Why Do Breakthroughs Never 'Breakthrough'
Why major scientific breakthroughs take decades to matter - and why that delay is often mistaken for failure.
I see versions of these comment all the time.
“Another random innovation I’ll never hear about.”
On the surface, it sounds like cynicism. Or trolling. Or boredom. But it’s actually something more interesting than that.
Some technologies give you feedback in months.
Others take decades.
Some may never resolve inside a single human lifetime.
The image below is how I’ve come to think about that difference.
At the top is what I’d call Shallow Tech (but most people call just “tech”) – things like commercial software. Fast iteration cycles, short feedback loops, and relatively forgiving failure modes. You could build essentially the same kind of product in 1987 or 2025 on roughly the same timeframe (<3-4 years). Even with the advent of Ai that can automate large parts of the process the launch time from idea to deployment has moved from a few years to a few months. It’s quick.
This is partly because the underlying constraints haven’t changed much - as it’s where most modern tech culture, capital, and deployment are optimised to operate.
Then there is Deep Tech.
Here to even start the clock on a launch time it typically requires a breakthrough in our understanding about how the world around us works - a fundamental step forward that proves an idea is possible under ideal conditions. Usually a significant step forward in physics, chemistry, or biology. These are the breakthroughs you hear about on the news (or through me).
Early results are fragile, context-dependent, and often irreproducible outside the lab and years can be spent stress-testing the idea and turning a one-off demonstration into a system that behaves reliably, safely, repeatedly, and at scale.
Beyond the breakthrough, the world itself must also reconfigure. Manufacturing processes must be invented, supply chains built, and regulators convinced.
CRISPR is a good example. In 2012, CRISPR-Cas9 was proven as a programmable way to cut DNA, but cutting DNA is not the same as treating a patient. Every edit carried the risk of off-target mutations, large chromosomal rearrangements, or unpredictable repair outcomes, any of which would be permanent. Years of work were required to improve enzyme fidelity, understand DNA repair pathways, and develop sequencing methods sensitive enough to detect rare but dangerous errors. Before CRISPR could move anywhere near a clinic, it had to be transformed from a powerful laboratory tool into a controllable medical instrument.
The first CRISPR-based therapy was approved in 2023. That puts the discovery-to-therapy gap at ~11 years.
What separates Deep Tech from even deeper categories are Venture timelines. Most Venture Capital fund managers (myself included) operate on a 10 (+2) model, meaning they want to invest and see a return within a ~10 year period. It’s reasonably arbitrary, built as much on the sustainability of human attention over a lifetime and a seeing a return while you are still around to benefit from it. But it informs investment appetite for ideas and it makes them harder (but not impossible) to start if they are perceived to require longer timescales.
At the bottom is what I think of as Mythic Tech.
Fusion lives here. Not because it’s impossible, but because its timelines are so long that belief, narrative, and institutional endurance matter as much as technical progress and where significant discovery and breakthrough are still required.
In 1987, the year when software began flying off the walls, researchers in Japan were sequencing genes in E. coli when they noticed embedded in the genome were short, regularly repeating DNA sequences, separated by equally regular gaps.
The structure was strikingly ordered, but with no apparent function. The repeats were carefully documented and published, then largely set aside – an anomaly recorded without interpretation. These “Japanese repeat units” became recognised as part of what we now call CRISPR. 25 years later it produced the first therapeutic.
The mistake we keep making is not understanding where ideas fall on this spectrum, and treating all three layers as if they should behave like ‘tech’.
We expect Deep Tech to iterate in four-year cycles.
We expect Mythic Tech to justify itself on venture timelines.
And when that doesn’t happen, we call it failure.
What’s shifting now is our literacy. We’re getting better at distinguishing between ideas that scale through iteration and those that advance through endurance. Some technologies aren’t meant to sprint; they have to survive long enough for the evidence to compound and the world to accept them.
When we stop forcing every breakthrough onto startup timelines, fewer things look like failures – and more of them start to look inevitable.
See you next week,
Dr Ben