and MCP tool in this project. For each subagent, say keep, fold into a skill, or turn into a tool, and why. Build a 10-example eval first, then fold them in one at a time and show me the score before and after each change. Stop if any task gets worse."
OK guys, this is interesting, but as usual the ~27B range (lets call 33B close enough) gets the attention.
nanbeige 4.3 33B is dense. even at q4 that's ~19GB. not happening on my 3060 without the same miracle that we had for Bonsai (looking at your @sudoingX).
so the bar for me is simple. if it doesn't fit in 12GB, it's a benchmark, not a tool. waiting on the 3B.
🚀Excited to announce Nanbeige4.3-33B!
To push the performance ceiling of small agents, our strategy in Nanbeige4.3 starts with building a powerful teacher first.
Nanbeige4.3-33B delivers SOTA agentic performance among open models at ~30B parameters:
* Strong results on agentic coding and cowork benchmarks.
* When integrated into mainstream agent harnesses, it tackles real-world tasks with cost-efficient inference.
* It will serve as a teacher for upcoming smaller Nanbeige models, starting with Nanbeige4.3-3B.
huggingface.co/Nanbeige/Nanb…
huggingface.co/Nanbeige/Nanb…
#LLM #Agent
Omarchy on everything!
Happy weekend☀️what's everyone's weekend project?
I finally dug out my 15-year-old Samsung Galaxy Tab. It can't really do anything anymore, so I'm going to try putting Linux on it, and if it survives that, the Omarchy launcher I've always wanted to try.
Pretty sure it's 32-bit ARM, so I genuinely have no idea if any of this will work. Worst case, it fails as a Linux tablet instead of failing as an Android one.
USB debugging is on. No going back now.
