Between 2022 and 2025, I dedicated myself to writing Framed. I deprived myself of certain comforts. I still don’t own a couch or a television. I also avoided jumping into new side hustles or business ventures (except for a bit of sportswriting).

Now that I’m finally coming up for air in March of 2025, I’m ready to admit that I really missed out.

Although people are certainly still buying fake followers on Instagram, the landscape for weekend side projects and micro SaaS businesses has changed quite a bit. People are making money in different ways. There are new opportunities and a lot of new competitors. My book addressed that in the late 2010s, there was a wave of digital entrepreneurs who bootstrapped businesses thanks to innovators like Stripe. Stripe made processing subscriptions—critical to a SaaS business—really easy.

Now, in the “twenties,” there’s a new wave of innovative tools. In broad terms, AI. These tools are so powerful that they could come close to commoditizing the entire Shark Social operation! “Shark Social” was my Instagram growth business in Framed. An AI agent in the right hands could have been used to handle pretty much all aspects of my social media business that generated hundreds of thousands of dollars in revenue.

Aspects of Shark Social (from 2017) and the impact of AI (from 2025):

1. Front-end website design with SEO-optimized content: Easier implementation, but wouldn’t have saved us much time because we still would have written the copy and sourced the images ourselves.

2. Integration with Stripe and signup flows: Easier implementation, would have saved a younger Tim many hours of development time.

3. Customer retargeting and cancelation: Easier implementation, would have saved plenty of hours of development time and support/maintenance time.

4. Ticketing system and customer support: Probably would have saved us several hours per week, considering how much effort we put into writing documentation and support scripts.

5. Proxy management: Could have saved us ten hours per month and quite a bit in development costs, compared to my end-game solution, dubbed “Loincloth.”

6. Burner account management: Although burner accounts (throwaway accounts used for testing) could have been managed by AI, or AI could have written code to better manage them, procuring burner accounts and setting up tests would still be best left to humans.

7. Marketing: Wouldn’t have made a difference, we had a productive, low-cost writing team.

8. Off-the-shelf automation solutions (“FollowLiker”): With access to open source client libraries and a well-written technical spec, a fully-featured desktop app like FollowLiker could have been developed by AI.

9. The automation system (“Mako” — alias from my book): Same as above—not a trivial implementation because my system was more distributed, but there is nothing—nothing—I did in early versions of Mako that AI couldn’t code right out of the box

10. Debugging, testing, and problem solving: This space required a ton of domain knowledge and creativity, and was further resource constrained by burner accounts and proxies. This is where AI would have failed. However, I admit that using AI to work through certain problems would have been better than trawling BlackHatWorld.

11. My advanced automation system (“Mako 2” — alias from my book): AI would have saved me at least ten hours of coding (and the end result would have been much cleaner), but it couldn’t have saved me all the painstaking time testing and generating (what I feel were) original insights. However, Mako 2 was never run for all customer accounts, so, yes, I suppose the whole business could have been created and run by AI agents.

I can’t say whether AI (either in design, development, or operation) would have been “better” in every case, but I’m admitting that this lucrative business could now be implemented by pretty much anyone with the domain knowledge.

I believe that the most consequential project I’ve ever worked on could now be implemented by one person, in less than a month.

What’s Vibe Coding?

Over the last few months, AI-assisted “agentic” coding tools have proliferated, and the term “vibe coding” has gained momentum.

They call it “vibe” coding because it’s not really software engineering. It’s more like listening to EDM music while basking in a summer breeze. The operator is “vibing” while expressing in plain English what the intelligent coding tool should be doing and then periodically checking its work. The engineer isn’t really coding and the AI agent has the productivity of roughly three mid-level engineers.

I figured that if I’m going to be competitive in this era of entrepreneurship, I must learn how to “vibe code” for tasks that I’d otherwise need to outsource (or spend an entire afternoon on).

I purchased a domain name, loaded up like $30 worth of credits for the best “fast” AI models, downloaded Cursor, and began vibe coding a “hub site” for runners. The site would include product reviews, book recommendations, useful calculators, and perhaps even analysis tools using biometric data.

RunThen.com

After a week of vibe coding, I’ve launched RunThen.com.

I wrote zero lines of code myself.

I wrote zero lines of AWS infra code myself.

I wrote zero lines of CI/CD and containerization code myself.

I wrote zero lines of database and CMS schema code myself.

Cursor wrote over 9000 lines of code for RunThen, which included:

• A game that measures reaction time

• A calculator for running pace (in my opinion, the best one on the internet) and a calc for time math

• User registration with email (with verification) or with Google

• Cloud infrastructure (ALB, ECS, ECR, RDS, etc)

• Editable user profile pages

• Contact form mail forwarding (with ImprovMX)

• Book and product review pages (with Sanity CMS)

• Integration with Google Analytics, OAuth, and HotJar

• SEO optimized metadata, structure, and SSG

• Decent quality written/filler content

• Captchas, throttling, and other security features

• Fully responsive design

• Clean(ish) linted code

• Dockerfiles, GitHub pipeline for deployment, and cloud best practices for security and reliability

If I was to copy and paste these features in asking a developer for a quote, the absolute lowest projection would be around $3000 worth of dev time. Frankly, it could be that much just for implementing signup and login at a moderate level of complexity! Of course, ChatGPT is validating these numbers, but nobody in their right mind would do this for a couple hundred bucks!

It cost me $30 in “fast request” credits. Notably, 20% of that fee (and 33% of the time I spend on this) was spent unraveling absurdly incorrect code created by the agent.

When coding a weekend project, implementing an entire user registration/login process from scratch could take an entire day. Spending most of my career on “special projects” and away from full stack engineering, I’ve actually never coded a user signup flow of this complexity.

If one step of the project takes an entire day, and that day happens to be Sunday, then you’re out of time without having done anything useful! So, in the past, I’ve always skimped on things like user registration or straying from Wordpress’ CMS. Vibe coding has unlocked this crazy workflow where I can bring ideas to life quickly with pretty much no cognitive load and at a trivial cost.

Vibe coding is vibe coding because there is no cognitive load. An experienced engineer with an agentic code assistance tool can be 5x as productive without ever truly getting “stuck” and with having the fortitude to “vibe” for much longer hours.

RunThen is not a serious business venture. However, hosting it will cost about $100/mo. If I keep building calculators and product reviews, I’m confident that I can break even. I could probably reach $400 in monthly recurring revenue by the end of the year—mainly through affiliate sales—if I really pushed myself.

If RunThen doesn’t earn money, I’ll turn it off. The real value is that I now have templates and sample code for many of the functions that are frustrating to develop for micro projects, such as CMS integration, AWS infrastructure (Terraform), and a user registration system.

Vibe coding dangles more grandiose opportunities, though. What if I pivoted from calculator site to…social media site? I could vibe code a Strava competitor. I don’t think it’d be particularly competitive, but those “Whartonite Seeks Code Monkey” dreams mentioned in my book are now easily achievable, perhaps by the Whartonite.

Continuing to Vibe

My takeaway from this is not that I’m going to create what I deemed, “The next _______” in my book, it’s that now there’s a trivial barrier between ideating projects and implementing those ideas. My weekend output can now be just as professional as that of a dedicated full stack team, at somewhere between 1/100th and 1/1000th of the cost.

Don’t get me wrong, that’s with tons of asterisks, and it comes after almost a decade of writing code, configuring systems, and creating written content. I’ve spent the last few years missing out, and now it’s time to catch up.

Do you vibe code? Do you want me to write more about vibe coding projects? Let me know!