“Naively optimistic” – this was my opening to a blog post I wrote for Errant Science five and a half years ago, when Overleaf was approaching its fifth anniversary. Five years later, Overleaf is now ten, and after leading the team there for close to a decade I’ve recently moved into a new role at Digital Science. So I figured it was time to write the follow-up to that blog post, and Matthew kindly agreed 😊

That original post was published on the last day of July 2017, only a fortnight after a pivotal moment in Overleaf’s growth: on the 17th July 2017 the Overleaf and ShareLaTeX teams joined forces to create a combined community of over two million users. It’s hard to believe that event was over five years ago, and I had to explicitly count it out when my co-founder recently pointed out that more than half (!) of Overleaf’s existence is post-merger.

Today, a decade after we first launched our online collaborative writing platform, the world stands on the cusp of a new revolution; the explosive rise of ChatGPT (and now GPT4), and before that, Dall-E and StableDiffusion for image generation has created a whole new field of content creation based on prompt engineering – being able to tailor your requests to these new large language model (LLM) based systems. The AI field itself is also seeing a rapid growth in new industry-led research (e.g. OpenAI, Microsoft Research, Meta, Google), conceptual observations (e.g. theory of mind) and new services (e.g. ScribbleDiffusion) all looking to further develop, understand and ultimately to exploit the potential seemingly on offer. 

This creates a huge opportunity for innovators who can use this new technology to solve a real-world problem – ideally solving a problem that they themselves encounter daily, as there’s no better early validation than finding yourself naturally using your prototype/MVP because it makes your life easier. Once you have an idea for how to solve a problem, and some evidence that this is a problem a lot of people have, the next simple step is to found a start-up company, right?

Yes and no. I’ll just come out and say it: running a business is hard work. I know that sounds like a “well, duh!” kind of statement, but it feels worth saying, because when we talk about start-ups we often focus on the exciting, glamourous opportunity it presents, rather than the hard parts that come with it. Perhaps I should expand it to say:

which, as Mark Hahnel pointed out to me (he read an early version of this post!), is a variation on the saying “ideas are cheap, execution is expensive”.

However you phrase it, this is where your naive optimism is required; if you don’t have a glass-half-full attitude, and the ability to focus on what needs doing right now rather than worrying too much about the future, it’s almost impossible to make a start-up a success, at least in my experience.

In the case of Overleaf—having founded it as a new SaaS start-up with John Lees-Miller in late 2012—the early days really were as fast-moving as you’d expect for a new company. We’d quit our previous jobs to work on it full time, and it quickly became all-consuming; when you’re just two people working on something, the more hours you put in really do translate into more direct output in a way that’s just not as first-hand once a business has scaled.

Our first setback was failing at the interview stage at Y Combinator, which I covered in part one of this blog – it’s where the optimism had to kick back in (and it did) as soon as we got back to London following that trip.

We went on to be accepted onto the relatively new Bethnal Green Ventures (BGV) start-up accelerator programme in London, which focused on “Tech for Good” – businesses aiming to have a transformative societal impact as well as a commercial opportunity.

And, whilst we were in a strong position at the time – we had users! – we were still very green when it came to knowing wh