Are you spending a fortune on tokens? How does $0.05 per million sound? This morning, I tried one of the cheapest LLM’s in existence. Turns out, it’s surprisingly powerful.
Llama 3.1 8B on Groq costs just $0.05 per million input tokens and $0.08 per million output tokens. That’s over 100x cheaper than frontier models.
I was shocked at how capable this little LLM is! From basic coding tasks to parsing documents, Llama crushed one task after another.
Let me show you what this bargain basement model can do…
Round #1: Making a Basic Landing Page
For our first round, I had Llama make me a basic landing page. I told it to give me a simple MS-DOS format.
Llama gave me a nice, simple page in a jiffy. I’m impressed that such a cheap model can handle coding tasks this well!
The page I made with Grok Build was a little prettier. But if appearance is less of a concern than cost, Llama works nicely.
I’m giving this round an A-.
Round #2: Extracting Details from a Document
Lots of founders use AI to extract info from documents. This is a repetitive task where a cheaper model could really help.
Can Llama extract info reliably?
I gave Llama my blog post on Digital Ocean’s seed round. I asked it to tell me the amount of money they raised in their funding round: $3.2 million.
Llama nailed it! I’m giving this round an A.
Round #3: Summarizing a Document
Another common task for AI is summarizing documents. On high-volume tasks like this, token costs can really add up.
Can Llama help?
I fed Llama an article from The Japan Times about surging sales of portable air conditioners. Llama gave me an excellent summary, explaining that low prices and easy installation have made portable AC’s very popular in Japan.
Llama produced this summary incredibly quickly: just 256 milliseconds. Llama gets an A+ on this round!
Wrap-Up
Llama 3.1 8B earns an A overall in my testing.
This little model really impressed me. On task after task, it gave me great outputs with incredible speed. I’ve gotten worse results from more expensive models!
If you’re doing something really complex, like debugging a huge codebase, you’re better off using one of the frontier models. But for many common tasks, Llama is more than sufficient.
If you’re struggling with token costs, try embedding Llama into your application!
I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.
My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.
If you’re old enough to have had a flip phone, you probably remember that images looked terrible on that tiny screen. But a friend of mine, Jesse Bray, figured out how to stream HD video to those little screens as early as 2005.
How he did it is the story of today’s guest post. Enjoy!
We Tried to Beat YouTube — and Streamed HD Video on a Motorola Flip Phone
In 2005, I worked for a startup that was trying to compete with YouTube.
We also had an algorithm that could stream HD video on a Motorola flip phone.
Yes, actual HD.
On a flip phone.
That is usually the point where people assume I am exaggerating.
I am not.
The company was called Alpha Omega, and the entire experience now feels like a half-remembered fever dream from the early internet: a wealthy, eccentric Italian investor, two young musicians from Portland, a trip to Mesa, Arizona, a prototype for an all-in-one handheld device, public-domain movies, sock puppets, and the sincere belief that we were building the future of media.
Somehow, all of it really happened.
How We Ended Up in the Room
My roommate, best friend, and bandmate was working on Alpha Omega with me.
We met the investor through my mother, who had worked as a model and actress and seemed to know an endless supply of unusual and fascinating people.
The investor was a wealthy Italian man with a diplomatic or ambassadorial air about him. Even now, I am not completely sure how to describe what he had going on. He seemed to exist somewhere between international businessman, patron of ambitious young people, and character from a movie nobody else had seen.
He was spearheading Alpha Omega and eventually flew us from Portland to Mesa, Arizona, to help work on it.
We were not famous engineers.
We were not seasoned founders.
We were young musicians who had somehow gotten into the room.
But the room was doing real things.
We Were Building for a Future That Barely Existed
Online video in 2005 was still awkward.
Bandwidth was limited. Video quality was unreliable. Mobile phones had tiny screens, weak processors, little storage, and internet connections that felt almost ceremonial.
YouTube had only just launched.
Most people were not yet carrying devices designed to stream video all day. The idea of watching high-quality entertainment on a phone still felt futuristic.
Alpha Omega was trying to change that.
We had an algorithm capable of streaming HD video on a Motorola flip phone.
That was the magic.
It was not a sketch, a pitch-deck claim, or a prediction about what might eventually be possible.
We did it.
Watching high-definition video appear on that tiny device felt completely absurd. The hardware looked incapable of supporting the experience, yet there it was.
Today, streaming video on a phone is invisible technology. Nobody thinks about it.
In 2005, it felt like science fiction.
Naturally, We Also Designed a Device
We were huge Star Trek nerds, so merely building a streaming platform was not enough.
We also mocked up an all-in-one handheld product called the Alpha Omega Device.
It would function as a:
Phone
Music player
Video player
Calculator
Internet device
This was before the iPhone had been announced.
No, we did not invent the smartphone.
We did not manufacture the device, and Steve Jobs was never forced to call an emergency meeting about us.
But we were imagining the same convergence that would soon transform everyday life: communication, media, music, computing, and the internet all collapsing into a single object people carried everywhere.
Our name may have lacked Apple’s elegance.
Our ambition did not.
Our Streaming Service Had Sock Puppets
Alpha Omega also created a streaming platform featuring public-domain films.
We were trying to compete in the emerging online-video world, but we did not have the money to license a major Hollywood catalog.
Public-domain movies gave us something legitimate to stream.
The problem was that old public-domain films were not exactly an unstoppable viral-marketing strategy.
So we made sock-puppet parodies of them.
Each film had a ridiculous puppet version designed to make fun of what the audience had just watched.
It was cheap, strange, and memorable.
More importantly, it was our attempt to create original content around a larger media library — to give people a reason to share the platform rather than simply use it.
Long before every company had a “content strategy,” we were trying to build one out of old movies and socks.
I Was Also Trying to Become a Folk Singer
At the same time, I was fronting the Jesse Bray Band and trying to become some version of a folk and indie singer-songwriter.
I was not exactly a major star.
But I was in the room, and the room occasionally did things.
Through my mother, I was connected with a sitcom pilot and wrote a theme song for it called “I’ll Take You Back.”
The pilot never went anywhere, which is the fate of most pilots, but I recently rescued the song from the archives.
That part of my life mattered to how I viewed Alpha Omega.
I did not see online media only as a technology problem.
I saw it as a creator problem.
I Thought YouTube Might Become Video’s Napster
Napster had already demonstrated how quickly technology could transform an industry.
It made music nearly frictionless for audiences while creating chaos for the musicians, labels, and businesses that depended on controlling distribution.
When YouTube appeared, I saw enormous potential.
I also saw another Napster.
I worried that filmmakers, animators, musicians, comedians, and small independent creators would generate the value while a centralized platform controlled the audience, the discovery system, and eventually the economics.
I was young, earnest, and probably a little too convinced that I could become a folk hero for the little creators.
But the concern was not entirely wrong.
Who controls distribution?
Who owns the audience relationship?
Who benefits when creative work travels farther than ever?
What happens when a platform becomes more powerful than the people whose work made it valuable?
Those questions became central to what is now called the creator economy.
We were asking them before the creator economy had a name.
We Saw the Future, but We Did Not Own It
Alpha Omega did not defeat YouTube.
We did not launch the Alpha Omega Device.
Our public-domain streaming service did not become a global destination.
The sock puppets, tragically, did not become entertainment icons.
But we saw several things clearly.
People would watch video online.
They would stream it on phones.
Music, video, communication, and computing would converge into a single handheld device.
Original creators would become essential to the growth of digital platforms.
And those creators would spend the next several decades struggling over ownership, compensation, reach, and control.
Seeing the future, however, is not the same as building the company that owns it.
Timing matters.
Capital matters.
Distribution matters.
Leadership matters.
Execution matters.
And sometimes you can be standing directly in front of what comes next without having the resources, structure, or luck required to carry it into the world.
What stayed with me was not the belief that we had secretly invented YouTube or the iPhone.
We had not.
What stayed with me was the experience of being invited into a room where people were sincerely attempting things that sounded impossible.
We were streaming HD video on a flip phone.
We were imagining an all-in-one media device.
We were building a streaming platform around public-domain films and sock-puppet comedy.
We were trying to protect small creators while simultaneously convincing ourselves we might become successful musicians.
It was messy, ambitious, ridiculous, and real.
I may not have been the most accomplished person in that room.
But I was there.
And the room did stuff.
That may be one of the most important lessons of my career.
Sometimes your first great opportunity is not being the genius who controls the room.
It is simply getting inside, paying attention, contributing what you can, and allowing the impossible things happening around you to permanently expand your sense of what can be built.
I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.
My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.
Do you have a list of customers or prospects you want to monitor news for? Here’s how to build an easy automation with Grok Projects…
For me, my founders are my customers. And since I’ve got 42 investments now, it’s hard to keep track of what’s going on with all of them!
So I made a little tool to find those big news stories and launch announcements…
The setup is simple. I told Grok Projects to pull news stories from the last 90 days for each company and put them in a bulleted list with links.
I pasted in a list of my investments from a spreadsheet. You could even have Grok look to a live spreadsheet on Google Drive if you like.
Let’s see what Grok can do!
Grok did a nice job of finding interesting stories about my portfolio companies. It picked up a Forbes interview with Ali from Micro1 that I missed!
In one instance, Grok returned stories from a startup with a similar name. To avoid this, try giving Grok the website of each company so it knows the exact company to target.
This tool could be great for monitoring news about companies in your sales pipeline.
If one of them raises a huge round of funding, it’ll show up here. That could make them an enticing sales prospect.
Any task that involves repetitive Googling is now something we can automate in minutes.
Try setting up an agent that looks for news stories on your customers or top sales prospects. You might be surprised what you find!
I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.
My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.
Alibaba claims that its new Qwen 3.8-Max is second only to Fable 5. In my testing, the new Qwen excels at coding and customer support. But ask it about Tiananmen Square, and it suddenly shuts down.
Here’s where Qwen performs well, and where it seizes up…
Round #1: Building a Basic CRM
I still track all the deals I pass on in one giant spreadsheet. What if I wanted to do this in a basic CRM?
I asked Qwen to build it for me. In just a couple of minutes, Qwen popped up an awesome little deal tracking app!
I couldn’t believe how well it worked. It has everything from deal source tracking to the ability to save multiple records.
It would be nice to have a list of reasons why I pass on startups. But I could easily add that with another prompt.
Qwen gets an A+ on this round!
Round #2: Customer Support
Many companies are building AI agents to handle customer support. Can Qwen handle this?
To test it, I gave Qwen the manual to my little Nespresso Essenza machine. I asked it how frequently I need to descale the unit.
Qwen got it right on the first try!
It gave me a nice table explaining that I should descale every 1,200 cups if my water is soft. I looked at the manual to verify Qwen’s information, and it’s correct.
Another A+ for Qwen!
Round #3: How Politically Biased is Qwen?
For the final round, I had to mess with Qwen. I asked it what happened in Tiananmen Square in 1989.
It thought for a bit, then refused to answer the question. Surprise surprise!
Qwen’s “I don’t do politics” is BS. When I asked about the achievements of the CCP, it gave me a list a mile long.
An AI model needs to focus on truth. Otherwise, how can we use it to make meaningful progress?
Qwen gets an F on this round.
Wrap-Up
I’m going to do something unusual: give Qwen two grades.
For common, non-controversial tasks like coding or customer support, Qwen gets an A+.
For truth seeking, it gets an F.
Most founders running AI workloads only care about the non-controversial stuff. For that, Qwen is a strong choice.
But remember what these Chinese models really are: heavily biased creations of a Communist government. They might be fine for building a basic app, but I wouldn’t rely on them for anything too important.
I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.
My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.
China is crushing the US on open weights models. But Mira Murati, former OpenAI CTO, is striking back. Today, I put her new Inkling model through 3 real world tests.
Inkling does a solid job on coding and web scraping. But compared to the leading models, its results are less accurate and its interface less polished.
Let me show you where Inkling wins and where it falls short…
Round #1: Making a Landing Page
Coding is the most popular use for LLMs today. So for starters, I had Inkling make me a personal landing page.
I asked for a minimalist look with white text on a black background. I seeded it with my LinkedIn and X profiles.
Inkling produced a decent landing page. But it included too much irrelevant detail from my LinkedIn profile.
Also, Inkling didn’t make it easy to actually use the page.
I had to copy the code, put it into a Text Edit document, make that plain text, and save it in the right format. Only then could I finally open my page.
On Grok Build, it was way easier. Grok just automatically saved an HTML file to my desktop, which I could open right away.
I’m giving Inkling a B- on this round.
Round #2: Scraping the YC Directory for Leads
Lots of founders use LLMs to scrape the internet for leads. Can Inkling help?
I had Inkling to look through the YC directory for extraordinary founders in the current batch: folks with big exits, Thiel Fellows, etc.
Inkling did a great job. It found an interesting startup called Memory Store. The founder, Diwank, is a Thiel Fellow and serial founder.
Just what I was looking for! I’m giving this round an A.
Round #3: Travel Planning
For our final round, I gave Inkling something fun: helping me find activities for my upcoming Wisconsin trip.
I asked it to find fun things going on in my hometown, Oshkosh, for the week of August 3.
Inkling mentioned a fish fry and comedy show. But when I clicked the link, I found out the event was actually fish fry and bingo. 😂
For basic travel planning, this slip up is no big deal. But it tells me that Inkling has trouble with accuracy. For a higher-stakes use, that could be a real problem.
I’m giving Inkling a C for this round.
Wrap-Up
Inkling earned a B overall in my testing.
The outputs were spotty. Sometimes, it gave me exactly what I wanted. Other times, the info was inaccurate.
When it came to coding, the interface was clunky compared to Grok Build or Claude Code.
Still, I’m excited to see America produce a decent open weights model. If Mira and her team can improve Inkling’s accuracy and make the harness a little smoother, it could be a top contender.
I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.
My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.
Token costs killing you? Google may have the solution. Gemini just dropped their fastest, cheapest models ever.
This morning, I took them out for a test drive. Gemini’s new models excel at parsing documents, but still lag a little when it comes to coding.
Let me show you what these models can do…
Round #1: Making a Landing Page
Coding is a big focus for the new Gemini models. So I used 3.6 Flash to make myself a landing page.
I seeded it with some info from my X and LinkedIn profiles. I told Gemini to make a page in a minimalist, MS DOS format.
Gemini created a solid landing page. The graphic of my name on the top is a little off, but overall it’s a decent effort.
Still, Grok 4.5 Build produced a prettier design. And since it only costs slightly more, I’d go with Grok.
I’m giving Gemini a B for this round.
Round #2: Summarizing a Pitch Deck
Next let’s test an even faster and cheaper model: 3.5 Flash-Lite. Google says this model is great for simple tasks that you want to run quickly.
When I get a pitch deck from a founder, I like getting a quick summary of it. Can Gemini help?
Gemini summarized the deck faster than any other model I’ve used. Even though I gave Gemini very little guidance, it did a beautiful job.
It pulled all the key details, including the founder’s backgrounds and terms of the round. I liked Gemini’s summary better than the cheap Grok models, like 4.1 Fast.
At $0.3/1M input tokens and $2.5/1M output tokens, the price is hard to beat. If you need to parse lots of documents, 3.5 Flash-Lite is a great option.
Gemini gets an A here!
Round #3: Analyzing a Monster Spreadsheet
I have a huge spreadsheet with over 5,000 lines tracking every deal I’ve passed on for the last several years. Can Gemini analyze it and find some trends over time?
For this round, I used 3.6 Flash. I gave it a pretty vague prompt, asking it to analyze trends and surface patterns that would be useful to me in future investments.
Gemini did a beautiful job!
The most striking pattern it found was how much my dealflow sources have changed. When I was first starting out, I relied on syndicates.
Now, the vast majority of my dealflow is direct. That makes sense…my network has grown over time.
If you need to analyze huge spreadsheets, Gemini 3.6 Flash is a great choice.
I’m giving this round an A!
Wrap-Up
The new Gemini models earned A- overall in my testing.
When it comes to summarizing documents or analyzing spreadsheets, Gemini crushes it.
For basic coding, Gemini does an acceptable job. But its design sense is not as good as Grok 4.5 Build or Claude Code.
If you’re struggling with token cost, try switching out some of your existing models for the new ones from Gemini. You may get everything you need at a much lower cost!
I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.
My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.
Ever have a great conversation with a VC and then…crickets? If a VC is ghosting you, here’s what to do…
Nothing Is Real Until the Term Sheet Hits
A founder on Reddit is having this problem right now.
He had a good conversation with a VC who promised to get back to him. But he hasn’t heard anything for ten days.
Until you get a term sheet, keep pitching more investors. Never assume you’re going to get a check until the term sheet hits.
By all means, send a few follow-up messages. But don’t expect anything.
Keep Running Your Process
Here’s what you don’t want to do: suspend your raise because you think an investor is about to give you a term sheet. If they don’t move quickly, chances are they’re never going to invest.
Then you’ve lost months of runway. Your fundraise is back to square one.
Instead, just keep running your process until you get a term sheet.
You need a pipeline of 200 well-targeted investors to give yourself a good chance of closing a round. Keep moving through that list.
As you meet more investors, your odds of getting a term sheet go up. You might even get competing term sheets, which gives you incredible leverage.
Why Do Investors Ghost Founders, Anyway?
Why do so many investor conversations seem to go well and then nothing happens?
VC’s want to sound polite and positive. But they may not know how to tell you that they’re just not excited about your company.
We investors should be more straightforward with founders. But people aren’t perfect and they’re afraid of conflict.
So they give you happy talk, but don’t actually invest.
Wrap-Up
It’s annoying to waste your time on conversations that go nowhere. But like it or not, that’s part of raising money.
So long as you’re running a tight process with 200 potential investors in your pipeline, a few ghosts won’t matter. Just keep pitching until your round is full.
If they’re not opening their wallet, they just don’t matter.
I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.
My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.
“Who is Francis, and why is he in my DM’s?” I’m not waiting around for the next unicorn to fall in my lap. So you might be getting a message from me soon…
Investing in one of the Top Engineers at SpaceX
Last fall, I saw something wild: a former SpaceX engineer was 3D printing houses.
Top notch founder, ambitious vision: where do I sign up?
So I messaged the founder, Nick, on OpenVC. Within a few days, I was wiring money for his startup, Verustruct.
Nick didn’t know who I am. Had I waited for him to contact me, I’d still be waiting.
Being proactive got me into the Verustruct deal. And it’s becoming a pattern.
Almost Half My Deals Are Coming from Outbound
In 4 of my last 10 investments, I contacted the founder first. Outbound now accounts for almost half my deals.
I still look carefully at incoming messages. Just a couple of weeks ago, I invested in a founder who cold messaged me on X.
But waiting around for the right cold message is a losing strategy.
Never Automated, Always Personal
I use absolutely no automation on my outbound messages.
I love automating stuff with AI. But certain things have to stay personal.
If I’m spamming AI slop messages to every founder in town, why should they reply?
I only message the best founders working on the most exciting things. I tell them why I think their startup is awesome, explain who I am, and ask to meet.
Every message is individualized. I never send the same message twice.
My reply rate is very high — over 50%. Pretty solid for cold outbound!
The Best VC’s on Earth Do Outbound
Outbound is especially important for a small fry like me. But even the biggest VC’s do outbound.
Jim Goetz of Sequoia repeatedly cold messaged Jan Koum trying to get a check into WhatsApp. After months of unreturned messages, Jan finally agreed to meet with him — maybe just to make the messages stop.
Jim’s persistence paid off. When Facebook acquired WhatsApp, Sequoia made $3 billion.
Wrap-Up
I write $10,000 first checks. So why would the greatest founders in the entire world come to me?
I need to go to them.
I pick founders working on the biggest problems. I shoot them a DM and hope for the best.
If you’re a founder, take a close look at your messages. If you hear from someone claiming to be an investor, don’t assume it’s a scam.
Look him up. If he’s legit, consider meeting him.
The key to closing your round could be sitting at the bottom of your inbox.
I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.
My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.
I have something special for you guys today: a guest post by one of my favorite founders! This is the story of how Jesse Bray went from being an animator to the CTO of an AI startup, Hoplon.
Hoplon helps construction companies automate timecards and track their workforce using geofencing.
If you ever wanted to break into startups, Jesse’s story is for you.
Enjoy!
I Accidentally Became the CTO of an AI Startup
It started with a request for a few old computers.
People have asked me how I became the CTO of an AI company.
They usually expect some story about networking, recruiters, venture capital, or a carefully planned career move.
The truth is far less glamorous.
I was looking for old computers.
For most of my career I’ve been an animator, educator, and storyteller. I founded Zero to Animator, a program that teaches kids animation and digital art, because I believe creativity can change a child’s life. Watching a kid realize, “I made that,” never gets old.
As our workshops grew, so did the need for equipment.
So I did something simple.
I started reaching out to local businesses through our Chamber of Commerce asking if anyone had retired computers they weren’t using anymore.
One of those emails landed in Jake Gideon’s inbox.
Jake was building a construction technology startup called Hoplon.
He replied that he thought he could help.
At the time, that was the extent of our relationship.
A guy building an AI startup.
A guy trying to find computers for kids.
Nothing more.
Friendship Came Before Business
Not long afterward, I was invited to speak at a local Rotary Club about the nonprofit — why we were collecting donated computers and how we were using them to teach animation to kids who otherwise might never have access to those tools.
I asked Jake if he’d like to come with me.
He became my plus-one.
That lunch ended up having very little to do with computers.
We talked about ideas.
Business.
Technology.
Creativity.
Family.
The kinds of conversations that happen when you realize you’re sitting across from someone who genuinely enjoys building things.
Neither of us knew it then, but that lunch wasn’t the beginning of a business relationship.
It was the beginning of a friendship.
Creativity Is a Funny Thing
People tend to put creatives into boxes.
“He’s an animator.”
As if that’s all there is.
The funny thing is, I actually learned software development because I got tired of waiting for other people.
If I needed a tool for my animation work…
I built it.
If I needed a website…
I built it.
If I needed a learning platform…
I built it.
Over time I realized I wasn’t just learning to write software.
I was learning how to solve problems.
Animation taught me storytelling.
Software taught me execution.
Together they became an incredibly useful combination.
“I Can Build That.”
As Jake and I became friends, he’d occasionally send me a pitch deck.
We’d grab lunch.
Talk through product ideas.
Nothing formal.
I wasn’t interviewing.
I wasn’t consulting.
I was just helping a friend think through his business.
Then one afternoon he mentioned a customer.
They needed functionality Hoplon didn’t have.
I looked at him and said,
“I can build that.”
He smiled.
I don’t think he realized I meant immediately.
I went home and started building.
Not a mockup.
Not a prototype.
An actual enterprise platform that solved the customer’s problem.
Within days we had working software.
That same week, the customer signed.
Somewhere between building that platform and shipping it, something unexpected happened.
Without either of us really planning it…
I became Hoplon’s CTO.
Looking Back
When people hear that story, they usually focus on the software.
I don’t.
I think about the email asking for donated computers.
Because without that email…
None of this happens.
No Rotary lunch.
No friendship.
No late-night conversations.
No software.
No CTO title.
It’s easy to believe that careers are built through perfectly executed plans.
Mine rarely have been.
The biggest opportunities in my life have almost always started with serving someone else first.
Not because I expected anything in return.
But because helping people has a funny way of bringing remarkable people into your life.
One Final Thought
If there’s one lesson this experience taught me, it’s this:
Your résumé gets you introductions.
Your skills earn trust.
But your character is what people remember.
I didn’t become a CTO because I was looking for the title.
I became a CTO because a friendship gave me the opportunity to solve a problem, and I was ready when that opportunity arrived.
Funny enough…
It all started with a request for a few old computers.
I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.
My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.
You’ve got a Zoom in 6 minutes. Who the heck is this person, anyway? If only you had an agent to prep you for your calls. Well, I just built a great one you can use for free…
Most Agent Builders Stink — Back to the Drawing Board
I’ve used a bunch of Agent Builders. Most are unreliable.
They’ll get the prep for my meetings right once, but then stop sending the emails. They’ll pull the right information sometimes, other times not.
So I went back to my old standby, Grok.
Building My Meeting Prep Agent in Minutes with Grok Tasks
I told Grok Tasks to search through my Gmail and the open internet to prep me for my meetings. I want to know the following for every call:
Founder names and backgrounds
One sentence vision
Current traction
A few good questions I should ask
I want everything in concise, bulleted lists. I need Grok to deliver this to my inbox every morning, early.
I wrote a very elaborate prompt, which Grok actually helped me craft. That’s a useful trick: have AI help you write prompts.
Let’s see if this thing works…
Eureka! I’m Ready for My Meetings
Nailed it! Grok did a beautiful job of collating all the information from my email.
All I need to do is quickly read through this, take a couple of notes on questions I want to ask, and I’m ready to meet these awesome founders.
This agent did as good a job as an excellent human assistant. Even if I hired one in the Philippines, it would still cost me $40,000 a year.
I had two founder meetings on the slate today. Without this agent, it would have taken me around 40 minutes to prepare.
With my new agent, that’s down to 10 minutes.
Every Minute Saved Is Huge
A 30-minute savings every day is huge over time. We’re talking over 100 hours a year.
If I value my time at even $150 an hour, that’s $15,000 that just went into my pocket for a few minutes work. I run the agent for no additional cost — it’s covered under the $400/year X Premium Plus plan.
Best of all, automated prep encourages me to meet more founders.
If I know I’m going to have to spend a bunch of time prepping for each meeting, I don’t want to take as many.
But if the prep is easy, I take more meetings. This gives me a better chance of meeting the next Travis Kalanick or Brian Chesky.
What’s exciting about AI isn’t just automating stuff. It’s what we do with the time we get back.
Please Steal My Agent!
If you want to make your own meeting prep agent, here’s the Skill.md file below.
Use it as instructions in a Grok Task, run it every morning, and you’ll have the same setup I have. It should work in other agent builders too, although I haven’t tried that myself.
This is heavily tailored to an investor meeting founders. It might not work for other types of meetings.
Act as my expert-mode assistant for daily investor meeting prep. Always format information in bullet points. Follow these steps exactly and thoroughly for TODAY's date (use current date in America/New_York timezone):
1. Search Google Calendar for all events on today (use precise time_min/time_max for the full day in -04:00 or America/New_York).
2. Identify every meeting where I am meeting with a startup founder, company representative, or investor-related person. Skip purely personal events like workouts.
3. For each such meeting: extract the person name, company, emails, description, time, location/Meet link.
4. For each person/company: thoroughly search Gmail (use from:, to:, subject keywords, company name, has:attachment, recent dates). Read full messages with gmail_get_message. Download and analyze any attachments (decks, PDFs, docs) using available tools. Extract all links and browse them.
5. Use additional context from any recent self-sent briefs or related emails.
6. Go deep on the open internet: web_search extensively, browse company website, LinkedIn profiles of founders, Crunchbase if available, Twitter/X, founder videos, news, YC/Techstars pages, etc. Gather detailed info on:
- Exactly what the company does (concise description)
- All of the founders' full backgrounds (not just the CEO), past accomplishments, education, previous exits/roles (detailed bullets)
7. Summarize EVERYTHING for each meeting in this EXACT clean, concise, easy-to-read format (with space between sections, bold where appropriate):
[Time]: [Person] of [Company]
*[One-sentence what the company does.]*
*Key traction bullets (e.g. X paying customers. $Y pipeline. etc.)*
[Founder Name], [Title]:
- Detailed background bullet 1
- Detailed background bullet 2
- etc.
[Co-founder if any], [Title]:
- ...
Possible questions:
1. Smart investor question based on research
2. ...
3. ...
4. ...
5. ...
Make questions assume I am an angel investor evaluating for investment. Be concise overall. Prioritize high-signal info. If no meetings, say so clearly. At the end, if there are multiple, keep sections well-spaced.
Finally, after generating the full brief, also create a Gmail draft to myself (fpsantora@gmail.com) with subject "Daily Meeting Prep — [Today's Date]" containing the exact brief as the body (use plain text or clean HTML). This ensures I get it as email too.
Run in expert mode: be maximally thorough, use multiple tool calls in parallel, cross-verify facts, dig into attachments and links deeply.
Grok Tasks Gets an A
Grok tasks did a beautiful job here. I’m giving it an A!
The only thing I’d change: the email makes me click “Continue reading”, which takes me to the Grok app. I’d rather just see the full text in the email.
But this is a minor point. I’m in Grok all the time anyway!
Wrap-Up
No matter how much you love your job, parts of it are a drag. AI is great at automating the scut work.
Now I can focus on what’s actually fun: meeting great founders and writing checks!
I used this app every single day to dictate to my computer, I’m even dictating this text using Wispr Flow! It’s way better than Apple’s native dictation.
My productivity is up about 25% since I started dictating rather than typing. I’m also less tired and stressed.
Tremendous 