Support this publication to show you appreciate and believe in them. As their writing reaches more readers, your coins may grow in value.
Top supporters
1.
0xcbab...dc70
43.7M
2.
0x65de...c951
35.9M
3.
AVC
18.5M
4.
feyth.eth
17M
5.
cryptoartist
10.5M
6.
0x0d6b...5ec0
10.2M
7.
0xfb7c...b627
10M
8.
0x1081...68d2
9.8M
9.
0xc4b7...03bf
9.5M
10.
0x190b...34c5
7.7M
Subscribe
Support AVC
Support this publication to show you appreciate and believe in them. As their writing reaches more readers, your coins may grow in value.
Top supporters
1.
0xcbab...dc70
43.7M
2.
0x65de...c951
35.9M
3.
AVC
18.5M
4.
feyth.eth
17M
5.
cryptoartist
10.5M
6.
0x0d6b...5ec0
10.2M
7.
0xfb7c...b627
10M
8.
0x1081...68d2
9.8M
9.
0xc4b7...03bf
9.5M
10.
0x190b...34c5
7.7M
AVC
Jan 8
Live Captions
I picked up my new Meta "Display" smartglasses a few days ago and have been playing with them a bit since. The one feature that feels really important and powerful to me is "live captions." Most people are familiar with closed captioning on TVs and in theaters, where the speech is translated into text and shown at the bottom of the screen. The Display smart glasses have this feature on the lower portion of the lenses. You can turn it on to understand someone speaking your native language bett...
When USV commits to investing in a startup, we negotiate a term sheet and then hand over the details to our lawyers. The startup hires a lawyer, and we hire a lawyer. The startup's lawyer prepares the closing documents, and our lawyer reviews them. In addition, our lawyer conducts "legal due diligence," which primarily involves reviewing existing contracts, stock issuances, the charter, and other relevant legal documents.
This process is expensive and made worse because the startup typically pays for both lawyers.
This is how it has always been done since I started in the business in the mid-80s, and I have always been uncomfortable with how expensive it is.
So I decided to run an experiment over the holidays.
We committed to lead a round of financing for a company in mid-December. We negotiated a term sheet and signed it before everyone departed for the holidays. I reached out to a law firm that I have used many times for this sort of thing and asked for a quote to handle our side of the deal. The quote came back at $50k.
So I said, "screw it" and decided it was time to try something different.
I fired up Google's NotebookLM, which allows users to create "notebooks," which are a large collection of documents that can then be used to run AI queries.
I put a large collection of "closing binders" of investments USV has made over the years, particularly companies I worked on, into one Notebook. I added the signed term sheet to this Notebook.
To create a second Notebook, I pointed NotebookLM at the data room that the startup we are investing in provided for legal diligence. That data room had every legal document the startup had entered into, including those with its employees, since it got started.
When we received the draft closing documents from the startup's lawyer, I added them to the first Notebook and asked for a legal review of the draft documents against the body of legal documents we have signed over the years, and most importantly, against the term sheet we had signed. I asked for a memo that outlined all of the issues with the draft documents and highlighted the most significant ones.
I then turned to the second Notebook and asked a series of questions like "tell me about the structure of the company and its subsidiaries and who is on the board of each of them" and "give me a list of every employee, the stock they have been issued, and all of the agreements they have signed" and "are their arbitration clauses in every agreement the company has signed?" I spent about half an hour asking these sorts of questions and put the answers to each into a Google Doc.
1/ Gemini passes ChatGPT in terms of DAUs, MAUs, and tokens consumed in the first half of 2026, making Google the king of AI.
2/ The Democrats take control of the House in the Nov 2026 elections, bringing to an end Trump's complete control of the US government.
3/ Smartglasses finally reach product market fit in 2026, but it won't be Meta that delivers the winning approach.
4/ A majority of venture capital deals close without lawyers on either side due to standardized documents (like the NVCA ones) and AI tools for review and legal diligence.
5/ A drunk driver using full self-driving legally challenges a DUI charge and wins.
6/ The SpaceX IPO, the largest ever, marks the market top, and we are in a bear market by year's end.
7/ Despite the loss of tax incentives, EVs and solar adoption significantly inflects in the US, reflecting better economics and utility for consumers and businesses.
8/ Blockchains disappear behind better consumer interfaces that allow users to use, spend, trade, and send tokens without concerning themselves with which blockchain they are on.
9/ An AI-generated song is nominated for Song of the Year in the 2027 Grammy nominations.
I like to bookend the New Year holiday with two posts, one looking back at the year that is ending (What Happened) and one looking forward to the year ahead (What Will Happen). This is the first of these two posts. The second one will run tomorrow.
Here's a top ten list of things that happened this year that really matter.
1/ The end of globalization. The Trump Tariffs represent a fundamental change in trade policy, and if they remain the approach of future White Houses, it will mark a return to protectionism and the end of the Free Trade era in the US, which has been the default policy of the US for my entire adult life. Whether you are for or against this change, it is massive and means that the rest of the world will now have to pay to access our markets in the US, making the reshoring of critical infrastructure possible. One thing I am less sure about but super interested in is whether tariffs can become a significant revenue generator for the US Government, as was the case until 1913 with the introduction of the modern federal income tax system. If tariff income can make a significant reduction or elimination of the federal budget deficit in the US, then that would be another major economic shift (away from deficit spending).
2/ Google got its mojo back. From Waymo robotaxis running wild on the streets of many of the most populous cities in the US to Gemini coming hard after ChatGPT, 2025 was a banner year for Google/Alphabet and a reminder that owning our data is key to owning us.
AVC
Jan 8
Live Captions
I picked up my new Meta "Display" smartglasses a few days ago and have been playing with them a bit since. The one feature that feels really important and powerful to me is "live captions." Most people are familiar with closed captioning on TVs and in theaters, where the speech is translated into text and shown at the bottom of the screen. The Display smart glasses have this feature on the lower portion of the lenses. You can turn it on to understand someone speaking your native language bett...
When USV commits to investing in a startup, we negotiate a term sheet and then hand over the details to our lawyers. The startup hires a lawyer, and we hire a lawyer. The startup's lawyer prepares the closing documents, and our lawyer reviews them. In addition, our lawyer conducts "legal due diligence," which primarily involves reviewing existing contracts, stock issuances, the charter, and other relevant legal documents.
This process is expensive and made worse because the startup typically pays for both lawyers.
This is how it has always been done since I started in the business in the mid-80s, and I have always been uncomfortable with how expensive it is.
So I decided to run an experiment over the holidays.
We committed to lead a round of financing for a company in mid-December. We negotiated a term sheet and signed it before everyone departed for the holidays. I reached out to a law firm that I have used many times for this sort of thing and asked for a quote to handle our side of the deal. The quote came back at $50k.
So I said, "screw it" and decided it was time to try something different.
I fired up Google's NotebookLM, which allows users to create "notebooks," which are a large collection of documents that can then be used to run AI queries.
I put a large collection of "closing binders" of investments USV has made over the years, particularly companies I worked on, into one Notebook. I added the signed term sheet to this Notebook.
To create a second Notebook, I pointed NotebookLM at the data room that the startup we are investing in provided for legal diligence. That data room had every legal document the startup had entered into, including those with its employees, since it got started.
When we received the draft closing documents from the startup's lawyer, I added them to the first Notebook and asked for a legal review of the draft documents against the body of legal documents we have signed over the years, and most importantly, against the term sheet we had signed. I asked for a memo that outlined all of the issues with the draft documents and highlighted the most significant ones.
I then turned to the second Notebook and asked a series of questions like "tell me about the structure of the company and its subsidiaries and who is on the board of each of them" and "give me a list of every employee, the stock they have been issued, and all of the agreements they have signed" and "are their arbitration clauses in every agreement the company has signed?" I spent about half an hour asking these sorts of questions and put the answers to each into a Google Doc.
1/ Gemini passes ChatGPT in terms of DAUs, MAUs, and tokens consumed in the first half of 2026, making Google the king of AI.
2/ The Democrats take control of the House in the Nov 2026 elections, bringing to an end Trump's complete control of the US government.
3/ Smartglasses finally reach product market fit in 2026, but it won't be Meta that delivers the winning approach.
4/ A majority of venture capital deals close without lawyers on either side due to standardized documents (like the NVCA ones) and AI tools for review and legal diligence.
5/ A drunk driver using full self-driving legally challenges a DUI charge and wins.
6/ The SpaceX IPO, the largest ever, marks the market top, and we are in a bear market by year's end.
7/ Despite the loss of tax incentives, EVs and solar adoption significantly inflects in the US, reflecting better economics and utility for consumers and businesses.
8/ Blockchains disappear behind better consumer interfaces that allow users to use, spend, trade, and send tokens without concerning themselves with which blockchain they are on.
9/ An AI-generated song is nominated for Song of the Year in the 2027 Grammy nominations.
I like to bookend the New Year holiday with two posts, one looking back at the year that is ending (What Happened) and one looking forward to the year ahead (What Will Happen). This is the first of these two posts. The second one will run tomorrow.
Here's a top ten list of things that happened this year that really matter.
1/ The end of globalization. The Trump Tariffs represent a fundamental change in trade policy, and if they remain the approach of future White Houses, it will mark a return to protectionism and the end of the Free Trade era in the US, which has been the default policy of the US for my entire adult life. Whether you are for or against this change, it is massive and means that the rest of the world will now have to pay to access our markets in the US, making the reshoring of critical infrastructure possible. One thing I am less sure about but super interested in is whether tariffs can become a significant revenue generator for the US Government, as was the case until 1913 with the introduction of the modern federal income tax system. If tariff income can make a significant reduction or elimination of the federal budget deficit in the US, then that would be another major economic shift (away from deficit spending).
2/ Google got its mojo back. From Waymo robotaxis running wild on the streets of many of the most populous cities in the US to Gemini coming hard after ChatGPT, 2025 was a banner year for Google/Alphabet and a reminder that owning our data is key to owning us.
There is one issue that came out of all of this legal work that I need to understand better and possibly change in the documents. I scheduled a call with the company and its lawyer to go over that. Otherwise, I came away from this process confident that the company's legal affairs are in good shape and the closing docs reflect the term sheet we agreed to and mirror the customary provisions and protections USV receives in investments we make.
While this did take about two hours of my time, we did not incur any legal fees. NotebookLM is either free to use, or comes with USV's Google Workspace subscription. I honestly don't know the answer to that.
There is one more thing we can do in the VC industry to make this process even better. We can all agree to use standard docs like the NVCA documents that are publicly available to use.
With standard documents and Notebook Lawyer, prediction number four in my 2026 predictions can easily come true.
4/ A majority of venture capital deals close without lawyers on either side due to standardized documents (like the NVCA ones) and AI tools for review and legal diligence.
All we need is startup founders to demand this. And VCs to have the willingness to say yes.
This VC has already done that.
10/ The $AVC writer coin 10x its holder base which reaches 10,000 by year end.
Happy New Year everyone. I hope 2026 is a fantastic year.
4/ The end of writing code. I get to sit next to a software engineer in the USV pit, and boy has it been fun to watch Spencer make software. He doesn't write code the way I did when I was his age. He uses agents to generate the code and then stitches software together the way a prefab house is assembled. My partner Albert wrote about the significance of this shift yesterday and it is huge. Albert said, "Coding agents are doing to source code what compilers did to machine code: push the code below the interaction surface". I spent fifteen years working tirelessly to get computer science classes into the NYC public school system and the hardest part was teaching kids new languages that they needed to master to instruct machines. Now they just need one of many coding assistants and an understanding of how to do the stitching. That is a skill we still need to teach, but it is an easier skill to teach and the results come more quickly. Like Instagram made everyone a photographer, AI will make everyone a coder. And that's a great thing.
5/ GLPs are making us healthier. This is not a new trend. GLPs have been with us since 2005. But in 2025, we went from 6% of the US on GLPs to 12% and surveys suggest that 25% of us will be on one by the end of 2026. This is leading to a massive drop in obesity and diabetes, twin scourges of the last twenty-five years, but also massive drops in alcohol and drug addiction, which may also be a factor in the largest one-year drop in serious crime that the US has ever seen.
6/ We soured on our phones. The most significant shift has been the movement to take phones away from kids in schools, and the resulting increase in socialization, play, and more good stuff. But I think this is just the tip of the iceberg. Our almost twenty-year love affair with the smartphone is over, and everyone I know wants to use it less and reduce their reliance on it. This is bad news for Apple.
7/ Scale is hitting its limits in AI. OpenAI's brilliant move was to use scale over everything else to train large language models. And it worked. We got LLMs that can do magical things. But in 2025 we saw scaling reach its limits and new tricks, like distillation, fine-tuning, and reinforcement learning, produce significant improvements. The emergence of DeepSeek in January 2025, which was trained on significantly less hardware, was the first shot across the bow in the war between brains and brawn. This is bad news for OpenAI and others who are raising endless amounts of capital to win the scale war.
8/ Sports betting goes onchain. Prediction markets, which started out as a way to speculate on politics, moved into sports betting and went vertical. While neither of the two main prediction market providers is truly "onchain", I expect that they will ultimately move there for many reasons, including cost, speed, finality, and more. What this means is the "house" will be replaced by a trustless decentralized system known as a blockchain, and as we saw with DeFi before, this will lead to better markets that are much less extractive and are 24/7 and fully global. We can argue whether betting is a vice or societal good, but it is a fundamental human behavior that has been with us forever and now it will be onchain.
9/ China is winning the next war. While the White House and seemingly everyone else in the US obsess about winning the AI contest, China is winning the next war, which is electrification (which, by the way, is what powers AI). China installed over 300 GW of solar in 2025, bringing its solar installed base over 1 TW by year's end. The US, by contrast, is expected to deploy about 300 GW of solar over the next five years. For anyone who thinks this is about climate change (it is), this is mostly about economics. It costs about $0.05 per KW to generate electricity with solar at current manufacturing economics, and it costs between $0.05 and $0.10 per KW to generate electricity with natural gas. China is building a less expensive energy generation system than the rest of the world. This is bad news for the US.
10/ Streaming won. It probably won a decade ago or more, but in 2025 we saw Netflix beat a bunch of legacy studios in the bidding war for Warner Brothers/HBO. Owning the end customer is generally the better business model. When you can own the production and the end customer, you win.
OK, that is it for the rearview mirror. Tomorrow we will look at the road ahead.
There is one issue that came out of all of this legal work that I need to understand better and possibly change in the documents. I scheduled a call with the company and its lawyer to go over that. Otherwise, I came away from this process confident that the company's legal affairs are in good shape and the closing docs reflect the term sheet we agreed to and mirror the customary provisions and protections USV receives in investments we make.
While this did take about two hours of my time, we did not incur any legal fees. NotebookLM is either free to use, or comes with USV's Google Workspace subscription. I honestly don't know the answer to that.
There is one more thing we can do in the VC industry to make this process even better. We can all agree to use standard docs like the NVCA documents that are publicly available to use.
With standard documents and Notebook Lawyer, prediction number four in my 2026 predictions can easily come true.
4/ A majority of venture capital deals close without lawyers on either side due to standardized documents (like the NVCA ones) and AI tools for review and legal diligence.
All we need is startup founders to demand this. And VCs to have the willingness to say yes.
This VC has already done that.
10/ The $AVC writer coin 10x its holder base which reaches 10,000 by year end.
Happy New Year everyone. I hope 2026 is a fantastic year.
4/ The end of writing code. I get to sit next to a software engineer in the USV pit, and boy has it been fun to watch Spencer make software. He doesn't write code the way I did when I was his age. He uses agents to generate the code and then stitches software together the way a prefab house is assembled. My partner Albert wrote about the significance of this shift yesterday and it is huge. Albert said, "Coding agents are doing to source code what compilers did to machine code: push the code below the interaction surface". I spent fifteen years working tirelessly to get computer science classes into the NYC public school system and the hardest part was teaching kids new languages that they needed to master to instruct machines. Now they just need one of many coding assistants and an understanding of how to do the stitching. That is a skill we still need to teach, but it is an easier skill to teach and the results come more quickly. Like Instagram made everyone a photographer, AI will make everyone a coder. And that's a great thing.
5/ GLPs are making us healthier. This is not a new trend. GLPs have been with us since 2005. But in 2025, we went from 6% of the US on GLPs to 12% and surveys suggest that 25% of us will be on one by the end of 2026. This is leading to a massive drop in obesity and diabetes, twin scourges of the last twenty-five years, but also massive drops in alcohol and drug addiction, which may also be a factor in the largest one-year drop in serious crime that the US has ever seen.
6/ We soured on our phones. The most significant shift has been the movement to take phones away from kids in schools, and the resulting increase in socialization, play, and more good stuff. But I think this is just the tip of the iceberg. Our almost twenty-year love affair with the smartphone is over, and everyone I know wants to use it less and reduce their reliance on it. This is bad news for Apple.
7/ Scale is hitting its limits in AI. OpenAI's brilliant move was to use scale over everything else to train large language models. And it worked. We got LLMs that can do magical things. But in 2025 we saw scaling reach its limits and new tricks, like distillation, fine-tuning, and reinforcement learning, produce significant improvements. The emergence of DeepSeek in January 2025, which was trained on significantly less hardware, was the first shot across the bow in the war between brains and brawn. This is bad news for OpenAI and others who are raising endless amounts of capital to win the scale war.
8/ Sports betting goes onchain. Prediction markets, which started out as a way to speculate on politics, moved into sports betting and went vertical. While neither of the two main prediction market providers is truly "onchain", I expect that they will ultimately move there for many reasons, including cost, speed, finality, and more. What this means is the "house" will be replaced by a trustless decentralized system known as a blockchain, and as we saw with DeFi before, this will lead to better markets that are much less extractive and are 24/7 and fully global. We can argue whether betting is a vice or societal good, but it is a fundamental human behavior that has been with us forever and now it will be onchain.
9/ China is winning the next war. While the White House and seemingly everyone else in the US obsess about winning the AI contest, China is winning the next war, which is electrification (which, by the way, is what powers AI). China installed over 300 GW of solar in 2025, bringing its solar installed base over 1 TW by year's end. The US, by contrast, is expected to deploy about 300 GW of solar over the next five years. For anyone who thinks this is about climate change (it is), this is mostly about economics. It costs about $0.05 per KW to generate electricity with solar at current manufacturing economics, and it costs between $0.05 and $0.10 per KW to generate electricity with natural gas. China is building a less expensive energy generation system than the rest of the world. This is bad news for the US.
10/ Streaming won. It probably won a decade ago or more, but in 2025 we saw Netflix beat a bunch of legacy studios in the bidding war for Warner Brothers/HBO. Owning the end customer is generally the better business model. When you can own the production and the end customer, you win.
OK, that is it for the rearview mirror. Tomorrow we will look at the road ahead.
