Mony Mony

I love AI, tech, investing, and forward looking statements.

[Mony Mony @monymonykay]

Every time I’m reminded that Jensen Huang is no longer CEO of Nvidia, I shed a tear.

[Mony Mony @monymonykay]

@levie excellent model for everyday gemini users

[Mony Mony @monymonykay]

Have your agent test on the AWS emulator first for quicker iteration - https://t.co/O4ie8xUAjs

Love this release.

[Mony Mony @monymonykay]

Jensen is out as CEO of Nvidia. Crazy times.

[Mony Mony @monymonykay]

@StockSavvyShay Worst timing ever.

[Mony Mony @monymonykay]

Anthropic put a price on itself yesterday. 825 million shares at $70 to $78, up to $64.4 billion, listing on Nasdaq as ANT with trading expected November 19. The FT says bankers already expect to price above the range after strong first-day orders. Run-rate revenue hit $81 billion at the end of October, up $10 billion in a single month. Google and Nvidia are each putting in about $10 billion more in a private placement at the IPO price.

The same day, IBM and Microsoft launched Imagine Enterprise, a platform seeded with 6,000 prebuilt agents and assets, financial services first. SEI, about $2 trillion in assets, is the named customer. The pitch is agents that run industry processes, governed, inside regulated firms.

The market just priced the labs at $1.4 trillion. The buyers of the labs' output will not let the agents write yet. PYMNTS says agentic AI is in use at 89% of large US enterprises, and 78% of them would keep agents on read-only access. The firms I work with do not own the AI trade. They buy capacity by the token. Their question is what happens to their price when the vendor's funding gets more expensive.

Three things to ask before your next vendor renewal. Ask for the token price path tied to their funding, not just today's rate. Ask which of those 6,000 agents have actually run in production at a firm like yours, with a named customer. Ask what it takes to move one agent from read-only to write, and who signs for it.

[Mony Mony @monymonykay]

@levie The only limitation is token budget :). Once token cost drop, we’ll see a creativity explosion and usher in the golden era of mind to reality.

[Mony Mony @monymonykay]

This weekend the agent grew a memory. Anthropic published how Claude remembers between sessions. It writes notes, replays them offline while it "dreams," and checks the notes against the original transcript to fix the ones that got it wrong. Meta started rolling out memory models to every Muse and Meta AI user the same week.

The research is honest about the failure mode. One Claude played Pokémon and wrote 413 memory files across 225 folders. It earned one badge. Another wrote nine files and earned four. The first one kept saving things it could have read off the screen, and most of it went stale. At one point it wrote down the wrong side of a mountain, marked the job done, and walked off a ledge three times.

The firms I work with cannot switch memory on and hope for the best. Every note an agent writes is a record. It has to be retained, audited, and deleted on request, under the same permissions as the person who asked for it. Memory is a records problem before it is a feature. The same weekend showed the other side. Chainalysis said an agentic attack exposed names, password hashes and live tokens for all of its users, and took its products offline.

Before you turn agent memory on, ask your vendor four things. What does it remember. Where does it live. Who can read it. How do I delete it. Four answers on one page. That is the whole test.

[Mony Mony @monymonykay]

@levie ** every two months

[Mony Mony @monymonykay]

China wants to win in AI, they converted an exiting vendor into a spy. Jensen fell for the temptation. He fessed up. They investigate to determine what IP China got from
Her. Then AI competition intensifies. Now everyone understands why this is a global competition that will set the stage for the next 100 years.

We are just beginning this journey. More drama to come.

[Mony Mony @monymonykay]

@BillAckman @JensenHuang As soon as you get famous everyone will try to find dirt on you. It’s sad.

[Mony Mony @monymonykay]

@amitisinvesting Yall need to separate a persons ethics from their duty to their company and capitalism. Stay out of people’s private affairs.

[Mony Mony @monymonykay]

@OfficialLoganK We need to keep increasing intelligence while lowering cost. This distribution of intelligence will bring creativity and innovation like humankind has never seen before.

[Mony Mony @monymonykay]

@pdicarlotrader Why Peter?

[Mony Mony @monymonykay]

@unusual_whales This type of “reporting” should be banned. It’s just regurgitating headlines with zero understanding. Spreading false reporting. Sad to see

[Mony Mony @monymonykay]

@stevenfiorillo This is normal

[Mony Mony @monymonykay]

Every company is building their own AI chat agents for their end users but not realizing that end users don’t want to use their chat agents, they just want their own personal chat agents to talk with and interface with your services through a single funnel where they run everything in their life and business.

So don’t force a user to use your company’s chatbot on your website, just make it accessible to their personal agent.

It’s the same way with developers, they don’t want to switch interfaces, they want to remain in a single ide and do everything from there.

[Mony Mony @monymonykay]

Bro doesn’t know what’s coming next…

[Gegam @Gegam245074]

I don't understand how to enjoy agentic coding anymore. A year ago it was really exciting. You'd describe a task, and Opus 4.5 would come back with a result in 10 minutes. You could literally build a small project in 40 minutes. Now I give a task to Opus 5.5 and wait at least two hours. I scroll TikTok, go for a walk in the woods, head out to run errands. I come back, and it's still working. By then I've stopped caring about the project. I can't be bothered to check the result. I have to really psych myself up and force myself to switch back into work mode and describe something

In that time, I could have been coding by hand and enjoying it way more

Something in agentic coding has broken, and something needs to be done about it. I've noticed that I haven't touched agents in two weeks and have done practically nothing. I've never been so happy

[Mony Mony @monymonykay]

@brian_armstrong The single queue is the best setup for human view. Agents can execute in parallel but we need to review linearly for high critical decisions.

[Mony Mony @monymonykay]

@NASM106 Jobs are welcome