ExpertsFMThe Future Of Brain-Computer Interfaces with Science's Max Hodak
Y Combinator Startup Podcast · March 9, 2026
5 Selected Lessons:
priority · 2 framework · 1 reasoning_pattern · 1 heuristic · 1
BCI as a category, not a product
Source
Y Combinator Startup Podcast — The Future Of Brain-Computer Interfaces with Science's Max Hodak (March 9, 2026) · at 0:00 ▸ · at 3:49 ▸ · at 37:17 ▸
When this applies
when forecasting how a nascent deep-tech category will mature into products
The takeaway
Treat an emerging deep-tech category as a portfolio of differentiated approaches matched to differentiated applications, not as a race to one dominant product.
Why it works
Different neural targets, modalities, and risk/benefit profiles are suited to fundamentally different applications, so no single device design can serve all use cases — analogous to how pharma isn't one drug.
One level up
Resist collapsing a multi-application frontier technology into a single 'winning product' narrative too early.
Where it stops holding
This stops working when one modality turns out to dominate across nearly all use cases on cost/performance, collapsing the category into a near-single-product market.
Detail
hierarchy:
- Different modalities for different applications (ultrasound for 'digital ambien/adderall', electrical for retinal prosthesis, biohybrid for ultra-high-bandwidth/brain-to-brain applications)
- Biohybrid reserved for highest-end applications, backloaded relative to scalable near-term impact
- Scalable, lower-risk approaches (electrical stimulation, cochlear implants, motor decoding) will be deployed first and more broadly
rationale: Mirrors pharma's structure: a category of many specialized products rather than one universal solution.
In their words
I think that BCI we're going to come to see is not a specific product in the way that like pharma is not a product. I think there are going to be a bunch of BCI companies going after different applications where different types of probes will make sense.
Figure out what you want, then be relentlessly high-agency toward it
Source
Y Combinator Startup Podcast — The Future Of Brain-Computer Interfaces with Science's Max Hodak (March 9, 2026) · at 47:04 ▸
When this applies
when a person (especially early-career) wants to break into a field with no obvious entry point
The takeaway
Career progress in hard, gatekept fields requires two sequential things: first clarity about the specific goal, then sustained, resourceful persistence (finding backdoors, unconventional routes) to get placed where that goal requires you to be.
Why it works
Having a clear target lets opportunistic persistence pay off (finding unconventional entry points like an independent study backdoor into a lab); without the clear target, high agency has nothing to aim at.
One level up
Agency is necessary but not sufficient — it must be paired with a clearly chosen direction, or it has nothing to compound against.
Where it stops holding
This stops working when the chosen goal itself is poorly matched to reality (no viable path exists) or when persistence crosses into working in isolation without leveraging others (see the separate Transcriptic/Neuralink lesson).
Detail
hierarchy:
- First: figure out specifically what you want (Hodak: work on brain-computer interfaces)
- Then: be high-agency and persistent about placing yourself there even through unconventional routes (e.g. sneaking into a lab via an independent study in another department)
rationale: Hodak attributes most of his real education and head start to persistence in getting into the right lab early, which 'only works if you have a sense of where you want to go.'
In their words
I had a sense of what I wanted, and then I was very high agency towards that.
First-principles quadrant exploration before committing to a technical approach
Source
Y Combinator Startup Podcast — The Future Of Brain-Computer Interfaces with Science's Max Hodak (March 9, 2026) · at 21:37 ▸
When this applies
when starting a deep-tech company without a pre-baked technical asset and needing to choose among several plausible technical approaches
The takeaway
When no existing asset dictates your starting point, enumerate the full space of plausible technical approaches as a structured matrix, explore each quadrant for feasibility, and let empirical findings (not initial intuition) narrow the field — but in domains shaped by evolution, first-principles reasoning alone is insufficient and must be checked against what the biological/physical system actually does.
Why it works
By explicitly mapping out the full combinatorial space of plausible approaches (e.g. a 2x2 matrix of target cell type x stimulation method) and empirically evaluating each, a team avoids anchoring on whatever asset happened to exist and instead converges on the approach actual evidence supports — while still respecting that biology requires empirical/evolutionary understanding, not pure first-principles engineering reasoning alone.
One level up
Decouple company formation from asset-anchoring: build around a problem-space and a method of search, not around whichever IP happens to already exist.
Where it stops holding
This stops working when time/capital constraints don't allow exploring multiple quadrants, or in domains (like software) where first-principles reasoning alone is sufficient and there's no 'what did evolution actually do' nuance to check against.
Detail
procedure:
- Start from a broad thesis (neural-engineering view of healthcare) rather than a specific pre-existing asset.
- Identify near-term highest-value target (retinal prosthesis) as feasible 'circa now'.
- Lay out the full matrix of plausible technical approaches (cell type x stimulation modality).
- Explore all quadrants rather than committing early.
- Use empirical/evolutionary evidence to eliminate quadrants (e.g. optic-nerve stimulation requires an impractical million-degree-of-freedom per-patient calibration).
- Develop the two most promising paths in parallel (optogenetic in-house; electrical via acquisition) and commit resources once evidence narrows the field.
rationale: In biology, first principles reasoning gets you far but not all the way — you also need empirical/evolved understanding, so systematic exploration beats pure deduction or asset-anchoring.
In their words
I think we took this very, very like first principles approach. And you have to be careful with first principles in biology because first principles are not enough in biology.
Nature-inspired design: 'how would evolution solve the new-cable problem'
Source
Y Combinator Startup Podcast — The Future Of Brain-Computer Interfaces with Science's Max Hodak (March 9, 2026) · at 28:51 ▸
When this applies
when designing a novel high-bandwidth interface between two complex systems and lacking an existing standard connector
The takeaway
When you need a connector/interface for which no engineered standard exists, look for an existing biological (or otherwise evolved) analog that already solved a structurally similar integration problem, and copy its mechanism rather than inventing a de novo artificial one.
Why it works
Biological systems have already solved analogous high-bandwidth integration problems (e.g. the corpus callosum unifying two brain hemispheres into one experience) via growing living, self-wiring tissue rather than a rigid engineered connector; mimicking that process (seeding living neurons that grow and integrate) can achieve integration depths that a rigid wire-based approach cannot.
One level up
Evolution is frequently a better engineer than a human team for biological integration problems — use it as a design search method, not just an inspiration metaphor.
Where it stops holding
This stops working when there's no good biological analog to borrow from, or when the biological mechanism's side effects (e.g. uncontrolled growth) are unacceptable for the application, which Hodak himself flags as an open risk.
Detail
approach: Ask 'if nature wanted a new cranial nerve/ultra-high-bandwidth connection, how would it build it?' -> grow living neural tissue that wires itself into the host brain, rather than place rigid wires directly into brain tissue.
signals:
- corpus callosum as existence proof of high-bandwidth integration via biological growth
- lab observation that cultured neurons placed on neurons spontaneously wire together
rationale: Avoids genetically modifying the host brain (a one-way-door risk) and avoids wires directly in brain tissue, trading that for a different risk: unconstrained biological growth, observed in animal models but not yet in humans.
In their words
If nature wanted to build an ultra-high bandwidth brain-to-brain connection, like what would, how would it, or if you wanted to make a new cranial nerve? So instead of having an optic nerve or a vestibular cochlear nerve, it wanted to have like the internet nerve. Like, how would nature solve this problem? Is it would grow like a new nerve.
Event horizon of predictability: near-term forecastable, long-term opaque
Source
Y Combinator Startup Podcast — The Future Of Brain-Computer Interfaces with Science's Max Hodak (March 9, 2026) · at 24:27 ▸ · at 50:25 ▸
When this applies
when trying to forecast the future of a fast-compounding technology field
The takeaway
In any domain undergoing compounding technological change, trust your forecasting ability only up to a near-term horizon; beyond that horizon, treat the future as genuinely unknowable rather than extrapolating confidently.
Why it works
In a field undergoing exponential/compounding change, near-term trajectories remain extrapolable from current trends, but compounding effects make outcomes beyond a certain horizon genuinely unknowable — not just hard to guess, but structurally opaque, akin to how people living just before the Industrial Revolution could not have imagined the changes of its first 10-15 years.
One level up
Calibrate the confidence of any forecast to the compounding rate of the underlying technology, not to how far out the prediction nominally reaches.
Where it stops holding
This stops working if the technology's rate of change decelerates or plateaus, in which case longer-horizon forecasting becomes reliable again.
Detail
rule: Treat the next few years as forecastable, but treat anything beyond roughly a decade out, during a period of compounding technological change, as unforecastable.
threshold: Hodak's personal 'event horizon' is 2035 — 'I can't see past it.'
rationale: Analogized to people at the dawn of the Industrial Revolution being unable to imagine the changes of the steam engine's first 10-15 years.
In their words
by 2035, it's just like impossible. There's like, I can't see past it.