Our new AI code policy: No AI

While most companies out there are talking about how they're using billions of tokens, and how their developers are being 10x productive, and how they're firing half of their company because they don't need them...

... We're going the opposite direction. We're making Screenshotbot AI-free. i.e. We will not write any line of code in AI (I'll define the exact policy at the end of this post).

why would you do this!!!???

Just to be clear, this isn't an altruistic goal. We're also not making a statement on the quality of AI code. For the 10 months or so, I was heavily using AI, and I can confidently say AI improved my productivity significantly.

But we're a business after all, so why wouldn't I want a tool that improves my productivity?

It turns out, Enterprise Sales is inherently human. It's going to be a while before your AI agents make purchasing decisions.

And if you're part of the decision making process for a potential enterprise customer---whether you're evaluating, whether you're writing the check, or whether you're reviewing the documents---it rubs you the wrong way when you have lost colleagues (and friends) to AI while we here are boasting about how we use AI and demand your money at the same time.

Oh no no, we don't want to be an NFT art: We want Screenshotbot to be a da Vinci: meticulously hand-crafted, by humans, for humans.

It's not the way people do it!

Sometime in 2015 I got into woodworking.

But living in New York City/Jersey City, I really couldn't have a workshop in my apartment.

... or could I?

At some point I found the videos by Paul Sellers. He calls himself "A lifestyle woodworker" (I don't remember if he called himself that before: but he certainly was doing woodworking as a career, not just as a hobby).

But while other woodworkers used power tools, Sellers used all hand tools. Hand tools aren't just a slower version of power tools, it's a very different skill set. It's not cheaper: in fact it's more expensive than power tools: they're typically hand-crafted precision tools, and there are a lot of them. But it was quiet, it was calming, and it was fun.

I didn't get too far in my woodworking hobby and at some point during the pandemic I sold most of my tools when they were in high demand.

But it certainly taught me a kind of patience and calmness that have served me quite well in the startup life.

They say startups are a marathon, not a sprint. I don't think I'll survive the startup marathon if I'm not having fun along the way. And despite all the productivity gains from AI, AI is not fun.

But AI is fun! Look at all these things I can build!

I hear this argument all the time. And I have a story for this too.

When I joined undergrad, we had a computer lab with workstations connected to the network and we had this typing game that stored leader-boards on the NFS.(It's called typespeed, you can still apt-get install it if you want to try it out)

This dude Vipul was on the top of these leader-boards.

I was a bored undergraduate student, and one summer I wanted to beat Vipul. I wrote a script that read the output of typespeed and just wrote it back into stdin.

And soon I beat Vipul, and I took the top place on the leader-board.

Was this fun?

Maybe. Writing the script was fun, seeing my script work flawlessly was fun. But I wasn't really playing the game anymore, and that wasn't fun. And it certainly ruined typespeed for me, and possibly many others at college.

When I decided to do a startup, I don't think it was necessarily to make boatloads of money. I had a reasonable amount of savings and my wife and I had figured out a lifestyle to make the savings last for a very long time (we don't have kids). I could've made a lot more money staying in corporate jobs over the last few years.

I think I'm working on startups because I genuinely have fun doing it. Yes it's stressful. Yes I'm constantly trying to get on the leader-board. But there's a fun and excitement in that process.

You're making me feel guilty!

I really need to emphasize that I'm not judging you for using AI. I bet all of our customers, present and future, use AI significantly.

I recognize that if you're in a corporate job, you don't have the flexibility like I do to make a decision that intentionally reduces your productivity, just because "fun".

But I do hope that if you're in a place to make purchasing decisions for Enterprise SaaS, you'll give added weight to human built startups.

Our new AI policy

I've spent the last two weeks AI-free, and I've figured out a sweet spot for our AI policy. I'll explain the policy, and I'll explain my learnings from the last few weeks that guided this policy. To be clear, we're not zero-AI, at least not yet.

• Absolutely no code should be written with AI: This is the over-riding policy. In the last few weeks, when I wrote code by hand I ended up building debugging tools that I wouldn't have with AI. These debugging tools made it a lot easier to iterate on an otherwise slow task. It also forced me to reduce noise from logs, because manually going through logs is a lot more work than asking Claude to.
• Absolutely no documentation or content written with AI: This blog post was written by hand.
• It's okay and expected to build AI integrations: such as MCP servers, or other tooling that our customers can use. However, no AI can be used in building these integrations, only for testing it.
• It's okay to use AI to learn something new: As long as no AI generated code is committed back in.
• It's okay to use AI for one-off sample projects, as long as AI use is disclosed: Screenshotbot has integrations with a huge number of platforms, and it's impossible for me to be an expert in everything. AI will help me build sample apps for these purposes.
• [still considering] It's okay to use AI to debug a time sensitive major issue: I had one issue where I had to respond to a customer last week, and I had limited time since I had to head out to a social commitment, and I used AI to help me out. But this is where it gets tricky: when I forced my self to skip AI I ended up building a debugging tool that would help me debug similar issues better in the future.
• It's okay to use AI for non engineering things that are not my expertise: I used AI last week to debug an accounting issue with my books for taxes. I suspect I would also use AI to build artwork in the future, but ideally I'd like to just get better at a skill than rely on AI.
• It's okay to use AI for sales or marketing research: but not for email content.