Tokenmaxxing

Every founder, creator, and media operator should have a newsletter. And any time someone asks me how to get started, I send them to my buddy Nathan @ The Feed Media.

He leveraged his newsletter to go from zero to $1M in revenue in 10 months with 920 email subscribers (on beehiiv, obviously).

Nathan’s hosting a free workshop next Wednesday at 12pm & 7pm ET for BDE readers walking you through how to to turn a tiny audience into a profitable business:

How to select a niche and write your first newsletter
Hit 1,000 high-quality readers without paid ads / a big audience
The email flows that turn readers into buyers

I couldn’t recommend it enough (and it won’t cost you a dime). You can thank me later.

Save your seat

P.S. There’s a huge bonus for live attendees - find out more on the workshop page.

Every CEO is talking about how their company is adopting AI to improve their business. Very few share the behind-the-scenes of what that actually entails, or quantify the progress.

Fortunately for you, I’m not like most other CEOs.

It’s been four months since I began to heavily push AI adoption here at beehiiv. Today 85% of employees use AI daily. We’re also spending about $25k per month on tokens.

There’s a lot that’s working well, and some things that aren’t. Let me give you an inside look at the state of AI adoption at beehiiv.

It all started with this email I sent to the team back in February:

❝

I give all of this context with the caveat and assurance that we have an extremely healthy business with $30M in the bank and 3+ years of runway.

That said — we are unprofitable and burning ~$700K /mo. We have been stuck in this venture flywheel where we grow and burn capital until we need to raise more money, which reduces the optionality and control of the business. All of this is super common for venture-backed businesses.

A lot of that is due to our business operations. Every additional X thousand new users requires us to hire more people to support that growth (support team, customer success, DNC, security, engineers, PMs, designers, etc.).

It's an inevitable flywheel that plagues every growing organization, which compounds and leads to higher costs and more bloat.

But if we want to break free of this venture flywheel (we unanimously do), we need to eventually stop linearly scaling costs and instead find leverage to increase our output exponentially with the people we already have today. Fortunately, the latest advancements in AI truly are capable of augmenting many of our existing processes (and are getting better seemingly each week).

I've never been more confident that we could scale this business to $100M+ in revenue with just the 130 people we have in seat today. But for that to be possible it will take a lot of trust, buy-in, and open-mindedness from everyone.

We aren't going to show up to work tomorrow with all of us as AI experts, rather, it's going to take some time to develop these skills and familiarize ourselves with these tools. But the sooner we start, the better.

Transparently, I want to avoid unnecessarily expanding headcount, solely because we aren't using the tools available to us today to their full potential.

If engineers used to complete 5 tickets per day, with AI they should be able to complete 15.
If customer support answered 50 tickets per day, with AI they should be able to answer 100.
If customer success could manage 40 customers at a time, with AI they should be able to handle 80.

This isn't optional — this is the path forward. We are not expanding headcount until we have exhausted every possible solution with AI to augment our current output.

Personally, I find this to be an incredibly exciting opportunity for everyone at the company. We get to collectively tinker, experiment, and learn how to use what might be the most transformative technology of our lifetimes. We have an incredibly smart and supportive team and have created a safe space to trial and fail in an effort to help each other get better.

The top performing companies are already way ahead on AI adoption. We admittedly are a bit behind (we're currently sitting between the 50th and 75th percentile), but the #proj-ai-club is the first step to catching up.

Our goal has always been to become a top 1% company, and I'm excited for the team we have today to push us forward in that direction.

So tactically, what did we actually do?

First, we onboarded every single employee onto the company Claude account. Then, we created the AI Club (a dedicated Slack channel) to be the home base for all conversations related to AI. Here was my initial post…

Granted, my idea to neatly organize messages by using different slackmojis only lasted a couple of weeks before people started to just post whatever. No regrets, my OCD appreciated the effort.

The early days were filled with people asking questions and making requests to make the tools more powerful.

And also filled with moments of delight when people would discover what was possible, no matter how seemingly small the accomplishment.

The goal of the Slack channel was to create a space where people could ask questions, share updates, and learn from each other. Simply seeing what other people were building was helpful to inspire new ideas. Seeing Dan use Cowork to organize his inbox could trigger others to use Cowork for similar tasks.

The Slack channel was a huge success and helped build momentum, but Claude was still too intimidating and confusing for a lot of people on the team. Learning how to use Claude Code, granting it access to certain tools, managing API keys, etc.

So we launched AI Office Hours on Mondays. It was an opportunity to unblock people and offer engineering support to answer questions and guide people through.

We also scheduled AI Show and Tell on Fridays so people could demo their projects to the entire company. I intentionally scheduled it on Friday afternoon in hopes that these demos would inspire people to tinker and build things over the weekend.

One thing I was very intentional about from the beginning: I didn’t care about efficiency or cost-savings.

I’m sure there were tons of wasted spend doing simple tasks inefficiently or making large requests to the database. But as soon as you start over-explaining the differences between the models and other nuances around cost optimizations, you totally lose people and intimidate them.

My focus was on adoption. I wanted people to experience their first “a-ha” moment with AI and get inspired to build more. I’d much rather have too many people using AI and it costing us too much than no one using it because of their insecurities. We could always focus on cost optimizations later.

60 days after launching the AI Club, I sent a survey to better understand AI usage and proficiency on the team...

The survey results were clear:

Employees feel much more comfortable using AI. The average comfort jumped from 5.3 to 7.4. The floor jumped from 1 to 4.
85% of employees use AI at least daily. One-third of the team claims to use it “constantly.”
Two-thirds of the team saves 4+ hours per week. A quarter of the company claims savings of 8+ hours per week.

Last month we had our company offsite in San Diego. We hosted a 36-hour AI Hackathon that the entire company participated in. We broke into small teams and ended the week with “demo day” where more than 40 teams presented their projects.

The projects were wildly impressive. Many of them carried on beyond the hackathon and have since become live features in the platform or internal tools that our team uses frequently.

Outside of company hackathons, employees are frequently building new tools and processes in their day-to-day as AI proficiency continues to increase.

Jacob on our Customer Success team vibe coded an entire platform that manages all the customer relationships with our enterprise users, and automates reminders and other menial tasks.

This is the quintessential example of the kind of unlock that's possible with AI. This simply didn’t exist before, now it does, and our team is capable of so much more because of it.

It’s good enough that we would actually probably pay for it. But rather than having to depend on another company’s roadmap, Jacob can make updates to its functionality whenever the team’s needs change.

Another random example: a few weeks ago someone on the team scrolled past this post on X from Tibo.

Jess, who has been manually submitting dispute evidence for years, decided to build something similar for our use case. It went live last week and already seems to be working super well.

All that to say — I’d still consider the AI initiative at beehiiv to be a work in progress. There have been a lot of great outcomes:

Almost every team has their own version of what Jacob built for the Customer Success team: tools and dashboards custom-built for them that didn’t exist before.

Most of the team uses AI several times a day to perform tasks that were previously manual (or impossible). I can’t put a percentage on it, but it’s safe to say output per employee has definitely increased.

Related — due to the increase in output and efficiencies with some of these tools, we've deferred hires we otherwise probably would have already made. We're doing more with less (which should always be the preference).

But it’s not all good in the hood. A few caveats about the progress:

We’re spending about $25k per month on tokens. I mentioned how cost optimization wasn’t a focus of mine in the early days, but we may need to start prioritizing that sooner rather than later.

AI champions have certainly emerged on each team. Which is great to an extent, but also may discourage others on the team from building stuff. Using the example from above, now that Jacob has built the dashboard, there’s less of an incentive for others on his team to build or contribute to it.

Some engineers have used agents to 10x their output. Others use AI more modestly. There are still some concerns on code quality as well, and occasional pushback on how to use it on the engineering team.

Related — I don’t necessarily think our shipping velocity has increased all that much. It’s easy to vibe code tools that no one uses, but much harder to ship production code to 60K active users. I’d classify AI’s impact on our shipping velocity and output as “marginal.”

I meant what I said in that email to the team: I do genuinely think it's the most exciting time to build. And I think we can get to $100M+ in revenue with our current team of 130.

I also think it’s an incredible time to be an employee, on any team, and have the cloud cover to tinker, experiment, and contribute to the company in a bunch of new ways.

Claude, ship this newsletter. Make no mistakes.

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/tokenmaxxing

Credit: Henry Winslow

Shoutout Henry for the reader submission 🫡.

As you can tell, Henry likes mushrooms. He writes Tricylce Day which is one of the most popular newsletters about psychedelics. I probably wouldn’t be too productive in this office.

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

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

The beehiiv Summer Release 2026 is July 16th and going to be the best 30 minutes of your summer → RSVP.
Anthropic dropped its 2026 Agentic Coding Trends Report — "surge" staffing, non-engineers shipping real tools, the wall between people who code and people who don't quietly coming down. Basically everything I wrote about up top.
Evan Spiegel told John Collison that over two-thirds of Snap's new code is now written by AI. What comes after smartphones?
David Senra covers The Creative Genius of Rick Rubin on Founders podcast.