Hey - It’s Nico.

Welcome to another Failory edition. Huge thanks to all 35,000+ of you who have been reading Failory this year!

This will be this year’s last edition. We’ll be back next year with more content on building products from scratch and step by step workflows you can copy to validate ideas fast, ship simple products, and automate growth. Our goal is to help you turn ideas into working products without hiring and to automate growth tasks that take a lot of time and budget. Stay tuned!

Let’s get into today’s issue. If you’re short on time, these are the 3 key things you should know:

• Luminar, a public company making Lidar sensors for self-driving cars, has filed for bankruptcy — learn more below

• The best growth tactics of 2025

• Microsoft released Trellis.2, an image-to-3D model generator — learn why this matters below

This season, give yourself the gift of compliance done right ^AD

Delve’s AI Copilot makes SOC 2, HIPAA, and ISO compliance smoother than ever during the holiday rush. Copilot reads your policies, gathers evidence automatically, completes tasks, and keeps every requirement moving so your team can stay focused on shipping and growth instead of audit prep.

The proof is magical.

• Lovable achieved SOC 2 in just 20 hours.

• WisprFlow closed Mercury and Superhuman in 1 month after accelerating compliance with Delve.

• 11x unlocked $1.2 million in ARR by reaching enterprise readiness faster.

If you want to wrap up the year with confidence and a clean audit path, Delve gives you everything you need to get compliant on time.

Use code MBXMAS for $1,500 off until December 31.

Book your demo now

This Week In Startups

🔗 Resources

The best growth tactics of 2025

Who cares about Mission and Vision?

The Founder’s Playbook for Performance Culture

Stop guessing. Start learning from real customers and accelerate your path to product-market fit. * 

📰 News

Apple opens up its App Store to competition in Japan

Luma releases a new AI model that lets users generate a video from a start and end frame

Google launches Gemini 3 Flash

OpenAI released a new image generation model: GPT Image 1.5

💸 Fundraising

Vibe-coding startup Lovable raises $330M at a $6.6B valuation

Edison raises $70M seed for its AI-doing-science platform

Call center chatbot startup PolyAI raises $86M at a $750M valuation

Nuclear startup Last Energy raises $100M for its steel-encased micro reactor

* sponsored

Fail(St)ory

Eyes for Cars

Luminar was supposed to be the lidar company that made self-driving cars safer.

Last week, it filed for Chapter 11 after layoffs, lawsuits, and losing the one customer that mattered. 

What Was Luminar:

Luminar Technologies built lidar sensors for autonomous cars. 

Most autonomous systems today lean heavily on cameras. They’re cheap, familiar, and good enough for lane keeping and basic driver assist. But cameras struggle when the sun is low, when headlights flare, or when you’re asking software to judge distance at highway speed.

Radar fills some of that gap. It works in fog and rain and can track objects far ahead, but it doesn’t see shape very well. A motorcycle, a car, and a guardrail can blur together.

Lidar sits in between. It sends out laser pulses and measures how long they take to come back, building a depth map of the road in front of the car. The problem is that they are expensive, power-hungry, and hard to manufacture at scale. This was what Luminar was trying to solve.

Investors bought into that bet. Luminar went public in 2020 through a reverse merger.  It signed Volvo, pulled in Mercedes-Benz, and quickly grew into a $3 billion company on expectation more than delivery.

Volvo was the anchor. Safety is its whole identity, and lidar fit neatly into that story. What started as a small 2020 order kept expanding. First tens of thousands of sensors. Then hundreds of thousands. By 2022, Luminar was planning around 1.1 million units.

That commitment reshaped the company. Luminar staffed up, locked in manufacturing capacity, and spent heavily to meet projected demand. The business assumed Volvo volumes were real, durable, and imminent.

They weren’t. Mercedes-Benz exited in late 2024, pointing to missed requirements. That left Luminar exposed. Volvo wasn’t just a customer anymore. It was the business.

Then Volvo walked too. It terminated the supply agreement, citing contract failures and supply chain risk. It also made clear it could ship advanced safety features without lidar. 

After that everything collapsed fast. Luminar cut roughly a quarter of its staff, the founder stepped down as CEO, and last week the board chose to file for Chapter 11 bankruptcy.

The Numbers:

• 🚗 Peak valuation: ~$3B after 2020 reverse merger

• 👥 Layoffs: ~25% of staff, second cut in one year

• ⚖️ Assets: $100M–$500M

• 🧾 Liabilities: $500M–$1B

Reasons for Failure: 

• Extreme customer concentration: Volvo wasn’t just a big customer. It was the business. When Mercedes exited, Luminar had no buffer, no second anchor to absorb the shock. A single termination turned an operating problem into an existential one.

• Scaling ahead of proof: Luminar made massive upfront investments based on projected demand, not shipped reality. Equipment, facilities, and headcount were built for volumes that depended on one customer staying happy forever. 

• Leadership instability and distraction: At the worst possible moment, the company lost its center. The founder-CEO abruptly stepped down in May after an internal ethics inquiry, then stayed on the board, leaving everyone guessing who was really in charge. Months later, he launched a new venture and tried to buy the company back from the outside, while Luminar was already in distress.

Why It Matters: 

• Hardware startups don’t get mulligans on delivery. If you miss once, customers quietly design you out. 

• Safety narratives don’t override supply chain risk. Automakers will choose “good enough and reliable” over “best and unpredictable” every time.

Trend

Trellis.2

We have all been glued to image and video models lately. Sora. Nano Banana, Veo. Image and video have gotten insanely good in the last year.

But another category has been quietly catching up, mostly out of the spotlight: AI-generated 3D models.

This week, Microsoft released Trellis.2. And no, this isn’t just another model drop. It’s one of those moments where you stop and think, okay, this is starting to actually work.

That’s why it’s worth talking about.

Why it Matters

• 3D is not niche anymore. If 3D generation gets cheap and reliable, it leaks into marketing, commerce, AR, and product validation. Not just games and VFX.

• This changes who can build with 3D. You no longer need a 3D artist for every experiment. That collapses time, cost, and iteration loops for founders.

• It turns one asset into a content engine. A single 3D model can power ads, PDPs, AR previews, and social clips. That kind of reuse doesn’t exist with flat images.

What is Trellis.2

At a high level, Trellis.2 takes an image and turns it into a usable 3D object. Not a screenshot of a render. An actual 3D asset you can light, rotate, and drop into other tools.

The key difference is quality where it counts. Generated objects don’t fall apart the moment you inspect them. Materials behave like materials. Shapes don’t collapse if they’re slightly weird or asymmetrical.

This feels like a DALL·E 2 moment for 3D. Not because it’s perfect, but because it crosses a threshold. Things that were painful or impossible a year ago are suddenly easy enough to try.

The Space

But Trellis.2 isn’t the only player here. This space is suddenly full of teams shipping real 3D generation products, and it’s moving fast.

• Tencent Hunyuan 3D: A “commercial-grade” 3D generation engine that Tencent is pushing globally, with enterprise API access via Tencent Cloud.

• Meshy: One of the most practical tools right now, with APIs for text-to-3D and image-to-3D that are clearly built for integration into real apps and workflows.

• Luma AI Genie: The text-to-3D model from Luma AI.

• Tripo AI: A 3D model generator that leans hard into speed and multiple input modes (text, single image, multiple images, doodles).

The Trend

The real shift here is not “AI can do 3D now.” It’s that 3D is turning into a cheap, repeatable input instead of a handcrafted output. That unlocks very specific businesses that didn’t pencil out before.

Here are a few that feel interesting to me:

• AR without the AR team: Most companies want AR previews. Almost none want to hire AR specialists. A startup that takes a few product photos and spits out AR-ready assets for web and mobile closes that gap. Furniture, home decor, appliances, even B2B equipment. 

• The ‘Figma for physical products’ layer: Early-stage founders mock software in Figma. Physical products don’t have that equivalent. With fast 3D generation, you can sketch a product, spin it, test reactions, and kill bad ideas early. No CAD. No manufacturing. Just validation.

•  3D cleanup tooling:  Generated 3D assets often break rules: too heavy, wrong formats, missing materials, bad geometry. A startup that automatically fixes or flags issues before assets hit production saves time and money.

Help Me Improve Failory

That's all for today’s edition.

Cheers,

Nico