Six Months That Beat Five Years
What changed was not the technology. It was who could use it.
From 2020 to 2025, I built sign language technology through two ventures. A 3D signing avatar deployed for Airbus, Thales, the French government, the Deaflympics. With Spokhand, an audience on social media watching the work happen in the open.
Real work, really shipped. And everything took months. A demo was a quarter. A corpus was a year. An iteration was a budget line. You had an idea in January and saw it running, maybe, in November.
Then the last six months of CLERC produced more than those five years combined. Smaller team. Almost no capital. Ten hours of structured sign language video from native Deaf signers, 600 videos released open source, and two working use cases on top: a browser demo that reads 30 ASL signs at up to 71% accuracy on signers it has never seen, and a learning platform built from the same corpus.
Idea on Monday, running the following week. That is the actual rhythm now.
The objection
Of course you went faster. The tools did not exist in 2020. That is not you, that is AI.
Fair. Except the tools arrived for everyone at the same time. Twenty dollars a month, in almost every country on Earth.
So look around. If access is universal, the acceleration should be too. It is not. Five-person teams are shipping what departments used to ship, while organizations with a hundred times the resources are presenting the same slides as last year.
The barrier was never access. It is usage.
Curiosity. The willingness to be bad at a tool for a week. Patience to iterate instead of judging from a distance. A real project to point it all at. That divide does not follow intelligence or funding. Well-funded, brilliant people are standing still right now, and stubborn ones with twenty dollars and an obsession are moving like a company of fifty.
Written English is a second language
Here is a part of this that gets almost no attention.
Sign language moves through the hands, the face, the body. That is my first language. But entrepreneurship is played in written text: investor emails, decks, contracts, cold outreach, the endless formality between an idea and a signed deal. For many Deaf people, written English is a second language, learned without ever hearing it.
AI is a text machine. Which means it lowered the cost of building a company for Deaf founders more than for almost anyone else. Nobody designed it for that. It happened anyway.
I feel it every day. It belongs on the record, because it is going to produce a generation of Deaf founders that the previous decade structurally filtered out.
What multiplied
AI built none of this on its own. It is a multiplier, and a multiplier needs something to multiply.
Five years of mistakes were already priced in. I knew which corpus designs collapse at scale, which partnerships drain a small team, which demos photograph well and change nothing. Experience is the list of mistakes you no longer repeat. AI lets you act on that list ten times faster.
The community was already there. Our signers are native. Our annotators are Deaf. When a hard question comes up about the language, gloss boundaries, a regional variant, whether a facial movement is grammar or expression, the answer is in the room. No amount of compute replaces that. Every team without it pays in silent data corruption.
The network was already there. Years in this field mean the path to a university partner, a signer, or a hard technical answer is one message long.
Multiply all three and six months look like five years. Multiply zero and you get zero.
Day to day it looks unglamorous. Code written and reviewed in hours instead of weeks. Documents drafted, torn apart, rebuilt the same afternoon. Experiments cheap enough to run instead of debate. Speed is not one magic moment. It is a hundred small frictions removed every week.
0.1%
CLERC is at roughly 0.1% of the goal.
The goal is not a demo and not a product. It is the Deaf Digital Heritage: a structured, permanent corpus of the world's sign languages, built by the communities who carry them. One language at a time, one country at a time. Sign language preserved as living data, not archive footage.
Around 70 million people sign. Count the families, friends, and colleagues who live alongside them and you reach half a billion. Built from the Deaf community. Built for everyone.
The next piece of this series is about where that data comes from, and why the community producing it is the point rather than a supplier. The one after that is harder to write. It is about what worries me, and it is not what AI does to jobs or to interpreters. It is what AI could quietly do to the language itself.
The revolution will still be here in September. The only question is which speed you pick.
Follow @CLERC to track the build.