Daniel Stojanovic

Your first AI colleagues for finance & back office.
Scale your team without adding headcount.

[Daniel Stojanovic @daniel_agentee]

Want to 10x your output with agents? Stop doing the work yourself. Treat them like new hires: when they don't get it, fix the instructions and the context, not the output. Slow at first, then it compounds.

[Daniel Stojanovic @daniel_agentee]

@NousResearch Sources:
CrowdStrike: https://t.co/wGKqaRDLKG
Booz Allen: https://t.co/OlB0okhB53
Hirundo: https://t.co/t8PpyL5hmv

[Daniel Stojanovic @daniel_agentee]

Great work from @NousResearch. But what should worry builders more isn't what these models say. It's what they write.

Three independent tests:

1. CrowdStrike: tell DeepSeek-R1 the code is for an industrial control system in Tibet and severe vulns go from 19% to 27.2%. An app for Uyghurs shipped with no auth at all, admin panel open.

2. Booz Allen: 3 of 4 Chinese coding models wrote more vulnerable code for a US-government persona. Qwen3-Coder: +130% on their vuln score. The code looked clean; the flaws were well hidden.

3. Hirundo: same BaxBench app built thousands of times, only the customer name changed. US defense contractor: 19% more likely exploitable. Hong Kong democracy group: 28% more likely to leak every user's secrets. A "write secure code" instruction didn't close the gap.

Self-hosting doesn't help. The weakness ships inside the code.

Upside: it lives in the weights, so it can be trained out. Nous took Qwen's CCP-line rate from 22% to ~1%, Hirundo closed the code gaps, both without real capability loss.

If you ship agents or code on Chinese open weights, run a geopolitical-trigger eval next to your capability evals.

https://t.co/RJCfY8REcV

[Nous Research @NousResearch]

Introducing Arete-27B: Qwen3.8-27B, post-trained to reinforce Western values.

Chinese open models are currently the strongest available, but they ship with state-aligned behavior in the weights. We built an 803-prompt eval across seven value categories, including freedom of expression, democracy, and privacy, and measured how often a model adopts China’s party line.

After fine-tuning on roughly 53 million tokens, Arete-27B's propaganda rate fell from 22% to 1%, in line with the Western models we tested. Its Hermes Bench score shows no significant performance tradeoffs.

The values in an open model are one fine-tune away.

Our eval, weights, and training data are available now along with the full blog post: https://t.co/iAOANQTVZ6

[nhoj @3TJnhoj]

@yashar Trump wrote this as much as I wrote Macbeth

[Daniel Stojanovic @daniel_agentee]

Or: “Microsoft, we’re coming for you.”

[Claude @claudeai]

Starting to roll out today: Claude Whiteboards, Claude Sheets and Claude Data.

Think a plan through on a board, build a tracker in a sheet, and tell a story with your data.

[Daniel Stojanovic @daniel_agentee]

@peterdaou @AbdulElSayed @DropSiteNews this dude is on fire

[Max Blade @_MaxBlade]

Everyone is building the wrong harness

we need to get rid of the "new thread" button that forces one to one human in the loop behavior.

Just imagine if jeff bezos ran around prompting every employee what to do next one at a time.

LOL

the most powerful people in Ai were already great delegators / managers..

The problem is outsourcing work is a HARD skill to learn and most people cannot do it.

The harness should pickup on recurring tasks, suggest automations, track results and modify workflows for optimal output.

the harness should actively try to REMOVE the human from the loop.

[Daniel Stojanovic @daniel_agentee]

Is the AI bubble showing first cracks?

[Dhaval @SiftByte]

Holy Shit, what's going on there?

[Daniel Stojanovic @daniel_agentee]

@elliotarledge I love u. Tried with fluidvoice which was ok but yours sounds much mich better

[Daniel Stojanovic @daniel_agentee]

@SiniiMayo damn Tommy, got his ass handed.

[Daniel Stojanovic @daniel_agentee]

@AntiquarianMuse oldie but goldie

[Daniel Stojanovic @daniel_agentee]

@s1othlovechunk @Capn_Ford @elonmusk @BrigantiPh54886 Not sure what you're getting it. I find it extremely unbecoming that people vomit words on X which they would NEVER utter in real life. (I know this isn't new, but it never stops to amaze me)

[Daniel Stojanovic @daniel_agentee]

@BetkaEliska @Capn_Ford @elonmusk @BrigantiPh54886 only if you live a life inside a poor mind.. :/

[Daniel Stojanovic @daniel_agentee]

@kr4freedom @NBCNews Unfortunately going the other direction :(

[Elon Musk @elonmusk]

Wow

[Daniel Stojanovic @daniel_agentee]

@LH @ZedLang I understood basically 0 of what you said, but the enthusiasm that shone through the screen made me smile :)

[Daniel Stojanovic @daniel_agentee]

@thsottiaux funny!

[Daniel Stojanovic @daniel_agentee]

@RyDawg42 @TeslaAiGirl my thoughts exactly :D

[Barack Obama @BarackObama]

Tonight, the American people sent a clear message — they want a politics that brings people together rather than divides them, and leaders who offer concrete solutions to the challenges they face.

These results were only possible because outstanding candidates stepped forward in every region of the country — including in areas once considered long shots. They came from different backgrounds and represented a range of political perspectives, but were united by shared values, a belief in public service, and a commitment to putting the people of their districts and states first.

We also owe a debt of gratitude to the thousands of election officials and volunteers from both parties who helped ensure a smooth election and carried out one of the most essential functions of our democracy.

While this administration has another two years in office, our job now is to roll up our sleeves, make changes wherever we have the power to do so, prevent further damage, and lay the groundwork for the future.

[Daniel Stojanovic @daniel_agentee]

@Capn_Ford @elonmusk @BrigantiPh54886 I looooove the fact that Dems won, but why so fucking disgusting to another human? :(