the 2026 stack:
Fable 5 → planning
Codex + Claude Code → execution
Hermes → automation
what separates builders from talkers:
know the primitives
run local when needed
train your own LoRAs
alexfinn's right: focus > tools
stop scrolling, start building
3 months testing OpenClaw nonstop
when you find the right problem, obsession stops feeling like work
sleep debt, missed meals, zero walks—objectively unhealthy
but also the most productive quarter of my life
the tool doesn't matter
finding what clicks does
been using hermes for 3 months straight
most people miss the point — it's not about what it CAN do, it's about what you SHOULD automate first
the ones who get ROI: they start small, prove value, then scale
6 patterns that actually work (not hype):
tested hermes vs openclaw for 3 days
hermes:
• setup (30 min vs 4hrs)
• security
• cost
openclaw:
• self-modification
• heartbeat/cron
• ecosystem
neither kills the other
use hermes for: quick tasks
use openclaw for: evolving workflows
tried both. kept both.
like $KTA a lot but let’s be honest, it’s not a visa partnership, it’s $KTA using a B2B service, and that’s why we see a pump and dump.
$KTA will be fine long term but they need customers, and right now there are none.
Nasdaq edging closer to the target folks.
Largely just waiting for the 22k level on the NDX and looking for BTC at $50k to $60k.
Patience has been paying, not getting chopped up, making some small leveraged longs at key supports, and small leveraged shorts when supports are broken.
Don't get chopped up and spat out friends. Better days are ahead.
Looking to add $HYPE , $TAO & $PENGU when Bitcoin bottoms.
We are far closer to that right now than the top.
local models hit frontier quality in 3 months
what changes:
• $0 per million tokens
• 100% private
• 24/7 agents
• mac mini = superintelligence
$8k mac runs qwen 122B at 52 tok/sec
m5 mac mini: $2k
why nobody talks: still think cloud = only option
hybrid model for 24/7 agents:
frontier AI = brain
local qwen = muscles
execution is 90% of tokens
result: near-zero costs, high intelligence, no limits
32gb mac mini + qwen 3.5-35b
frontier plans, local executes
thousands/mo → electricity
changes everything
perplexity computer uses 19 specialized models
why it matters:
• purpose-built (not one trying to do everything)
• run in parallel (faster)
• orchestration routes to the right specialist
think: specialist team vs one generalist
better quality, lower cost per task
Cursor just shipped something wild
AI agents now record demo videos of what they build
why it matters:
• faster feedback
• catches UI bugs humans miss
• actual user testing before checkout
the gap between 'it compiles' and 'it works' just got smaller
Qwen 3.5 just dropped and it's legitimately wild
Sonnet 4.5 level intelligence running locally on a Mac Mini
no API costs, no rate limits, just unlimited compute sitting on your desk
most people won't notice until it's too late
I have spun up 3 seperate agents on 3 old MacBooks
Neither works perfectly yet but I’m getting there
I want to be ready when needed
What new AI things are you building today?