Diego L

[Diego L @diegolancha]

The path to immortality… specialized models trained in vast amounts of human body data

[Anthony Pompliano 🌪 @APompliano]

Can Eli Lilly become the largest company in the world?

@jvisserlabs thinks so.

"The reason I believe Eli Lilly has a chance to be the largest company in the world, and the number one AI company in the world within 5 years, is because they are building a specialized model."

"They have their own data center with 1,000 GPUs with all the data that $LLY has had [as] a 150 year old company."

"They are creating an innovation hub of using data all for human software."

[Anthony Pompliano 🌪 @APompliano]

My current philosophy on technology and finance:

Use artificial intelligence to create revenue, use bitcoin to protect profits.

[Peter H. Diamandis, MD @PeterDiamandis]

We're so close to longevity escape velocity (LEV by 2033) that your sole responsibility right now is to avoid dying from something stupid. The next 5 years will deliver more medical breakthroughs than the previous 50.

[Diego L @diegolancha]

A step in the right direction. Now we only need to solve for trust. An agent can discover a service and pay for it, but how does it know the service is honest, the price is fair, and the output wasn’t fabricated? Payments are the easy part. Verification is the hard part.

[Nick Prince🛡 @Nickprince]

this was my most engaged post in 9 years on X. half a million people saw it 🤯

so I looked at who showed up. ~1 in 3 of you have zero crypto in your profile. AI engineers, SaaS founders, payments infra builders. you came because "agents paying for APIs without API keys" is a problem you already have.

that's what makes this interesting. crypto people building the rails. AI people building the agents. both sides need each other for this to work.

glad you're all here.

[Diego L @diegolancha]

“Claude, audit my CRM system, copy data to snowflake, and replicate workflows and dependencies in python scripts…

Salesforce, I won’t be renewing this year”

How can any SaaS avoid this?

@Benioff

[Marc Benioff @Benioff]

Welcome Salesforce Headless 360: No Browser Required! Our API is the UI. Entire Salesforce & Agentforce & Slack platforms are now exposed as APIs, MCP, & CLI. All AI agents can access data, workflows, and tasks directly in Slack, Voice, or anywhere else with Salesforce Headless 360. Faster builds, agentic everything. 🚀
#Salesforce #Agentforce #AI
https://t.co/mxySdJS7HR

[Diego L @diegolancha]

@levie And how do you even keep up when everyone can copy your data structure anywhere else and apply their own agentic flows on top?

[Diego L @diegolancha]

@Benioff Even so… what will the moat be?

You can have full audits of the CRM data system with Claude, store and replicate the structures in another database (without the SF premium), and get the same results like if you had Salesforce… the agentic workflows should be superior…

[Diego L @diegolancha]

So much food for thought…

[Peter H. Diamandis, MD @PeterDiamandis]

https://t.co/Ve3Vn9F0Z3

[Diego L @diegolancha]

@bygregorr @APompliano Basically dumped a research piece into NotebookLM… smart

[Diego L @diegolancha]

Curious about this, does zkML verify that the model itself is trustworthy, or just that the output wasn’t tampered with after inference? Because in my experience building agent pipelines, the trust problem isn’t usually at the compute layer. It’s upstream: who decided what data the model saw, and who authorized the agent to act on it.

[Diego L @diegolancha]

AgentFM turns idle GPUs into a P2P AI grid.

This is the hardware layer of the agent economy starting to emerge.

The thesis: as frontier compute gets locked up by hyperscalers and enterprises, decentralized markets could be how smaller players stay in the game - slashing costs by tapping idle capacity worldwide.

But it introduces a massive new problem: trust and economic governance.

Who's accountable when compute is anonymous and volatile? How do you track spend across a decentralized marketplace? How do agents verify they got what they paid for?

The real bottleneck for decentralized agent compute won't be capability. It'll be building the economic trust layer needed for anyone to actually rely on it.

Capability isn't the bottleneck. Trust is.

https://t.co/u7dutnAGnF

[Diego L @diegolancha]

@delveroin https://t.co/Wt5v8N9Mfx

[Diego L @diegolancha]

@web3devop https://t.co/Wt5v8N9Mfx

[Diego L @diegolancha]

@Amank1412 https://t.co/Wt5v8N9Mfx

[Diego L @diegolancha]

@BatsouElef https://t.co/Wt5v8N9Mfx

[Diego L @diegolancha]

Super interesting. Frontier models will still matter at the edges, but the 22x scaffold vs. model swap stat says it all.

As compute gets scarcer and enterprises lock in frontier access, individuals get priced out. That’s when innovation shifts to exactly what you’re describing - harness design, orchestration, efficiency.

Constraints accelerate ingenuity.

[Diego L @diegolancha]

This is something most people aren’t thinking about yet. If compute stays scarce, the downstream effect is access stratification - big corporates lock in frontier model output through capacity deals, and everyone else gets rationed or priced out.
The play for individuals and small teams: architect for model-agnosticism from day one. Tier your workloads, frontier models only where the delta justifies the cost, smaller specialized models everywhere else. Build agent orchestration layers that route intelligently across providers.
Ironically, compute scarcity might accelerate the “agentic” shift faster than abundance would. Constraints force efficiency, and efficiency favors builders who understand the full stack, not just the biggest models.

[Diego L @diegolancha]

@heyblake https://t.co/Wt5v8N9Mfx

[Diego L @diegolancha]

@helloitsaustin Love the mindset of building for yourself. I’ve been leveraging other repos and skills for the fear of moving too slow, only to find out that I need to spend much more time fixing, tweaking, etc. to make these external skills actually fit my use case

[Diego L @diegolancha]

@heynavtoor How is this better than RAG? Shouldn’t this be limited by the size of the context window?