Top 8 pieces of advice for building a Generative AI company today

The world has gone crazy for everything AI and large language models. And for good reason. 

We have entered a new era where language can be understood by machines, where chat interfaces have created totally new UX paradigms and API access has made this technology cheap and ubiquitous.

But what’s the advice for founders building in this new world of LLMs - a world of constantly shifting norms, where the core tech is commoditised and where incumbents have a huge natural advantage?

At firstminute, we last week gathered an expert group of people including Mehdi Ghissassi, Director & Head of Product at Google DeepMind; Guy Podjarny, the founder of Snyk, the unicorn cybersecurity company and Viviana Faga, General Partner at Felicis.

Also there were founders of some of our portfolio companies building in Generative AI including Mistral (foundational model), Nabla (medical co-pilot), Klu (tooling, Resemble (synthetic voice), Granola (workplace of the future) and many more still in stealth. 

Company building

1) The first four pieces of advice were all about company building. And the first was one that is all too easy to forget: focus on the user not the technology. 

“There is going to be massive value creation in this space,” said Mehdi from Google DeepMind.

“But the best advice for founders is to focus on the customers and not fall in love with the technology... The goal should be to solve a problem using AI, not use AI to take on a space for the sake of it.”

Guy from Snyk added: “People need to be saying they are solving X problem using AI, not using AI to solve X… It’s a subtle difference, but a crucial one. The user has to come first.”

2) The second bit of advice was to work hard to build depth in the product beyond just the magic of the LLM itself - and that even clever workflows and fine tuning of the models is not enough. 

“Any problem worth solving will likely require more than just an LLM… and building more will definitely be needed for defensibility over time,” said Guy. 

3) The third piece of advice from the group, which came from founders such as Alexandre Lebrun, the cofounder of Nabla and Mehdi from Deepmind, was that focused apps were more likely to win more than broad infrastructure.

That’s because infrastructure tends to be a “winner takes all” market and so companies building there are 100x more likely to fail. A much easier path is to choose a very specific niche to build in first, even if the long term vision is to dominate everything. 

4) The fourth piece of advice was just that founders should expect FIERCE competition. The great thing about LLMs is that they disrupt everything, but the bad thing is that everyone knows it and there is no way you are the only one with the idea. Fast execution is key!

Product

The last pieces of advice were focused on product choice rather than company choices. 

So the fifth piece of advice was that products that were focused on assisting workers were much lower friction and more likely to gain quick adoption than products trying to build some kind of fully autonomous action. Right now at least, co-pilots are better than agents. 

Having said that, autonomy is the “true magic”, said Guy and where LLM technology can be really transformational. So a sixth piece of advice is that this should be the ultimate goal for many!

The seventh point was that one of LLM's true superpower is processing unstructured content (for structured content, other options may be better/cheaper).

And finally, the eighth point is that security & safety in LLMs is low, and will be for a while. That means that founders have to find tasks where the results are verifiable by people (humans in the loop) or where there are low consequences if there are mistakes!

What larger funds are looking for

Viviana Faga, General Partner at Felicis, a $3bn AUM fund that investors across Seed and Series A, also came to talk to the group about what investors are seeing right now.

First off, she made clear that Tier one investors are taking Generative AI and AI more seriously than ever. 

“We’re in the midst of the 4th tectonic technological shift,” she said. “The 1980s was about hardware, the 90s software and then the 2000s and 2010s it was all about cloud and mobile. We are now in the age of AI”.

She added that in terms of what they are looking for in Series A deals, they are definitely looking for strong traction, although it’s true that very strong founders are getting deals done with only relatively early traction.

Viviana also floated three theses for how they saw this space evolving and what they were excited about. 

1. Interested in the revolution in media and marketing. “We feel very strongly that a significant share of media, marketing, and entertainment will be AI generated.”

2. Personal AI agents will come. “AI agents are going to supercharge personal and professional productivity… Remember Steve Jobs talked about people having 1000 songs in your pocket with the iPod. We think people are at some point going to have a personal AI agent in their pocket.”

3. A race between open and closed source. “I think that the foundation model race between OpenAI and Google will shift to a race between closed source and open first models.”

firstminute view

At firstminute, we are investing a lot in Generative AI companies and we agree with Viviana Faga that the long term impact of the technology will be enormous.

But of course, elements of it today are over-hyped. In the short-term there are major barriers to adoption that need to be overcome getting this technology into production (e.g.....latency, cost, reliability and data privacy to name a few.) And for seed investors, we do think many will be burned by the space and invest poorly. We have to be aware that scale ups and cloud leaders in the public markets have their A* teams on Gen AI + they have massive distribution. This is not the same as cloud-first companies taking on legacy incumbents back in the early days of the cloud revolution.

Given these barriers for adoption, and the natural distribution advantages of incumbents, we have been focusing on investing across the stack, specifically in areas where the opportunities for creating value are accelerated by workflow-specific applications and new infrastructure paradigms.

• At application level, we have made investments in verticals with limited cloud champions capable of innovating quickly, such as AI for healthcare practitioners, comic creation, B2B independent resellers and biochemistry research.

• At the tooling layer, we have invested in 3 horizontal companies redesigning the user experience for creating intelligent workflows, end-to-end applications and machine learning pipelines: one focused on knowledge workers and two on software developers.

• And lastly, we have made an investment in an open source foundational model, Mistral, supporting the thesis that there should be a European champion as an alternative to OpenAI.

What are we looking for next?

1. When it comes to B2B software, we are looking at opportunities in vertical SaaS - products focused on the workflows of a specific user within a specific industry. However, we want to back companies here operating in verticals where there are limited cloud incumbents who can easily roll out an AI co-pilot feature to their existing customers. So areas where digital transformation has not penetrated quite as far. Think Construction, Healthcare, Insurance, Credit Management, Legal Services, Industrial Design & Engineering.....

2. We think GenAI could drive a renaissance in consumer software investing which has taken a backseat over the last few years when it comes to VC interest and funding. With new platform shifts, new consumer behaviors emerge - what will GenAI bring here? We have already seen AI Therapists, AI Lawyers, AI Tutors, AI Personal Assistants, AI Personal Trainers. There will be more weird and wonderful things to come for sure.

3. Also importantly, we think that UI/UX can be completely re-imagined with AI - e.g. a previously complex interface for users can become simple and intuitive - meaning users might be quicker to adopt and less likely to churn! This could be a new CRM, new email or new customer service tools.

4. And let’s remember it's not all about Generative AI.... The next wave of AI is also robotics and industrial digitalization — robots, avatars, and digital twins – where AI interacts with the physical world. We are investing here as well.

If you are building a Generative AI company, please reach out to us at sam@firstminute.capital, michael@firstminute.capital and lorcan@firstminute.capital.