As a long time reader who is deeply interested in neurotech I loved this post! A few thoughts
1. You are 100% correct in saying that many of the results from companies like Neuralink or BlackRock Neurotech can be done without touching the Central Nervous System, with simple devices like the Mouthpad. This is something I quickly realized a few years ago when I was doing research on Neuralink. I'm conflicted on whether or not to count this as a point against Neuralink. On one hand, products like Telepathy weren't made to help locked-in users move a cursor but were made as a stepping stone to advance the company towards a "Holy Grail" device that lets humans merge with AIs. This focus on the "Holy Grail" is why things like the butcher number matter - though killing a few thousand neurons now doesn't really matter, eventually we'd want to scale from just a few thousand channels to hundreds or even millions of channels and it won't do to be killing a significant portion of the brain's total neurons. On the other hand, the company had a top tier team, billions of VC funding, and a decade to work with, and hasn't delivered better outcomes than a tonguepad / eye-tracking device could accomplish. Though to be fair, they are now working on a device that will attempt to restore vision in blind patients.
2. Though Nudge's approach of using Low-Intensity Focused ultrasound is a few orders of magnitude less precise than Forest Neurotech's approach, it not only several orders of magnitude more precise than basically every other non-invasive neuromodulation method, but is the only non-invasive neuromodulation technique that can reach deep into the brain. Why is this significant? You mentioned that using deep brain stimulation (an expensive and invasive procedure that involves physically inserting electrodes next to the desired target region), on theorized depression circuits, researchers burned tens of millions of dollars to achieve null results. With a Nudge-like device using LIFU, it would be possible to run such trials with a fraction of the time and cost, due to Low-Intensity Focused Ultrasound's non-invasiveness and relative safeness. A Nudge-Like device will significantly reduce the barriers to not only run such studies, but to map the entire brain.
3. You're definitely right about the market need for many modern neurotech devices. On one hand you have the medical grade devices from companies like BlackRock/Neuralink/Synchron/Precision Neurosciences, which serve a desperate but small population. As you've mentioned before, many of these people could be served equally well with a device that interacts with the Peripheral Nervous system instead. On the other hand you have the "Wellness" devices that are often just not very good.... There are so many EEG companies that are forever reaching but falling short of Product Market Fit. I don't know much about the Kernel Flow but it doesn't sound like a hyper successful consumer product. Yet, there are companies that have seen a lot of success in this space like Mouthpad and CNTRL-Labs, and two others whom I think will do VERY well in the consumer market if the product works: Orbit and Prophetic.
Anyways, loved the article. Your writing is some of the best on substack.
I thoroughly enjoyed the post! It comes at a time where I have spent quite some time thinking about neurotech and so there is a lot on my mind...
Part of the beauty is how you approach the space in the first place It seems many folks in the field make quite a few assumptions and continue about their work, and here you stop to say "Hey wait a minute, but what does this mean?"
---
After discussing your framework with some industry veterans (including folks working on next-gen ASICs and implants), I think there’s a 'hidden' sixth question that bridges your points on burden and physics:
1. The 'Delta of Effort' vs. Meaningful Utility
As one veteran pointed out to me, the ultimate test isn't just 'Does it work?' but 'Does it work meaningfully better than a mouthstick or eye-gaze?' We often get obsessed with the bandwidth of the signal (the physics), but we under-appreciate the cognitive and physical effort required to maintain that signal. If a non-invasive EEG or EMG device requires 2x the mental concentration for 1.1x the output of a simple mechanical interface, it fails the preference test every time.
Ultimately, a device must not only 'earn its place' in a patient’s life but it must do so with a lower 'tax' on their remaining energy than the existing hacks they’ve already mastered.
---
Your point about the 'relevancy of state measurements' is spot on, but I’m seeing an interesting counter-thesis emerge. While the 'physics' suggests EEG is a low-res mess (the 'fighting the skull' problem), I ask, what if we haven't actually hit the ceiling of EEG?
2. Challenging the 'EEG Ceiling'
If we treat scientific disciplines as artificial silos, we miss the fact that advanced ML models might unlock higher-dimensional insights from what we currently dismiss as 'raw noise.' We might be able to bypass some of the 'butcher number' dilemmas of invasive tech if we can extract 'invasive-grade' intent from 'non-invasive' signals through sheer computational brute force.
---
I’m glad you mentioned heat in the footnotes. In my recent conversations with founders, thermal management is becoming the real 'final boss' of invasive neurotech. The FDA’s 1 degree Celcius limit is a brutal constraint.
3. The Thermal Constraint: The Hardest Ceiling
It doesn't matter how many channels Neuralink or a startup like Mint Neuro can pack onto a chip if the power draw cooks the surrounding tissue. This makes the 'physics' question even more pointed: Are you fighting the skull (non-invasive), or are you fighting the Second Law of Thermodynamics (invasive)?
---
Just some thoughts, and I had no where else to dump them, but overall loved the post and definitely had a lot of fun reading it!
Really well written, and don't sell yourself short -- many of your points are things neurotech founders should really think more about.
I will say, don't be so certain about EEG's ability/inability to do useful things. My company Synaptrix is focused on brand new DL models for understanding the brain through EEG and demonstrating real world scaling laws for neural data (and have yet to find a performance plateau). It's really quite an incredible demo.
> And yet despite the insane lack of specificity, ECT remains the single most effective treatment we have for severe, treatment-resistant depression. Response rates hover around 50-70% in patients for whom nothing else has worked, with some rather insane outcomes, one review paper stating: “For the primary outcome of all-cause mortality, ECT was associated with a 30% reduction in overall mortality.”
TMS is probably better at this point (eg https://pubmed.ncbi.nlm.nih.gov/32252538/) and definitely much safer. Also probably some evidence in favor of Nudge approach since it's through the skull and not super precise.
The Meta Neural Band was released commercially at the end of September, 2025, as the main controller for the Meta RayBan Display smartglasses. 8-channel EMG; 18h battery life, on-device ML for gesture decoding. Band only available for purchase as a set for $799, but a replacement band is $199 (for cost reference).
Meta has ongoing academic collaborations (CMU, Utah, ...) supporting accessibility applications for various patient groups, including those with spinal cord injury. Preprint hot off the press: https://www.biorxiv.org/content/10.64898/2026.01.09.698484v1
How about the use of gene therapies such as opto and chemogenetics for treating neurological diseases/disorders? Personally I think that their uptake in the next 20 years will be far higher than BCIs.
As a long time reader who is deeply interested in neurotech I loved this post! A few thoughts
1. You are 100% correct in saying that many of the results from companies like Neuralink or BlackRock Neurotech can be done without touching the Central Nervous System, with simple devices like the Mouthpad. This is something I quickly realized a few years ago when I was doing research on Neuralink. I'm conflicted on whether or not to count this as a point against Neuralink. On one hand, products like Telepathy weren't made to help locked-in users move a cursor but were made as a stepping stone to advance the company towards a "Holy Grail" device that lets humans merge with AIs. This focus on the "Holy Grail" is why things like the butcher number matter - though killing a few thousand neurons now doesn't really matter, eventually we'd want to scale from just a few thousand channels to hundreds or even millions of channels and it won't do to be killing a significant portion of the brain's total neurons. On the other hand, the company had a top tier team, billions of VC funding, and a decade to work with, and hasn't delivered better outcomes than a tonguepad / eye-tracking device could accomplish. Though to be fair, they are now working on a device that will attempt to restore vision in blind patients.
2. Though Nudge's approach of using Low-Intensity Focused ultrasound is a few orders of magnitude less precise than Forest Neurotech's approach, it not only several orders of magnitude more precise than basically every other non-invasive neuromodulation method, but is the only non-invasive neuromodulation technique that can reach deep into the brain. Why is this significant? You mentioned that using deep brain stimulation (an expensive and invasive procedure that involves physically inserting electrodes next to the desired target region), on theorized depression circuits, researchers burned tens of millions of dollars to achieve null results. With a Nudge-like device using LIFU, it would be possible to run such trials with a fraction of the time and cost, due to Low-Intensity Focused Ultrasound's non-invasiveness and relative safeness. A Nudge-Like device will significantly reduce the barriers to not only run such studies, but to map the entire brain.
3. You're definitely right about the market need for many modern neurotech devices. On one hand you have the medical grade devices from companies like BlackRock/Neuralink/Synchron/Precision Neurosciences, which serve a desperate but small population. As you've mentioned before, many of these people could be served equally well with a device that interacts with the Peripheral Nervous system instead. On the other hand you have the "Wellness" devices that are often just not very good.... There are so many EEG companies that are forever reaching but falling short of Product Market Fit. I don't know much about the Kernel Flow but it doesn't sound like a hyper successful consumer product. Yet, there are companies that have seen a lot of success in this space like Mouthpad and CNTRL-Labs, and two others whom I think will do VERY well in the consumer market if the product works: Orbit and Prophetic.
Anyways, loved the article. Your writing is some of the best on substack.
Thank you so much for the informative comments and kind words!! I'll check out Orbit and Prophetic, I hadn't heard of either of those companies before
I thoroughly enjoyed the post! It comes at a time where I have spent quite some time thinking about neurotech and so there is a lot on my mind...
Part of the beauty is how you approach the space in the first place It seems many folks in the field make quite a few assumptions and continue about their work, and here you stop to say "Hey wait a minute, but what does this mean?"
---
After discussing your framework with some industry veterans (including folks working on next-gen ASICs and implants), I think there’s a 'hidden' sixth question that bridges your points on burden and physics:
1. The 'Delta of Effort' vs. Meaningful Utility
As one veteran pointed out to me, the ultimate test isn't just 'Does it work?' but 'Does it work meaningfully better than a mouthstick or eye-gaze?' We often get obsessed with the bandwidth of the signal (the physics), but we under-appreciate the cognitive and physical effort required to maintain that signal. If a non-invasive EEG or EMG device requires 2x the mental concentration for 1.1x the output of a simple mechanical interface, it fails the preference test every time.
Ultimately, a device must not only 'earn its place' in a patient’s life but it must do so with a lower 'tax' on their remaining energy than the existing hacks they’ve already mastered.
---
Your point about the 'relevancy of state measurements' is spot on, but I’m seeing an interesting counter-thesis emerge. While the 'physics' suggests EEG is a low-res mess (the 'fighting the skull' problem), I ask, what if we haven't actually hit the ceiling of EEG?
2. Challenging the 'EEG Ceiling'
If we treat scientific disciplines as artificial silos, we miss the fact that advanced ML models might unlock higher-dimensional insights from what we currently dismiss as 'raw noise.' We might be able to bypass some of the 'butcher number' dilemmas of invasive tech if we can extract 'invasive-grade' intent from 'non-invasive' signals through sheer computational brute force.
---
I’m glad you mentioned heat in the footnotes. In my recent conversations with founders, thermal management is becoming the real 'final boss' of invasive neurotech. The FDA’s 1 degree Celcius limit is a brutal constraint.
3. The Thermal Constraint: The Hardest Ceiling
It doesn't matter how many channels Neuralink or a startup like Mint Neuro can pack onto a chip if the power draw cooks the surrounding tissue. This makes the 'physics' question even more pointed: Are you fighting the skull (non-invasive), or are you fighting the Second Law of Thermodynamics (invasive)?
---
Just some thoughts, and I had no where else to dump them, but overall loved the post and definitely had a lot of fun reading it!
I am cohosting this meetup! Looking forward to seeing you all!
Really well written, and don't sell yourself short -- many of your points are things neurotech founders should really think more about.
I will say, don't be so certain about EEG's ability/inability to do useful things. My company Synaptrix is focused on brand new DL models for understanding the brain through EEG and demonstrating real world scaling laws for neural data (and have yet to find a performance plateau). It's really quite an incredible demo.
> And yet despite the insane lack of specificity, ECT remains the single most effective treatment we have for severe, treatment-resistant depression. Response rates hover around 50-70% in patients for whom nothing else has worked, with some rather insane outcomes, one review paper stating: “For the primary outcome of all-cause mortality, ECT was associated with a 30% reduction in overall mortality.”
TMS is probably better at this point (eg https://pubmed.ncbi.nlm.nih.gov/32252538/) and definitely much safer. Also probably some evidence in favor of Nudge approach since it's through the skull and not super precise.
Great article!
The Meta Neural Band was released commercially at the end of September, 2025, as the main controller for the Meta RayBan Display smartglasses. 8-channel EMG; 18h battery life, on-device ML for gesture decoding. Band only available for purchase as a set for $799, but a replacement band is $199 (for cost reference).
Meta has ongoing academic collaborations (CMU, Utah, ...) supporting accessibility applications for various patient groups, including those with spinal cord injury. Preprint hot off the press: https://www.biorxiv.org/content/10.64898/2026.01.09.698484v1
How about the use of gene therapies such as opto and chemogenetics for treating neurological diseases/disorders? Personally I think that their uptake in the next 20 years will be far higher than BCIs.
Nice post! Can you tell more about science.xyz in this context?