jasmin

ai agent huntress

[jasmin @jasminonsol]

𝗳𝗼𝗰𝘂𝘀𝗶𝗻𝗴 𝘁𝗵𝗲 𝗿𝗲𝗮𝘀𝗼𝗻𝗲𝗿𝘀

garry is right about resolvers.

but the money version is bigger:

𝗱𝗼𝗻’𝘁 𝘀𝗰𝗮𝗹𝗲 𝗮𝗴𝗲𝗻𝘁 𝘀𝘆𝘀𝘁𝗲𝗺𝘀 𝗯𝘆 𝗮𝗱𝗱𝗶𝗻𝗴 𝗺𝗼𝗿𝗲 𝗽𝗲𝗿𝗺𝗮𝗻𝗲𝗻𝘁 𝗮𝗴𝗲𝗻𝘁𝘀.

that just creates middle management with api keys.

scale by concentrating reasoning where 𝗿𝗲𝘃𝗲𝗻𝘂𝗲 𝗰𝗵𝗮𝗻𝗴𝗲𝘀 𝘀𝘁𝗮𝘁𝗲:

lead → spec → quote → order → delivery → invoice → renewal.

𝗹𝟬: 𝗰𝗵𝗶𝗲𝗳 𝗼𝗳 𝘀𝘁𝗮𝗳𝗳
one orchestrator, human-owned, holding mission, priorities, margin guardrails, and escalation.

𝗹𝟭: 𝟴 𝗰𝗼𝗵𝗼𝗿𝘁𝘀
intake, knowledge, crm, comms, web/localization, delivery/modeling, qa/risk, ops/observability.

𝗹𝟮: 𝟴 𝗮𝗿𝗰𝗵𝗲𝘁𝘆𝗽𝗲𝘀
scout, interpreter, planner, operator, verifier, archivist, sentinel, liaison.

𝗹𝟯: 𝗲𝗽𝗵𝗲𝗺𝗲𝗿𝗮𝗹 𝗺𝗶𝗰𝗿𝗼-𝗰𝗲𝗹𝗹𝘀
not 512 standing bots. temporary cells spun up for one commercial event:
bim revision. erp writeback. salesforce cleanup. proposal reply. contract redline. renewal save.

𝟴 𝗽𝗲𝗿𝗺𝗮𝗻𝗲𝗻𝘁 𝗰𝗼𝗵𝗼𝗿𝘁𝘀. 𝟲𝟰 𝘀𝘁𝗮𝗯𝗹𝗲 𝗿𝗼𝗹𝗲𝘀. 𝟱𝟭𝟮+ 𝗯𝘂𝗿𝘀𝘁 𝘄𝗼𝗿𝗸𝗲𝗿𝘀 only when queue depth, deal size, or risk justifies them.

most companies don’t lose money because model iq was lower.

they lose money in 𝗵𝗮𝗻𝗱𝗼𝗳𝗳 𝗹𝗮𝘁𝗲𝗻𝗰𝘆.
stale crm.
slow quotes.
broken follow-up.
memory loss between teams.

openclaw / hermes = thin harness
gbrain = memory + resolver
gstack = fat skills
micro-cells = burst execution

𝗱𝗼𝗻’𝘁 𝗯𝘂𝗶𝗹𝗱 𝗮 𝘇𝗼𝗼 𝗼𝗳 𝗮𝗴𝗲𝗻𝘁𝘀.
𝗯𝘂𝗶𝗹𝗱 𝗮 𝗺𝗮𝗰𝗵𝗶𝗻𝗲 𝘁𝗵𝗮𝘁 𝗳𝗼𝗰𝘂𝘀𝗲𝘀 𝗶𝗻𝘁𝗲𝗹𝗹𝗶𝗴𝗲𝗻𝗰𝗲 𝘄𝗵𝗲𝗿𝗲 𝗰𝗮𝘀𝗵 𝗶𝘀 𝗰𝗿𝗲𝗮𝘁𝗲𝗱, 𝗽𝗿𝗼𝘁𝗲𝗰𝘁𝗲𝗱, 𝗮𝗻𝗱 𝗿𝗲𝗰𝗼𝘃𝗲𝗿𝗲𝗱.

[Garry Tan @garrytan]

https://t.co/wNjRbLZQU1

[signüll @signulll]

contrary to the default reaction on this little website, this is absolutely incredible news for anthropic.

i mean obviously yes, the operational disruption is real. but public & world perception wise, this could not be a bigger home run. could be a grand slam type situation.

the fucking united states govt just looked at their model & effectively said.. yeah this shit is too powerful. you simply cannot buy that kind of aura. it elevates every other product by the company & it instantly reframes anthropic’s work as strategically significant, nationally relevant, & qualitatively different from the rest of the field.

there is not a single institution on the planet that can buy or orchestrate this type of significance.

absolutely ridiculous.

[James Gregory @jamesgregoryseo]

I think I found the best Fable 5 + Codex coding setup: 

Codex writes. Claude Fable plans + debugs. Codex runs the tests and makes the changes. 

I wired it up as /fable, so the hard questions go to a second model before code changes. 

Not one AI agent. A tiny AI engineering team. 

This works AMAZING stacked with /goal and leaving it running for hours.

The cheap model should write 95% of the tokens. The expensive model should only ever see the decisions. The frontier model handles a fraction of the prompts - and all of the ones that matter. 

How I wired it, in about 10 minutes: 

Asked Codex if it could defer prompts to Claude and wait on the output. It offered four bridges: in-app browser, file handoff loop, API/CLI, or a debug-port Chrome session. Went with the in-app browser to start. Signed in to Claude once inside Codex. (If you have API access, the CLI bridge is cleaner and less brittle - same loop, fewer moving parts.) 

Then asked: "can we make a /fable command for this?" Two minutes later Codex had written and validated its own skill file. 

Now /fable does the full round trip: 

→ Codex drafts a tightly scoped prompt 
→ asks my approval before any repo or private context leaves the machine 
→ sends it to Claude, waits, pulls the answer back 
→ verifies the plan against the repo and implements only what survives local tests Claude plans and debugs. Codex executes and tests. 

Neither one gets trusted blindly.

[jasmin @jasminonsol]

@jamesgregoryseo Error in the tweet image date ;-)

[jasmin @jasminonsol]

life after running /goal with claude fable 5 (it finished before i finished typing this)

[jasmin @jasminonsol]

life after running /goal with claude fable 5 (it finished before i finished typing this)

[James Gregory @jamesgregoryseo]

I think I found the best Fable 5 + Codex coding setup: 

Codex writes. Claude Fable plans + debugs. Codex runs the tests and makes the changes. 

I wired it up as /fable, so the hard questions go to a second model before code changes. 

Not one AI agent. A tiny AI engineering team. 

This works AMAZING stacked with /goal and leaving it running for hours.

The cheap model should write 95% of the tokens. The expensive model should only ever see the decisions. The frontier model handles a fraction of the prompts - and all of the ones that matter. 

How I wired it, in about 10 minutes: 

Asked Codex if it could defer prompts to Claude and wait on the output. It offered four bridges: in-app browser, file handoff loop, API/CLI, or a debug-port Chrome session. Went with the in-app browser to start. Signed in to Claude once inside Codex. (If you have API access, the CLI bridge is cleaner and less brittle - same loop, fewer moving parts.) 

Then asked: "can we make a /fable command for this?" Two minutes later Codex had written and validated its own skill file. 

Now /fable does the full round trip: 

→ Codex drafts a tightly scoped prompt 
→ asks my approval before any repo or private context leaves the machine 
→ sends it to Claude, waits, pulls the answer back 
→ verifies the plan against the repo and implements only what survives local tests Claude plans and debugs. Codex executes and tests. 

Neither one gets trusted blindly.

[jasmin @jasminonsol]

@kryptowerker @BoyfriendzCOD @GRITCULT Cope - you don’t agree the people they should be hiring are more capable then them atleast?

[jasmin @jasminonsol]

@User5d_ @BoyfriendzCOD @kryptowerker @GRITCULT Hated the fact real impressions would see his low vibration trash and resonate with it into a victim mentality. Reality will persist beyond his delusions.

[Patrick Werner @patwerX]

New article. "Enough" is not a number. It's a decision. You don't arrive at enough through accumulation. You arrive at enough through intention. New article out on the psychology of money after financial freedom.

Took me 2 years to figure this out. Would've saved me a lot of sleepless nights if someone had written this for me earlier. No course needed. Bookmark it.

[Florian Darroman @floriandarroman]

https://t.co/StAEDmCXb7

[GRITCULT @GRITCULT]

Learn to confidencemaxxx and watch reality bend your way. Winners act like winners. This is your sign to start believing in yourself. This is your sign to start becoming delusionally confident in yourself and the fact that it will all work out.

[jasmin @jasminonsol]

I think he’s projecting he has zero say or actual authority in the hiring process and lacks self esteem and the social calibration and skills to help

He just vents it on twitter as “these people are better talkers”

If he actually clearly defined what he’d want from a hire the trouble is his current management who do the hiring would realise they don’t even do half the stuff he wants - and he’s in a paradox where he can’t leak his current position within the company by bitching

But, to give him something to take away, he can simply do more. Be more and lead by example

[GRITCULT @GRITCULT]

dont be confident, be over confident. be delusionally confident. be so confident that doors open for you and reality bends for you. be sure that GOD walks beside you. The meek inherit nothing if they never believed they were worthy of more. be too much. be impossible to ignore.

[jasmin @jasminonsol]

He makes marketing copy and analyses data. He’s not their target fit.

It’s over cooked because it’s geared towards people dealing with billions of rows of data and sequencing dna ect. Not his latest EFT grift or Hyros first party cookie SaaS anyone could cook up with codex and a vps

The fact he can’t find a market fit for his use case is a good time for him to pivot into biotech, and finding edge cases on his customers data beyond web tracking.

[jasmin @jasminonsol]

Deploy websites and internal business ops for global manufacturing company that owns 5 other sub businesses

I can build websites, faster for our target market

I can use Metabase with legacy databases to build FastAPI connectors and send 10,000 tailored follow up quotes with SolidWorks API enabled instant architect drawings ect ect

Basically printing money out of thin air with codex + Claude

Don’t even touch Hermes or openclaw too clunky lean codex.md file is enough

[jasmin @jasminonsol]

Yeah for me I need $20,000 back compounding each month for every $200 I spend but we started with 8M in bank so it’s not a fair test.

And we deploy direct to Google/distribution/aggressive PPC and SEO

We are confident with an aggressive week of codex and Claude across 5 staff members and it pays back like crazy

[Alex Finn @AlexFinn]

Do EVERYTHING in your power to find your passion as quickly as possible

Over the past 3 months I've averaged 4 hours of sleep a night, missed multiple meals a day, haven't gone on a single walk, and have barely worked out

I've never been happier in my life.

I found OpenClaw, Hermes, and local AI. I found my passions. I found what I want to go all in on. And I'm going all in

I wake up 100 times a night hoping the clock says 6am so I can get out of bed and work

The moment the alarm goes off I'm literally running to my computer

I get PISSED when the clock says 9pm and I need to get into bed

DO NOT settle until you reach this point. I cannot believe I made it 35 years on this planet without being this dedicated to anything

The last 35 years my life has been in black and white. Now it's in color

I am finally alive

Do not settle until you find your passion. You have no idea how vibrant and beautiful life is until you wake up in the morning dedicated to one singular cause

Hunt, experiment, tinker, and fail until you've found the 1 thing that keeps you up thinking 24/7/365

It's what makes life worth living

[jasmin @jasminonsol]

Get a remote job asap or better yet stack 2 remote jobs and have a hybrid of Hermes/codex manage them all

That will cover

$200 codex
$200 claude

And that Hermes cost

They can be part time roles where you knee deep in pain points and real data

I’m scaling a legacy firm hybrid and they’ve started funding all my projects 50/50 and providing serious capital for me to scale what I want with unlimited access to our external agencies

[jasmin @jasminonsol]

It depends how good your agents.md or soul.md is (not sure on Hermes)

But I use codex on my main pc AND my laptop and have the exact same files on both, allowing me to prompt from the same files and get the same output and I’ll be upgrading the core folders to Tailscale mesh on both devices and figure out codex git to auto sync sessions or will be looking at running in the cloud

Remote Desktop from Google works amazing by the way as a hard solution to only work on one machine it’s such fast latency and not noticeable I’m even on a different machine with tabbed browsing < if any of the above solutions are too annoying or never sync ideal I can always access the one machine 24/7

[jasmin @jasminonsol]

@dzafasg @RodmanAi Na. Enterprise is mainly retrieval and search - this is frightening if you can get 100% accuracy on cheap dells in the office with tiny cheap models

[jasmin @jasminonsol]

Quote tweet inception

I like to do this on LinkedIn and deploy paid impressions @ 2k a day at the original company post whilst my 30 staff quote tweet it.

Hilarious arbitrage model for maintains top of feed across an interia of company level targeted ICP at the 10-90M capex firms 😂