Built Different

Where discipline meets mindset. | Sport. Self-dev. Sharp thinking.

[Built Different @BuiltDiffly]

This man's credentials will leave you speechless.

7th degree black belt. World's fastest gun disarm. Former Marine.

All forged from being forced to kill at age 7.

The explosive secret to weaponizing your worst trauma into unstoppable strength:

[João Moura @joaomdmoura]

You can't build a simple agent system. You can only choose where the complexity lives

This is something I've learned over 20 years of building software, but it hits differently with agentic systems. Every architectural decision is really a decision about where you're willing to pay the complexity tax.

I see teams put it in the wrong place all the time. They build elaborate multi-agent orchestration layers with sophisticated handoffs and routing logic. Then they need to change something and it takes weeks because every adjustment cascades through the whole pipeline.

The teams that move fastest do the opposite. They keep the orchestration simple and push the complexity into their tooling, their observability, and their testing. Places where it actually works for you instead of against you.

With agents, this matters even more because you're iterating constantly. The system behaves differently with real users than it did in testing. Prompts need tuning. Edge cases show up. If your complexity is sitting in the orchestration layer, every one of those adjustments becomes painful.

Simple orchestration, smart tooling. That's the pattern I keep seeing win.

[Jon Willbanks @jonwillbanks]

Senior dogs that pace at 3am. Stare at walls. Stop recognizing their family.

Most owners call it "just getting old."

There's a name for what's actually happening.

And a 2018 clinical trial showed one specific intervention can help:

[Mark Woodland @MarkAWoodland]

Most new CEOs arrive to prove they are the smartest person in the building.

Satya Nadella did the opposite.

A thread on how he turned Microsoft around by killing one mindset:

[Martin Felando @MartinFelando]

Four broke men hired to haul nitroglycerin over mountain roads. One jolt and it detonates.

A French master made it in 1953. An American obsessive remade it in 1977.

Two films about what fear does to a man with nothing to lose.

The Wages of Fear (1953) and Sorcerer (1977).

[Marc Gravely @MarcGravely]

The Texas Supreme Court just redrew the map on appraisal in property insurance. If you handle first-party claims in Texas, this one matters.

I've been asked to join the ACCC Pop-Up Dialogue on June 18th to break down what In re ACE American Insurance Company means in practice.

[João Moura @joaomdmoura]

Most agent projects I see are negative ROI. Not because the agents don't work. Because they never make it to production.

We run most of our agents on GPT-4o mini. The models are fine. What actually kills enterprise agent projects is the trust gap. Governance they can't meet, data exposure they can't control, no observability layer. The org just won't let it go live.

We work with PepsiCo, the Department of Defense, DocuSign. Every one of them went through the same realization that building is actually the easy part and all the real work goes into making the system trustworthy enough to deploy.

If you can't observe it, you can't trust it. I keep saying it because I keep watching it play out.

[SammyArmstrong @SammyRArmstrong]

Jennifer Garner is 54 and just overhauled her training.

After years of chasing strength, she realised she'd ignored the one thing that actually predicts how long you live.

Here's what she changed, and the full routine behind it:

[Jon Willbanks @jonwillbanks]

Are we over-vaccinating our pets?

40 years of immunology research says yes. But not in the way most pet parents think.

Here's what's actually going wrong, and how to fix it:

[Jordan Saunders @jsaunders_]

A founder raised $94 million for his first company. It sold for about $30 million.

For his next one, he took no investors at all.

Today it makes over $200 million a year with around 40 people.

It's the strongest case against venture capital I've seen:

[Mark Woodland @MarkAWoodland]

Paul Polman's first act as Unilever CEO wiped 8 per cent off the share price.

Analysts called it reckless.

8 years later, it let him reject a 143 billion dollar takeover.

Here is how he built a company too valuable to buy:

[Marc Gravely @MarcGravely]

Texas spent $60 million on a high school stadium. Within two years of opening, it had to be closed.

The repair bill came to $10 million.

The engineers who investigated the failure acknowledged that doing it correctly during original construction would have added about $1 million to the project.

Think about that math for a moment.

One million dollars in upfront quality would have prevented ten million dollars in emergency repairs on a sixty-million-dollar project.

And that doesn't account for the disruption, the lawsuits, or the years of litigation that followed.

I wish this were an unusual story.

I've spent my career representing institutional property owners who find themselves in exactly this position. School districts, hospital systems, universities, apartment communities, commercial real estate owners. The details change but the trajectory is almost always the same.

A project goes out to bid. The lowest number wins. The board celebrates the savings. Construction begins. The contractor who submitted that low number is now operating on margins that are thinner than anyone on the owner's side realizes. They bid aggressively to win the work, and now they have to protect those margins during construction. So substitutions happen. Value engineering reduces cost but also reduces quality. Subcontractors get selected for price rather than for track record.

None of this is visible on the surface.

The building goes up. It looks fine. The ribbon gets cut. Everyone shakes hands and moves on.

Then, two to five years later, the problems appear.

Water intrusion. Foundation settlement. Building envelope failures. HVAC systems that don't perform to spec.

The kinds of problems that are invisible during a punch list walkthrough but cost millions to remediate once they manifest.

And by that point, the contractor is three projects down the road. The architect has closed the file. The engineer has moved on. Nobody involved in the original construction has any incentive to come back and make it right.

The owner is left standing in a building that was supposed to last 50 years, holding a repair estimate that rivals the original cost of construction.

One question could have changed everything.

Before accepting the project, before signing off on substantial completion, the owner could have asked:

Who audited this work?

Not the contractor's internal QA team. Not the architect who designed the project and has an interest in it appearing clean. An independent third party whose only job was to find problems before they became catastrophes.

That question, asked at the right time, is the most cost-effective investment an institutional property owner can make.

[João Moura @joaomdmoura]

Most of our production agents run on GPT-4o mini. The intelligence piece is basically solved.

What's blocking teams from shipping is architecture, deployment, monitoring, and trust. All the stuff that doesn't make for exciting launch day announcements.

Remember when every model launch was about context window size? 128K, 256K, a million tokens. Nobody talks about that anymore.

Same thing happened with reasoning.

We used to do chain-of-thought manually, now models handle it internally.

When the first reasoning models came out they stopped using tools altogether and tried to reason their way to every answer. That knocked CrewAI offline for a few days.

Andrew Ng told me he thinks we'll end up with just agents working in groups, no scaffolding at all. I'm not fully there yet but the trend is clear.

I'd invest in the stuff you can't spin up overnight: trust, production architecture, real customer data, and distribution. I'm more excited about that than any new model launch right now. Probably a funny thing for an AI CEO to say.

[Terry Lynch @terrybali]

A few weeks ago we hit 17.45m of 9.47% CuEqRec at Lion. For non-mining people, that's a freakishly high-grade copper-nickel-PGE intercept. Here's what those numbers actually mean:

17.45 metres is the width of the mineralized zone the drill passed through. That's not a narrow vein you need to chase. That's a thick, continuous body of metal.

9.47% copper-equivalent recoverable means that if you processed a tonne of that rock, nearly 10% of what you pull out is valuable metal. In copper, nickel, palladium, and platinum combined.

To put that in perspective: the average copper mine operates profitably at 0.4% grade. We're talking about grades more than 20 times higher than what's considered economic.

AI needs copper. Defence needs nickel. Defence contractors need PGEs. We have all three in one drill hole.

This is what "more millionaires made in mining than AI" actually looks like on the ground.

$PNPN $PNPNF

[Jordan Saunders @jsaunders_]

In 2020, Disney CEO Bob Iger retired and handed the company to his hand-picked successor.

Two years later, the board fired that successor and asked Iger to come back.

The stock had nearly halved.

Here's how he turned it around:

[Mark Woodland @MarkAWoodland]

Most career advice is tactical.

Carina Waye's advice is two lines.

Love hard. Show up.

Today she hosts the Kismet podcast. Here's her story:

[Mustufa Khan @mustufa4socials]

https://t.co/YbBKvx1Byk

[Martin Felando @MartinFelando]

An astronaut bound for Jupiter, listening to a calm computer refuse him.

An Irish rogue clawing into the English aristocracy, then losing it all.

Two Stanley Kubrick films on how far a man will go, and what it costs.

2001: A Space Odyssey (1968) and Barry Lyndon (1975).

[Mustufa Khan @mustufa4socials]

In one week, Elon Musk lost to Sam Altman in court.

He then filed to take SpaceX public at a $1.75T valuation. The biggest IPO in history.

People think they're buying rockets & Starlink. When they are buying Elon's bet against Altman.

Here's what's hiding inside this deal: 🧵

[João Moura @joaomdmoura]

We run agents on sales, marketing, and engineering internally at CrewAI. None of them use the same guardrails.

What I use is some combination of LLM as a judge (prompting another model to evaluate outputs), code guardrails (straight Python checking for specific things), and for complex use cases both. We also run a fine-tuned model for prompt injection and PII on the enterprise side.

No single approach works perfectly alone, but combining two or three handles about 99.99% of cases.

You can't spec out guardrails like a traditional feature. You ship the agent, watch it behave in the real world, adjust, and keep going. It feels more like training a new team member than deploying software.

Put miles on your agents. The guardrails reveal themselves once you've put enough time into it.

[SammyArmstrong @SammyRArmstrong]

Henry Cavill is 43 and still has the Geralt physique.

3 seasons of The Witcher. Greek-god build, start to finish.

Here's how he built it and how he keeps it: