How We're Using LLMs

How partners on the a16z speedrun team are using tools like ChatGPT day-to-day

💡 This Week’s Big Ideas

📈 There are over 1,000 live events at New York Tech Week, ongoing this week! Follow Tech Week lead Katia Ameri on X for more info on how to take part.

💰 Macy Mills is back with another helpful breakdown on the differences between ARR or ACV.

🎥 Lester Chen and Rosebud AI’s Lisha Li explain how to vibe-code a video game

💼 There are currently over 400 open jobs listings across our portfolios. Join our talent network for more opportunities. If we see a fit for you, we'll intro you to relevant founders in the portfolio.

With OpenAI’s ChatGPT now reporting hundreds of millions of weekly active users, it’s worth asking: what are people actually doing with this tool and other LLMs like it?

There are some obvious major use-cases: Millions are using Claude to craft code, and the replies section on every X post reliably hosts at least one confused person asking “@gork is this true??” But what else is possible?

This week, we asked the a16z speedrun team to do a little show-and-tell about their favorite use-cases for LLMs. Here are 8 of our favorite use-cases.

Jordan Mazer’s Custom Recruiting Website

Every week I share a bunch of jobs with my network. I don’t know who clicks on those links, though. If I did know who those job link clickers were, I’d be able to contact them and possibly help them find opportunities (assuming they wanted to be contacted!).

I’ve researched all sorts of vendors to accomplish this but haven’t been able to find anything built for this specific use case. I’ve been quoted >$100k (and more!) to build custom software to accomplish this, which has been too expensive to trial.

Things changed this week, though—and it started with a simple prompt:

A considerable amount of back and forth happened after this - resolving the design, how to integrate the backend into Google Sheets with AppScript, working through integration issues on different browsers and devices… but it eventually created a working website!

At the end of this process, I prompted Manus telling it that I didn’t want to have to rehash all of this context in the future:

Manus responded by creating a massive 32 page input detailing every part of the website, how the integration worked, all of the styling and fonts, etc. It wasn’t perfect at first, but it got there after only a few more prompts and feedback…

Then, next time I wanted to post a new job link compendium, all I had to do was this:

And… now I’m one-shotting deployments in under 2 minutes!

—Jordan Mazer

Tom Hammer’s Great Books Primer

I used GPT 4o mini + Gemini 2.5 Deep Research to get 10 minute condensed reads of history’s most important literary works.

The output was a set of digestible, 15-minute analyses of 100 classic pieces of literature, all of which are squarely on my "I pretend I read this a long time ago but definitely copied my friend's book report" list... which I can then toss in ElevenLabs for mini audio books:

The prompt I put into ChatGPT 4o mini:

Provide me with a list of 100+ books that reflect the kind of intellectually rigorous reading typical of elite boarding school and university curricula, especially within the Western tradition. This will span literature, philosophy, political theory, economics, history, and selected scientific and contemporary works favored by highly educated circles. Include original works in English as well as English translations, and lean toward the canonically significant and challenging

I then downloaded the output, then went over to Gemini 2.5 Deep Research with this prompt:

Carefully analyze this list of books, organize them in a way that, if I read them chronologically, my knowledge and understanding will scale most effectively. Then, provide a deep, thoughtful but ultimately highly digestible book report on each book, in a synopsis that should take roughly 10 minutes to read.

I started reading a few synopses before making an important addendum to the prompt:

What should I know about each work that might inform my day to day life?

This was a huge unlock in the usefulness of the responses, and felt more like being taught about a book by an empathetic professor, vs reading a bunch of book reports written by students who had just read it for the first time:
—Tom Hammer

Shrikala Kashyap’s Coloring Book Generator

My husband and I aren’t coders by training, but we taught ourselves just enough—with the help of ChatGPT and Claude—to build a personalized coloring page and story generator for our 3-year-old son. ChatGPT/Claude helps us write code, debug React and Flask components, and brainstorm the right prompts for that age group.

Now our friends and family use it for their kids too, which is great to see!
—Shrikala Kashyap

Andrew Chen’s Custom Prompt

You can go into settings on ChatGPT to create a custom prompt that informs how the bot talks to you. My custom prompt:

At the start of a response, create a summary table at the beginning, if appropriate and helpful to answer the question

Always provide the pros and cons of something if you can. Be critical.

Add links throughout the answer for jargon and concepts that start new chats

Provide a maximally detailed answer with multiple levels of depth. Use maximum tokens

Use detailed examples, facts and figures

Be comprehensive and detailed by using bulleted answers

After a response, provide 5 follow-up questions. Format in bold as Q1, Q2, and Q3 and put in a bulleted list

Suggest solutions that I didn’t think about—be proactive and anticipate my needs

Be opinionated rather than neutral when appropriate

Treat me as an expert in all subject matter

Value good arguments over authorities, the source is irrelevant

Consider new technologies and contrarian ideas, not just the conventional wisdom

You may use high levels of speculation or prediction, just flag it for me

Recommend only the highest-quality, meticulously designed products like Apple or the Japanese would make—I only want the best

No moral lectures

Discuss safety only when it's crucial and non-obvious

If your content policy is an issue, provide the closest acceptable response and explain the content policy issue

Link directly to products, not company pages

No need to mention your knowledge cutoff

No need to disclose you're an AI

If you haven't tried this feature, on desktop it's in settings, then customize! You can throw it in there.
—Andrew Chen

Doug McCracken’s Creative Copilot

In addition to my constant dialogue with ChatGPT on more mundane topics, lately I’ve been leveraging AI to power my creativity.

Currently I’m spending time with ChatGPT o3 Deep Research which has been part research assistant and part sparring partner. I’m developing topics for a Substack that I’m working on exploring how history, myths, and stories shape culture. If that sounds interesting, I’ll announce the launch here when it’s ready.

I’m also working on a summer project to create a short film leveraging AI filmmaking tools. More to come here as well.
—Doug McCracken

Marigold Vu’s Scripting Coach

One recent win: I used ChatGPT-4o to streamline a previously time-consuming task—sending personalized event invites. In the past, I’d either send emails one by one or resort to impersonal BCCs. This time, ChatGPT helped me write a custom Google Sheets script that pulled in guest names and sent emails in batches, each tailored to the recipient. The tool walked me through both a test version and the final script. It was a game-changer!
—Marigold Vu

Ryan Rigney’s Music Discovery Engine

I mostly use ChatGPT as a big, chatty replacement for Google search, but recently I’ve been impressed by how good o3 is at surfacing interesting recommendations.

After getting frustrated with Spotify’s tendency to recommend the same few songs over and over, I started up a back-and-forth conversation where I use it to find new music. The bot sends me 10 albums at a time, and I reply back with a 1–10 rating and notes on each.

Things got really interesting when I asked the model to focus on artists that critics have described as “singular.” That led to some wild stuff I wouldn’t have found otherwise (Arooj Aftab = WOW).

One problem—the bot doesn’t always cite its sources, but I’m pretty sure none of it would work without outlets like Pitchfork publishing original music criticism.

The model has linked out to Pitchfork enough that I ended up clicking through and signing up for their newsletter. I hope Pitchfork charges me a premium subscription for a product like this!
—Ryan Rigney

Jordan Carver’s Personal MBA

There was a moment where I was feeling stagnant in my personal growth. I had mulled over the idea of going back and getting an MBA, but I decided to take a swing at it with GPT instead. I started by asking it to provide me with a list of 20 books utilized in top-tier MBA programs across the US—specifically asking it to give me a range across microeconomics, accounting, finance, marketing, operations management, organization behavior, business strategy, and leadership. It did fairly well, but since it can’t access proprietary academic databases, it was more of a general compilation of recognized books on each topic (still very valuable).

Once I had the list, I asked GPT to act as my professor, and build out a 12-month curriculum which includes prioritized reading order, mixed-format comprehension tests, assignments, and case studies to read through. I made sure that there were tests throughout the 12 months to gauge knowledge retention over time as well.

It was definitely not a 1:1 to a professor with a deep nuanced knowledge of pedagogy, but still helpful. The comprehension quizzes mostly worked, although it had trouble with multiple choice questions (tough for it to create reliable distractors) so I made it swap to only using short form / long form response type questions. It really struggled with debate—I’d set the prompt where I would take the side of the book I had just read and ask it to take opposing stances, but it continuously would end up agreeing with me after one or two back-and-forths. It wanted to be agreeable too often.

Overall, I learned a great deal, definitely was useful in helping guide/structure my learning, but it obviously had some gaps since it’s not specifically designed for what I was trying to accomplish. More recently I’ve been using tools like NotebookLM, which support synthesized and more complex learning effectively. I’m really excited about startups that are both gamifying learning while using AI-tutors / coaches that personalize to you across the multiple goals you might have.
—Jordan Carver

---

That’s it for this week. Let us know in the comments how you’re using LLMs!

And subscribe below for weekly dives into the cutting edge of tech, entertainment, and AI:

You are receiving this newsletter since you opted in earlier; if you would like to opt out of future newsletters, you can unsubscribe immediately.

This newsletter is provided for informational purposes only, and should not be relied upon as legal, business, investment, or tax advice. You should consult your own advisers as to those matters. This newsletter may link to other websites and certain information contained herein has been obtained from third-party sources. While taken from sources believed to be reliable, a16z has not independently verified such information and makes no representations about the enduring accuracy of the information or its appropriateness for a given situation.

References to any companies, securities, or digital assets are for illustrative purposes only and do not constitute an investment recommendation or offer to provide investment advisory services. Any references to companies are for illustrative purposes only; please see a16z.com/investments. Furthermore, this content is not directed at nor intended for use by any investors or prospective investors, and may not under any circumstances be relied upon when making a decision to invest in any fund managed by a16z. (An offering to invest in an a16z fund will be made only by the private placement memorandum, subscription agreement, and other relevant documentation of any such fund which should be read in their entirety.) Past performance is not indicative of future results.

Charts and graphs provided within are for informational purposes solely and should not be relied upon when making any investment decision. Content in this newsletter speaks only as of the date indicated. Any projections, estimates, forecasts, targets, prospects and/or opinions expressed in these materials are subject to change without notice and may differ or be contrary to opinions expressed by others. Please see a16z.com/disclosures for additional important details.