Narmour

Nathan Armour | Systems Architect & Edge AI Expert for SMEs | Tools not Waifus 🤖

[Narmour @narmour]

I predict that real builders will converge on this. Maximum intelligence means little if we don’t have maximum truth seeking to go with it. This is the new frontier. The open source community are already ahead on this.

When it comes to real agency what do you really think the bottleneck is?

[Elon Musk @elonmusk]

Grok is maximally truthful

[Narmour @narmour]

@Bonks_n_Stonks @Mayhem4Markets @chefnrg Sounds like he has an M&A team that needs regular feeding 😂 he’s created a demon in his org!

[Narmour @narmour]

@no_stp_on_snek @gregorykennedy Literally! Just go full sci-fi with it. Instead of creating its own category it’s trying to be something it’s not.

[Narmour @narmour]

@gailcweiner @benlsage 😂

[Narmour @narmour]

@Mayhem4Markets @chefnrg That’s not a bad thing when you have a bag big enough to cover the losses. He’ll hit a home run soon enough if he keeps swinging the bat!

[Narmour @narmour]

@albadiaaz2 The guy is jacked! 😂 that’s a few years of consistent training with no PEDs.

[Narmour @narmour]

That’s not the reason though, is it?

It has more to do with their models being heavily locked down and unable to assist on frontier/SOTA work. The lack of transparency around it, plus the abrupt “violation” messages on prompts - that’s the daily UX for any serious builder right now (since Opus 4.6 that I know of).

Their broader tactics as a company (lobbying, regulation pushes, etc.) are a whole separate issue.

[Narmour @narmour]

Grok 4.3 (and the upcoming ones) are now their first truly frontier-scale models after rapid scaling on Colossus. The real leverage has always been in infrastructure and distribution, not just raw model intelligence - Elon has emphasized this.

The Cursor acquisition + heavy focus on orbital data centers are the big money moves. You don't need to lead every benchmark by 3-6 months when you control the compute, power, and deployment edge (especially with deals like the Anthropic compute partnership).

How much extra 'intelligence' do most users actually need day-to-day vs. reliable, fast, affordable access at scale?

[Narmour @narmour]

@no_stp_on_snek @MatthewBerman Interesting. I'd like to know more about this.

[Narmour @narmour]

Pull the weeds up by the roots. Everything downstream solves itself.

The government-corporate cronyism that has only intensified since the 2008 bailouts is the real root issue here. This point resonates with a lot of pro-capitalist, pro-market entrepreneurs (and non-entrepreneurs) who want genuine free markets instead of a rigged system.

[Narmour @narmour]

@kimmonismus The recent Mythos/Fable situation + Anthropic's consistent push for heavier regulation/gating probably isn't winning them points with a pro-growth administration.

[Narmour @narmour]

@1337hero @ForrestPKnight Spitting facts right here.

[Narmour @narmour]

How are people actually coding in 2026?

Is the frontier really just brute-forcing LLM calls in autoresearch loops?

I’m still spending ~95% of my time building detailed implementation plans and docs before handing anything off to coding agents. I’ve tried the brute-force/agentic loop approach a few times, but I always end up going back to heavy upfront planning. Hasn’t really changed for me since early 2025.

What’s your actual workflow looking like right now?

[Narmour @narmour]

@BladeoftheS This account still exists? 😂 I see you have excelled at reductionism yet again! I suppose the world needs someone like you to fix it, right? 😂

[Narmour @narmour]

@initjean Memory prices will drop eventually. Probably after the AI wars. There’s huge demand for memory production and it’s happening. In the meantime though buy a GPU! A 3090 can host a surprising amount of local intelligence and that will only improve.

[Narmour @narmour]

@gailcweiner There’s certainly something to be said for those who witnessed firsthand the full onboarding of the internet and computers into the workplace. Adaptation became an early survival skill for that cohort. Youth is great, but it comes with the cost of living through your first cycle.

[Narmour @narmour]

@TheAhmadOsman Hmm, interesting. I would like to know more about this. Are we just benchmaxxing here 🤔

[Narmour @narmour]

@mert Ah shit…

[Narmour @narmour]

@DaleCloudman @arXmedes @cb_doge Exactly. All of the frontier model ‘thinking’ loops etc. That’s a lot of valuable training data to tap into.

[Narmour @narmour]

I'm back baby!

Why do we even need a traditional PC anymore when the real computer is the inference engine?

The future PC should *be* the inference stack first - everything else secondary. We've basically been using computers wrong this whole time because we were missing the most important piece: on-device intelligence.