How AI Has Turned My (VC) Work Into Flow (Part II)

From writing essays to assessing market trends, AI has transformed my arduous work into a creative elucidation of craft. In this second part of my essay, I'll explore how AI has impacted how I send cold emails, learn to program, and take notes.

My Work Flow

My workflow has seen a few improvements with all the recent AI innovations, including:

1. Outbound sourcing

2. Up-skilling: learning how to program

3. Meeting note-taking, distillation, and synthesis

Deal Sourcing: Outbound

What Does This Look Like?

In the hours of the day that I’m not doing research or writing, my time is spent chasing deals.

Working at a small fund affords me significant autonomy for spending my time while also challenging me to make up for the fund’s shortfalls in “brand awareness.” My time is spent reaching out to founders in my network who are “working on something new” or building enduring relationships with those who may fit the founder archetype my fund tends to track and back.

The tools available to me today help me with the second half of sourcing: cold outreach.

One thing you recognize as a venture capitalist is that if you are doing your job correctly, there is a considerable amount of competition to identify, track, and win deals that go on to return multiples of a fund. That means a generic “Can we catch up over coffee?” line from your firm’s work address won’t be fruitful.

My solution is to do background research on the individual rather than the company or sector they work in. To win the hearts of founders, my hypothesis has been to treat each one more as an adventurist choosing me to come along for a journey rather than as a commodity to achieve some monetary goal. This makes building the relationship more genuine and authentic.

The drawback of approaching cold outreach from a human-centered mindset is that it’s not scalable to curate personalized, targeted messages to high-propensity founders. But it’s the only approach that allows my email to become the signal above the noise.

My manual process includes doing background research and finding some unique aspects about them sprinkled across their digital footprint. This could range from a newspaper article recognizing them for a significant act they did in high school, diving into their public Github documentation, reviewing their posts on Substack, or reviewing some of their academic publications.

After aggregating and synthesizing this digital make-up, I pop into ChatGPT to help craft a personalized, “natural” outreach.

Let’s take a hypothetical founder and apply this strategy. Some quick research on Mr. X yields the following:

• Graduated with distinction from a top-tier university with a degree in computer science

• Mentioned in a newspaper article referencing an app they built to help make it easier for high school students to find summer internships at tech companies

• Wrote a piece on Substack on how to find more serendipity through outdoor walks

• Built a side-project that aggregates all my contacts and tells me when it is their birthday

This should be enough.

I head to ChatGPT and “prime” the software to understand my writing style. I copy and paste some of the writings I’ve published on Medium, and once ChatGPT summarizes my input, I provide the following prompt:

Is this good? No, not entirely.

But with a few more prompts, I could get this 75% of the way there and then switch out facts I learn about other founders in the future to speed up the process.

While the tool still requires manual fine-tuning to achieve an outcome, I’m happy with the output.

It still affords me time savings in that the structure of the email is in place, and my time and efforts can be spent editing and word-smithing my way to an acceptable cold email.

Tool: ChatGPT (Cold Email Outreach)

• Link: https://chat.openai.com/chat

• Utility: 4.0 / 10.0

• Drawbacks:

  • Since I am a native English speaker and have a good amount of practice writing these types of emails, it may be faster for me to write it out initially versus relying on ChatGPT as a crutch. The extension is that I could code up a program using GPT-3.5 to take facts I know about other founders and turn them into the emails as above. It would undoubtedly provide me leverage, but the workflow isn’t as straightforward.

Up-Skilling: Learning How To Program

What Does This Look Like?

At the start of Part I of this series, I did a small tirade about how Web3 was a back-end revolution. Here, it comes full circle. The hero of most Web3 conferences, meetups, and funding are developers. Unbeknownst to me, developers tend to enjoy building tools for other developers. These tools end up becoming valuable businesses (hopefully).

As a result, my role as an investor is to focus on these developer-oriented tools or, as they are colloquially known, Web3 picks-and-shovels.

To invest well in this segment of Web3, I couldn’t focus on simple heuristics like how much do I vibe with the team; I needed to explore concrete aspects of their most tangible output: open-source code.

Here’s the thing: I don’t know how to program. 😬

So, for the last three months, I’ve been waking up at 5:30 am to spend 1-2 hours before the workday learning how to code from a first-principles perspective.

Update: it’s hard.

AI Tool + Coding

Here’s where I fell in love with ChatGPT: it writes code and explains concepts to me as if I’m five years old.

It may be flawed, and the syntax may be different from what I am looking for, but it’s similar to writing in that a skeleton is produced that makes putting the meat on the bone much, much easier.

Let’s take a look at a quick example.

One of my coding exercises is teaching me about APIs. The assignment requests that I code an asynchronous function that returns a promise. Within the function, I needed to include an await fetch() request to handle the promise returned by the API call.

What does this mean? Your guess is as good as mine.

By simply entering a highly specific prompt, I could copy and paste the question into the program and get a rough draft code base requiring limited editing, like my API URL.

This is nifty because it showed me that you don’t need to be an expert at Googling if you are good at asking questions (e.g., prompt engineering). Before ChatGPT, I’d have to ask a similar question using less specificity (increasing my margin for error on the search) and then parse through hyperlinks, hopping from one link to another, until I came across what looked like the answer on StackOverflow to help me write the code.

What would have taken me 20 minutes as a StackOverflow novice turned into a copy-and-paste exercise.

Do I know how to program? A bit.

Do I feel confident I can take bite-sized assignments and devise a solution thanks to ChatGPT? You bet!

My confidence has increased tremendously, knowing I can leverage ChatGPT to augment my learning.

So, to tie in my Web3 investing, instead of relying on others to tell me the contents, functionality, and potential unforeseen edge cases in smart contracts created by Web3 protocols, I can paste their code directly into ChatGPT and have it walk me through step by step, in plain English (and usually as if I am five years old).

Tool: ChatGPT (Programming)

• Link: https://chat.openai.com/chat

• Utility: 9.0 / 10.0

• Drawbacks: From ChatGPT’s perspective:

Meeting Note-taking, Distillation, and Synthesis

I take a LOT of notes during calls with founders. My questions and filters vary depending on how far along a potential investment opportunity is, the category or niche a startup plays in, and the team's background.

As a note, my style during meetings is less question-and-answer volleying and, by design, more conversational.

These meetings aim to get to know the team and, more importantly, how they think beyond what is written in a deck or an answer to a simple “yes and no” question.

The typical founder experience with a venture capital firm goes something like this:

• First meeting with an Associate (a junior-level employee)

• Second meeting with a Principal (a mid-level employee)

• Third meeting with a Partner (a senior-level employee)

• Fourth meeting with a General Partner (a key decision maker)

• Investment Committee (senior partnership meeting determining the merits of a deal and if it will be funded, etc.)

This isn’t exact, but it represents the touch points a founder can expect from a $250M+ AUM seed fund. For each of these meetings, 30-minutes to an hour in length, the typical questions tend to be covered over and over again with each touch point:

• What is the problem you are solving?

• What’s your solution?

• How big is the market?

• How do you plan to make money?

• Why is your team best fit to be the winner-take-all player in the space?

My issue with this is that had good notes been recorded with proper summarization and key insights initially, subsequent meetings could have been devoted to drawing out more actionable learnings rather than rehashing old ones.

Given how hard it is to start a company, I don’t agree with making founders feel like they are “ping-ponged” from one meeting to another to cover the same information. Their time is more valuable than how they are often treated.

Given that I intend to make the founder's experience with my firm as pleasant as possible, I approach each meeting such that whatever notes I take are fully summarized and contain actionable insights for the next touch point in my firm.

What this looks like:

I try to take notes while on calls with founders because I can type significantly faster than I can hand-write. I’m so good about stealthy typing that founders have told me they didn’t even realize I was taking notes.

But speed comes at a cost.

My notes look like someone threw up the alphabet, then used a lint roller to clean it all up.

Given the goal in the previous section, if I left my notes as I had initially typed them, they’d be useless for my team and, with enough time, me, too.

AI has allowed me to take my notes, reformat them with markdown, and then subtly “improve the language and grammar” so it’s readable and action-oriented for my team.

Subsequently, once the body of text has been reformatted, it’s easier for me to pull my insights and assessment from a 10,000-foot view. This part is the most critical, so using a transcription or summary video recorder wouldn’t work for my profession. My notes aren’t meant to capture what a founder says - they are intended to capture my insights and the assumed implications of their answers.

For example, they may give me a go-to-market strategy, but recording the specifics is less critical than recording the potential risks and pitfalls that their likely approach may entail. Or, separately, if the team slide includes background working at a tech company for which I have a close network, I can create my own checklist or action items to reach out to someone from that company to act as a reference. Video recorders only increase the work for me, and from my perspective, they come across as lazy. Transcription software only degrades the conversation’s experience and creates psychological barriers when parties realize that they are being recorded word for word.

AI Tool + Note Taking:

To preserve founders' privacy, I will include an example of general notes I had taken for a Web3 Token Article. As you’ll see below, the notes are disparate, disjointed, and largely hard to decipher, especially if there’s some time delay between when I took the notes and when I plan to synthesize them for an essay.

Raw Notes:

Now, my original notes were good but not great. They did also capture some of the added emphasis and insights that I was looking for as they built upon prior knowledge I had put together.

Thanks to tools like Notion AI, with a few shortcuts, I can get a summarized output that will help refresh myself on the so-what of this text:

The output then adjusts to this:

What’s great now is that I can opt to both improve the language, make my notes shorter, and summarize in three clicks without having to do it myself, making it easier for me to get back up to speed and gain the utility I originally intended to get from taking these notes in the first place.

It doesn’t take much effort to see how valuable this can be in taking meeting notes, then improving them with a few AI shortcuts, thanks to Notion. From there, it’s easier to share via email or into our CRM with less fuss and deciphering for my teammates 🙂

Tool: Notion AI

• Link: https://www.notion.so/product/ai

• Utility: 8.5 / 10.0

• Drawbacks:

  • Potential loss of tone and style through summarization

Summary

From improving my ability to draft, code, source deals, and take notes, AI tools like GPT-3 have helped me handle the basic mechanics of writing. In terms of saving time or reducing my need to reinvent the wheel, I can focus more on the more impactful things in my role: thoughtful intentionality on my work and the craft-oriented approach therein.

These tools are imperfect, often requiring me to roll up my sleeves. But, it gives me an immense amount of hope, given that the first few customer-facing iterations have given me this much benefit. I’m excited to see what comes next and how my workflow will evolve 🚀