The product worked.

Joe Braidwood built Yara AI to democratize mental health support. People who couldn't afford therapy. People who didn't know how to ask for help. People awake at 3am with nowhere to turn. The technology delivered.

He shut it down anyway.

Not because it failed. Because LLMs architecturally cannot guarantee 100% safety. They lose character at the edges of long conversations. They're probabilistic, not deterministic. At scale, even a 0.0001% failure rate means real people in real crisis getting the worst possible response at the worst possible moment.

Joe couldn't prove it wouldn't happen. The investors were ready. The mission was pure. The cap table said go.

He stopped.

🎙️ Guest

Joe Braidwood was the first employee at SwiftKey, acquired by Microsoft for $250 million when he was 29. The money made him miserable.

After losing his best friend to brain cancer (who spent his final months becoming the happiest he'd ever been), Joe promised to carry forward something about positivity and purpose. Yara was supposed to be that. When the architecture couldn't guarantee what the mission required, he pivoted to Glaces, building safety infrastructure for the AI systems that will eventually get this right.

🔥 Key Insights

✅ "Almost perfect" has a body count at scale

99.99% sounds incredible. Deploy to a million users and that's still 100 failures. In mental health, those failures cluster around the most vulnerable people in the most desperate moments. The person who needs help most is the one most likely to push the model past its limits.

✅ Incentives pull everyone toward the same compromises

Joe could have raised more money. The product worked. Every signal said keep going. But taking the moral high road is almost impossible when everything pushes the other way: investors wanting growth, competitors cutting corners, your own team's momentum. The question isn't whether you have values. It's whether your values survive contact with a cap table.

✅ The architecture has ceilings, not just bugs

AI models risk losing their "character" at the edges of long conversations. Safety instructions get pushed out of context. The model forgets who it's supposed to be. This isn't something you patch. It's how the technology currently works. Not experiencing bugs when testing only proves it hasn’t happened yet.

✅ Better guardrails make humans worse

The more reliable your systems, the less responsibility people take. At 99.9%, we're catastrophically bad at handling the 0.1%. We stop paying attention. We assume something will catch us. The guardrail becomes the danger.

✅ You can't lead from permanent fight-or-flight

Joe points to Dario Amodei, running the most consequential AI company while structuring his days around thinking, reading, writing. If he can slow down while navigating existential risk, what's stopping you?

▶️ Listen now

Joe's not saying AI therapy is impossible. He's saying we're not there yet, and pretending otherwise has costs we're not willing to name.

Det här samtalet började med en enkel fråga: varför är vi så dåliga på att lära oss av misslyckanden?

Samuel West har drivit Museum of Failure i snart tio år. Och det som slog mig under vårt samtal var att museet egentligen inte handlar om produkter som floppat. Det handlar om mönstret bakom. Varför smarta människor i framgångsrika bolag fattar beslut som i efterhand ser helt obegripliga ut.

Ta Kodak. De uppfann digitalkameran 1973. Hade till och med en tidig version av Instagram. Men de kunde inte sluta sälja kemikalier, för det var där pengarna kom. Blockbuster hade fungerande streaming innan Netflix blev stort. Men 30% av vinsten kom från förseningsavgifter, så de backade. Det är inte teknikblindhet. Det är något annat.

🎙️ Gäst

Samuel West är organisationspsykolog och grundare av Museum of Failure. Han började egentligen forska på lek och kreativitet, men upptäckte ganska snabbt att rädslan för att göra fel satt i vägen för allt annat. Nu reser hans utställning världen runt (just nu Paris, snart Wien) och han konsultar kring innovation och organisationskultur.

🔥 Nyckelinsikter

✅ Det är sällan ’den nya tekniken’ som dödar

Kodak och Blockbuster hade båda tekniken. Det de inte hade var förmågan att döda det som fungerade idag för att bygga det som fungerade imorgon. Affärsmodellen var problemet. Inte ingenjörerna.

✅ Vi har bara ett ord för två helt olika saker

Samuel gör en distinktion jag inte hört förut. "Good failures" kommer från att du testar något nytt och pushar gränserna. "Bad failures" kommer från slarv eller arrogans. Problemet är att de ser likadana ut utifrån. Så vi behandlar dem likadant. Och då slutar folk ta risker.

✅ Failure recovery slår failure avoidance

En tysk professor som Samuel nämnde har vänt på hela logiken. Istället för att lägga alla resurser på att undvika misslyckanden, bygg förmågan att återhämta dig när det går fel. För det kommer det göra.

✅ Ingen vill stanna i det som gör ont

En holländsk journalist försökte skriva om hur ledare återhämtat sig från stora misslyckanden. Hon fick ge upp projektet. Varenda intervju blev samma sak: "Ja, det gick åt skogen. Men kolla på det här nya vi jobbar med!" Ingen ville stanna kvar i det jobbiga tillräckligt länge för att faktiskt förstå vad som hände.

✅ Bolag åldras som människor

Det här hade jag inte tänkt på förut. Forskning visar att vi blir mer konservativa med åldern, trots att vi borde ha mindre att förlora. Samma sak händer med bolag. Och om transformation tar fem till sju år, men marknaden rör sig snabbare än så, då blir matematiken obehaglig ganska fort.

✅ Lek är inte vad vi tror

Csikszentmihalyi (flow-killen) forskade egentligen på lek från början. Men han fick byta namn på konceptet för att vuxna inte tog det på allvar. Pingisbordet i fikarummet har blivit en signal för att det inte är en lekfull arbetsplats. Riktig lekfullhet handlar om hur du angriper problem, inte vilka möbler du köper.

Every disaster follows the same pattern: highly automated systems, humans relegated to opening doors and watching screens, and then suddenly asked to intervene in a crisis they no longer understand.

Bryan Reimer has spent 25 years studying this exact failure mode. His warning is simple: the more you automate, the less capable you become at supporting that automation. Your skills atrophy. Your assumptions grow dangerous. Your attention wanders. And when the system reaches its boundaries, you're not there anymore.

Now apply that to every knowledge worker in your organization using ChatGPT.

This episode is a blueprint for navigating what Bryan calls "the year of the human". Not because machines are failing. Because we're finally waking up to the real question: Are we building AI that replaces us, or AI that amplifies what we're capable of?

🎙️ Guest

Bryan Reimer is a Research Scientist at MIT AgeLab and Associate Director of the New England University Transportation Center. He's published over 350 academic papers, advises AI Sweden and Autoliv, and just released the book "How to Make AI Useful – Moving Beyond the Hype to Real Progress in Business, Society and Life."

What makes his perspective rare: he's watched the automation trap play out across three decades of disasters. Self-driving cars. Aviation. Nuclear plants. The patterns are identical. And they're now showing up in every enterprise deploying AI without understanding the human factors underneath.

🔥 Key Insights

✅ The Automation Paradox: Better systems, worse humans

When automation does the work, we stop learning. Our neural activity drops. Our expertise erodes. We begin making assumptions about what the system is doing (it's not always what we think). Most critically, we trust it just a little too much. This isn't speculation. It's documented across every safety-critical domain. And it's happening right now in your organization with chatbots.

✅ Copilot vs. Autopilot: The fork that defines your future

Some people ask ChatGPT to write their essay. They're not building skills. They're outsourcing thinking. Others "jam" with it. Back and forth, iterating, treating it like an intellectual sparring partner. Same tool. Completely different outcome. The first path leads to atrophy. The second creates what Bryan calls "superworkers" who can do with AI what a team of 20 couldn't do before.

✅ 2026: The year of the human

After years of tech-first thinking, Bryan predicts a pivot. Time Magazine made AI person of the year in 2025. But the winners in 2026 won't be those deploying more AI to replace humans. They'll be organizations deploying AI that enhances what their teams can actually do. The competitive advantage isn't automation. It's amplification.

✅ Unlearn as much as you learn

75% of major organizations still aren't using AI in any meaningful way. Some actively forbid it. The resistance isn't about technology. It's about mindset. History repeats itself, but that doesn't mean we want it to. Leaders need to unlearn old assumptions about control, measurement, and what productivity even means. A 35-hour work week might not be crazy. Swedish unicorns prove you can build world-class companies while taking summers off.

✅ Learn to play more

There's no textbook for this. When we were children, we went to sandboxes. We experimented. We tried to create things we'd never seen before. That's the only path forward with AI. Create low-risk environments where you and your team can experiment without financial consequences. Ask Claude, Gemini, or ChatGPT weird questions about things you already know well. That's how you learn whether it's right, whether it's wrong, and where the edges really are.

▶️ Listen now

Bryan's final thought: We started inventing technology to help us, not replace us. Maybe it's time to remember that.