Focus is advice for people who have not found the overlap
ยท 6 min read
I get told to focus roughly once a week. It comes from people who mean well and who are usually right. Somebody at 22 running four companies looks like somebody who cannot finish things, and that pattern is common enough that the warning is a reasonable default.
I want to argue against it in my own case, and then give you the test that would prove me wrong, because an argument like this is worthless without one.
What it looks like from outside
On paper it reads badly. Biotech Wallah does cognitive screening. Healoncal does dermatology from a phone camera. Neurento builds audiometry hardware. Convolity does supplement and drug interaction reasoning. Four companies, four organ systems, four markets. Plus research posts on a cancer registry and on wearable physiological signals.
Described that way it sounds like someone collecting hobbies. If that were the real structure I would deserve the advice.
What it actually is
Those are not four problems. They are one problem appearing in four places, and the problem is this: get a measurement that a clinician will accept out of hardware the patient already owns, then make the output legible enough that acting on it is defensible.
Every one of those companies is a rerun of that sentence. Cognitive screening pulls voice, eye movement and task response off a stock phone. Dermatology pulls lesion features off the same camera. Audiometry is the odd one out only because a phone speaker cannot be trusted for a calibrated tone, which is exactly why that one grew hardware. Supplement interaction is the same shape moved from sensing to reasoning, where the input is a person's own health data instead of a sensor stream.
The parts that are genuinely hard are shared. Signal quality on uncontrolled consumer hardware. Per-metric explanation instead of a single score, because a clinician who cannot see why will not act. Deployment where bandwidth and device quality are not assumptions you get to make. Documentation written on the expectation that a regulator reads it later. I solve those once and they pay out four times.
Four companies is not four times the work when three quarters of the hard part is the same part.
The version of this I actually believe
I think my capacity for parallel load is the strongest thing I have, not a liability I am managing. It is the reason I can hold a clinical trial, a registry build, a hardware bring up and a review cycle in the same week without any of them stalling, and I would be worse at all four if I were only allowed one.
Some people go deep and some people go wide, and the wide ones get told for years that they are unserious until the moment the breadth turns into the whole advantage. I stopped treating it as something to apologise for.
The costs, which are real
Switching is not free. There is a warm up tax every time I move between a signal processing problem and a regulatory document, and on any single one of these I am slower than a founder who does nothing else. If you put me next to a person doing only dermatology AI, they will out ship me on dermatology AI. That is simply true.
The failure mode is also not the one people warn me about. Nobody drops a company. What actually slips is the unglamorous middle: the follow up email, the second revision, the tedious data cleaning nobody is waiting on. Interesting work pulls attention away from necessary work, and with four of them the necessary work has four places to hide.
So the honest accounting is that breadth costs me depth per unit and costs me follow through, and I buy shared infrastructure and shared learning with it. I think that trade is worth it. I would not pretend it is free.
The test
Here is how you tell whether you have an overlap or a collection. State the shared substrate in one sentence, without naming any of the projects.
If you can do it, the breadth is real and the focus advice is aimed at a version of you that does not exist. If you cannot, you have four things and a story you tell about them, and you should listen to the people telling you to pick.
Mine is the sentence three sections up. I can write it without mentioning a company name, and I can point at the same code and the same documents doing work in more than one of them. That is the whole basis of my claim, and if it stopped being true I would want somebody to say so.
What I would tell someone else
Default to focus. It is right most of the time, and the base rate on people who think they are the exception is not flattering.
But the advice gets given as a law when it is a heuristic, and heuristics have domains. Focus is what you recommend to someone who has not found the overlap yet. Once the overlap is real, insisting on one lane is just asking somebody to give up an advantage they already built.
If you want to check the substrate claim rather than take my word for it, the projects are listed on the front page with what each one actually shares.