I've spent nearly forty years watching projects die and I've noticed something that never quite gets said out loud in the post-mortems: the failure is rarely the whole story. Sometimes the wreckage contains the best thing you'll ever build.
The Verge has a good piece on exactly this happening at Apple. Project Titan, the self-driving car that burned through an estimated ten billion dollars and got cancelled in 2024, apparently left behind the neural processing architecture that now underpins the AI chips in current iPhones and Macs. Apple needed silicon that could process huge volumes of sensor data in real time, safely, on-device, with almost no latency and no cloud dependency — because a car that phones home to a server before deciding whether to brake is a car that kills someone. That's an extraordinarily hard engineering problem. Harder, in some ways, than making a chatbot sound clever.
The car failed. The chip didn't.
When Titan got shut down, the obvious narrative was "Apple wasted a decade and billions of dollars." And on the balance sheet, sure. But the on-device neural engine work didn't get binned with the car — it got absorbed into the core silicon team and repurposed. The same low-latency, privacy-first, no-cloud-needed processing that was meant to stop a two-tonne vehicle from hitting a pedestrian is now quietly running Face ID, computational photography, and on-device language model inference on the phone in your pocket.
This is the thing about hard R&D: you very rarely know in advance which bit of it is the actual asset. Apple thought it was building a car. It was actually building the on-device AI stack that would let it credibly compete in a market — generative AI — that didn't even exist as a consumer category when Titan started.
Why this matters if you're not Apple
I don't have ten billion dollars of R&D budget lying about in Worcestershire, and neither do most of the people reading this. But the pattern scales down fine.
When I built the first version of RSSMasher years ago, the actual goal was a fairly narrow content-aggregation tool. What came out of solving the boring, unglamorous plumbing problems — reliable feed parsing, deduplication, rewriting content at scale without it reading like a spam bot wrote it — ended up being reusable in ways I hadn't planned. Bits of that plumbing are still running, in different clothes, inside MarketMasher and Article2Video today. The "failed" first version of the product wasn't a failure at all. It was the R&D phase for everything that came after, I just didn't label it that way at the time.
Indie builders do this constantly without noticing. You set out to solve problem A, you fail at A, but you've accidentally built a solid, general-purpose solution to problem B, which turns out to matter more. The trick — and it is a trick, because hindsight makes it look obvious and foresight never does — is noticing what you've actually built rather than mourning what you didn't.
The lesson isn't "never give up"
I want to be careful here, because the tempting takeaway is some limp motivational poster about persistence. That's not it. Apple was right to kill the car. Ten years and ten billion is a brutal amount to spend on something that never generates a pound of revenue, and a company with infinite resources still eventually has to say "enough." The lesson isn't that failure is secretly success if you squint. It's that hard technical problems, tackled properly, tend to generate transferable capability even when the original target turns out to be wrong.
That has a practical implication for how you structure R&D, even at a small scale: build the hard, general infrastructure first, and treat the specific product as a hypothesis sitting on top of it. If the hypothesis is wrong, you haven't wasted the infrastructure — you've just proven you need a different product to put on it. That's a much cheaper way to fail.
If you're building anything ambitious right now — an app, an AI tool, a SaaS product that's fighting you every step of the way — it's worth asking honestly what you're actually accumulating underneath the thing that isn't working. There's a decent chance the real asset isn't the car. It's the chip.
— Wayne