Science nonfiction

The startup playbook is being rewritten in real time. AI lets tiny teams build, ship, and grow faster than any generation before, compressing years of milestones into months.

Join Alex Kantrowitz, Founder of Big Technology, and Vanta's SVP of Engineering, to learn the patterns they're seeing before they become conventional wisdom.

You'll learn:

Register to save your spot. Can't make the time? Register anyway, and you'll get a link to the recording.

If Tyler from 12 months ago saw what Tyler from last Thursday was doing, he'd have thought it was 2030. But it's not 2030 — it's 2026, and the future is here right now.

Let me explain.

Each quarter I give an hour-long presentation to the entire company recapping the prior quarter and previewing the upcoming one. It's a forcing function to grade our progress and realign everyone so we're all swimming in the same direction.

I've been exclusively using Figma Slides for the past year or so. If you're still using Google Slides, you have no idea what you're missing. And if you're still on PowerPoint, ngmi.

Before I dig into what happened next, let me take a quick detour and share our current AI setup at beehiiv. I'm far from the most AI-pilled person on your X feed, but this context will help explain the rest of the post.

I have Claude up on my desktop pretty much at all times. We use Fellow at beehiiv as a notetaker that joins every meeting and saves all the transcripts and recordings (Granola, Fireflies, all sort of the same). Then we use Claude connectors to sync all of the following via MCP:

I'm forgetting a few, but the point I'm trying to make is that context is king. Whether you're using ChatGPT or Claude — it's only as useful as the context you feed it. To make the most of all that sexy super intelligence, Claude has unfettered access to every conversation and piece of data across our entire company (minus sensitive user data like PII, passwords, and payment info, of course).

Ok, so back to last Thursday. I set aside 10 hours to build the Q4 presentation I was scheduled to give on Friday. To my surprise, Figma's MCP (finally) started supporting Slides.

Rather than carefully crafting one slide at a time like I had my entire life up until that moment, I just dropped in bullet points describing the data and charts I wanted on each slide. Sometimes I dropped a link to a specific doc (for example, a new product spec), or just mentioned a specific Slack channel (like #topic-podcast-gtm).

It took about an hour to transfer the narrative in my head into a bunch of slides that looked like the one above. More or less, I had a completed outline of a presentation. Then I fired up Claude Code, fed it my previous quarterly presentations, provided some design direction, gave it access to Figma Slides, and told it to get to work.

In about 30 minutes, it one-shotted a deck that was better than any I have ever put together (and I am damn good at making decks). On one slide I simply asked for a chart showing the growth of subscription revenue on the platform. It took my notes, queried the database for the raw data, cross-referenced the relevant Slack channels and product meeting transcripts, read our knowledge base and product updates, and created something that looks like this…

A few weeks ago I shared a post from Lenny's Newsletter (written by former Apple design lead Anshu) that explained how to better use AI for design. It totally blew my mind. This whole time I'd been prompting Claude to "build me a landing page" like everyone else, then getting frustrated with the cookie-cutter results.

To summarize the post: AI is trained to make the safest choice at every step, which is why most AI designs look the same. The fix is to push it off those defaults — use random strings as seeds for truly unique designs, write wild prompts steered by your own taste, and generate real images instead of gradients.

But most interesting to me was the idea of launching subagents as critics to provide an objective critique of whatever you're building. You can even use those subagents to recursively improve the design until the critic is satisfied. Genuinely mind-blowing.

The principles in that post apply way beyond design. I now routinely launch subagent critics on everything I do to get an objective "second set of eyes" on the output and keep pushing the boundaries of what I'm producing.

Back to the presentation. After the initial one-shot, I launched two critics: a design critic to recursively improve the aesthetic of the slides, and a narrative critic to make sure the presentation nailed my objective and the talking points I'd set out to deliver.

As a final step, I went through each slide one by one and offered my own critique. Changing the copy, swapping data, improving the layout, etc. Mind you — I'm using Wispr Flow and just talking to my computer in plain English as it takes my every word in stride and makes edits instantly.

It's not that I didn't work hard and put serious brain power into the final product… I did. All in, it took about five hours to perfect it. But during those five hours I was legitimately just talking to my computer, spawning virtual data analysts and designers in real time.

The outcome is the best deck I've ever laid eyes on (no joke). I genuinely can't believe I used to create shapes with my mouse and drag and drop things around the screen like a fucking caveman.

But that's not all. While I was waiting for Claude to make edits, I decided to download Muse and see what all the hype was about. Unless you've been living under a rock for the past couple months — the newest frontier is personal agents. xAI launched Grok Bot, Meta launched Muse, and an upstart called Instinct launched its own personal assistant.

A personal agent has always been the holy grail of super intelligence. It's been depicted in sci-fi movies for decades, and with each new advancement in AI we've inched closer and closer to it becoming reality.

Well, we're a whole lot closer to sci-fi becoming science nonfiction than people think. While Claude was cooking on my presentation, Muse started sourcing candidates for a new role and helping me plan an upcoming trip.

For the new role, I had Claude give Muse a synopsis of every conversation I've had with our CMO about the position, and fed it a bunch of candidates who were (and weren't) a good fit to calibrate. Then it went off to work.

For the trip, my friends and I are planning to go to Punta del Este in Uruguay for New Year's Eve, but holy shit is it difficult to navigate. They don't really use Airbnb, so it's all handled through local agencies. Between housing, flights, concerts, and logistics, it was honestly making my head spin. So Muse started doing research and contacting those agencies on my behalf.

By the way: if you have any connects in PDE, shoot me a reply. I appreciate how hard my little Muse is working, but I'll take all the extra help I can get.

Somewhere during this experience I sort of put it all into perspective. I'm literally talking to my computer to pull data and update slides while my phone is simultaneously sourcing candidates and feeding me housing options for an upcoming trip.

The future is here.

And we're actually just getting started. Two weeks ago Meta announced its new wearables at Connect 2026: new VR glasses, two new Ray-Ban models (including its first pair without a camera), and a pendant-style AI device called the Muse Charm. All of which connect to our homie Muse. I'm still not convinced glasses are the ideal form factor, but they're certainly getting better with each iteration.

Muse is fun to use… but right now it's only as smart as the limited context I've fed it, and it's trapped inside my phone.

Squint and you can see a world where the glasses get stylish enough that I just have Muse on me all the time. Going back to context being king — it could literally see and hear every interaction I have, and I could activate Muse on demand. It knows who I am, where I am, what I'm talking about, my calendar, my priorities, and everything else going on in my life.

Meanwhile, everyone likes to give Apple shit for being late to launch anything relevant in AI, and Siri (deservedly) is the butt of every joke. But I actually think they're incredibly well-positioned.

Instead of spending billions on inference and data centers to build models of their own (which are quickly being commoditized by open source models)… they own the hardware in the real world and have access to the billion wealthiest people on the planet.

They're rumored to be releasing AirPods with cameras in late 2027. Same as the smart glasses seeing and hearing everything you're doing — the same can be said for smart AirPods you wear all the time. And AirPods are arguably already the most successful wearable ever (an estimated ~$20B in annual revenue).

Smart AirPods with cameras plus a personal agent… and now you have the plot of Her. But Her isn't decades away, it's right here. All of the trends are converging on it as we speak: smarter models, lightweight wearables, personal agents, better battery life, etc.

Sure, there are a lot of privacy concerns with this future. Back in the day, your parents could get blacked out at a bar and take a shit on the floor, and unless you were there to witness it, no one would ever know. Today everyone has a smartphone (i.e. a hi-res camera and recording device) and access to social media. The internet is a paper trail of everything you've ever said and done.

I'm not here to argue whether or not we should compromise our privacy for the performance gains and conveniences this superintelligence can deliver (at least not in this post).

But I am here to argue that the future is now. I feel like I experienced something new last week, and I just can't go back to the before times. I don't think anything is ever going to be the same again.

What an exciting time to be alive.

If you enjoyed this post or know someone who may find it useful, please share it with them and encourage them to subscribe: mail.bigdeskenergy.com/p/science-nonfiction

If you don’t mind the building swaying back and forth due to the wind, this office would be pretty chill.

Think you can generate a better office? Reply with your submissions 📨.

Some of my favorite content I found on the internet this week…

Turn on, tune in, drop out. Click on any of the tracks below to get in a groove — each selected from the full Big Desk Energy playlist.

Share this newsletter with your friends, or use it as a pickup line.

👉 Your current referral count: {{ rp_num_referrals }} 👈

Or share your personal link with others: {{rp_refer_url_no_params}}

Enjoyed this newsletter? Forward it to a friend and have them signup here.

Until next Tuesday 🕺🏽

📥 Want to advertise in Big Desk Energy? Learn More