Both started because I was repeatedly annoyed by a similar feeling: working through a process that felt unnecessarily difficult and broken. With recipes, it was forgetting ingredients, burning food, and repeating the same boring meals. We indexed the world’s recipes and built a seamless workflow on top of the content after we structured the long text files into structured data using lots of Machine Learning algorithms. Users loved it. We got 4.9 ratings, Google Play and Apple Store awards, and Samsung acquired the business after we were lucky to have three large companies send us inbound interest. With Samsung, I had the chance to scale from about 30 to 120 people in nine months and see how a global organisation thinks about product, markets, and risk. That experience is now baked into how I build.

What people talk about less is what happens after a successful exit. There’s a quiet, awkward gap where you’re supposed to be “happy”, but inside I felt I had just sold and lost part of myself. Because I had founded Whisk as part of appearing on the BBC Apprentice, getting to the final and pitching it to Lord Sugar, my personal identity ended up being more closely tied to the business than felt comfortable in the low times. After selling Whisk to Samsung, many people considered me to be a “successful founder”, and that felt good. But a few years prior, we had many moments of almost running out of money, and I think that was even more stressful because so much of my personal identity was tied to Whisk and vowing on national TV “to prove Lord Sugar wrong”.

I hired a coach partly to work through what to do next in my “career”. The question we came back to was: what actually makes you happy? My answer: being able to create things (startups, art, gardening) and do it with people I like spending time with (a great team). I’m not good enough at the art of gardening, so the answer for me lies in startups.

So, as I considered starting another company, there was definitely some “second-time founder anxiety” at play… people were watching. I now wanted to make the next thing “bigger” or more impressive than the last one. Ultimately, two things motivated me to try again. Firstly, I imagined how you’ll feel in ten years if I had never tried again. Secondly, as AI accelerated between 2023 and 2024 with the release of GPT-4, I realised that the huge jump in reasoning and capability depth meant really exciting new software opportunities were going to be possible.

So I considered problems that really annoyed me and landed on legal. I’d spent over a million pounds on legal fees. The work was important, but the process felt slow, opaque and painful. Many founders I spoke to had a similar story. And that was the itch that became GitLaw.

On a technical level, legal contracts looked strangely familiar. Recipes and contracts are both long, dense documents written for humans, not machines. Unstructured, they are hard to automate. Once you add structure, you can add workflows on top of it. GitLaw allows us to generate first drafts from trusted templates, analyse and review existing documents, surface the right clauses to negotiate, track obligations, and automate things like signatures and reminders.

Community is another parallel. Whisk eventually became a mix of publisher content and user-generated recipes. Hundreds of thousands of people shared their own dishes. With GitLaw, we’re building an open library of vetted templates, the kind of documents I wish I’d had when doing my first investment round, option scheme or commercial deal. The AI doesn’t start from a blank page; it starts from market-standard contracts contributed and reviewed by real lawyers, then adapts them to each use case.

Both startups are “AI-heavy”, but that means something very different today. With Whisk, we had to build a lot of proprietary models. We bought GPUs from the US (we literally flew to the US to buy the latest graphics cards and ran the models in our Birmingham-based office), trained systems ourselves, and dealt with a lot of infrastructure pain because there was nothing off the shelf in 2012. Today, world-class language models are accessible via API in minutes. That’s amazing for founders: a team that’s raised a few million can access capabilities that used to require hundreds of millions. The flip side is that models are no longer the main moat. Differentiation now comes from workflow design, domain context, data quality, and how good your evals are.

There are a few broader lessons that carried across both journeys. First, distribution beats cleverness. Whisk’s growth inflexion points came from unexpected distribution events, like TV exposure and publisher partnerships, not just “better algorithms”. At GitLaw, most traction so far has come from social, founder-to-founder referrals, and the credibility that comes from being open and transparent about what we’re building.

Second, written knowledge compounds. Whisk has been a leader in distributed work from 2014 onwards, a long time before it became “the norm” after COVID in 2020. We instilled best practices to always document thinking so that others could learn and digest it. With GitLaw, we’re remote and async by default, which forces documentation. But today, that isn’t just important to make our distributed team work well together; it’s training data. Specs, decisions and playbooks are exactly the artefacts AI agents can use to uplevel automated work later.

The lesson across both startups is simple: when you find a problem that won’t leave you alone, build the thing that makes it go away. And I'm also really enjoying the journey.

For more startup news, check out the other articles on the website, and subscribe to the magazine for free. Listen to The Cereal Entrepreneur podcast for more interviews with entrepreneurs and big-hitters in the startup ecosystem.