What AI Still Can’t Replace

At first glance, artificial intelligence and startups are a match made in heaven. A powerful tool for a new venture to accelerate processes, achieve greater operational efficiency, and uncover market insights? It sounds too good to be true.

And it is. AI will hit the startup landscape in ways we can’t yet fathom, particularly impacting software-only startups.  

AI’s effects are already making themselves felt in other fields. Recently, I spoke with a colleague who proclaimed consulting a “dying industry”, with traditional consulting firms like McKinsey confronting serious problems as AI begins to offer rapid analytical services. Nor are tech platforms themselves immune: as users increasingly turn to ChatGPT for answers, the future of Google is being called into question.

With the business world undergoing seemingly endless transformation, we’re left with the question: what can’t AI replace? And what does it mean for the future of startups?

Not all startups are created equal

An important point when considering how AI will shape the startup landscape is how startups can differ. On one point, they all share common ground: each business has barriers with which to defend itself. They may have one barrier or many, good or bad, but they’ll have barriers all the same. 

In the case of B2B SaaS, we typically encounter barriers in the form of producing code and distributing a product. It’s pretty rare for these companies to enjoy network effects, i.e., gaining further value the more users they have. Their core offering is software and code.

If one thing is certain, it’s that AI will lower the cost of code for more or less everyone. This is massively impactful for B2B SaaS distribution. Why? Because AI opens up the opportunity for each client to be able to create the solution for themselves, almost for free. The way in which B2B SaaS is reliant on code as one of its defence mechanisms makes it vulnerable to AI because the tech can effectively reproduce that code off its own bat.

Conversely, companies with network effects are not impacted by AI in the same way. You might be able to copy Airbnb at close to zero cost, but that doesn’t mean you can kill it. The key difference here is the fact that Airbnb has strong network effects. As its network of hosts has grown, so has the number of travellers – and vice versa. This is a valuable asset that helps defend the business from being replicated.  

Furthermore, AI poses a risk when it can replace some physical activity of a business or person. For the most part, it’s hard for the tech to do that. So, when you’re considering a line of defence against AI, it helps to have physical operations. For example, Uber has some massive offline operations that support the activities of its couriers: bikes, bags, local marketing, and so on. 

It’s these components – network effects and an offline presence – that make a business resistant to the advances of AI.

Caution: commoditization incoming

Software-only startups are more vulnerable to automation than those with offline features. This is because the advance of LLMs and robotics creates long-term risks for these startups in the form of commoditization. 

AI enables anyone to vibe code customer solutions in a matter of days. You once needed bespoke software to create products like chatbots; these can increasingly be built using LLM API’s. Highly specialised engineers were once seen as key to building software; now that intelligence is outsourced to the providers of LLMs. 

In the past, no business could afford to even build a basic website featuring a payment solution, so Shopify was created. It’s the same for all CRMs, such as Salesforce, as software was created to fill a gap in the market.

But now you can build your own Salesforce for your company with Loveable. Set aside a week and you’ll be good to go.

AI is democratizing access to the specialized knowledge, technology, and skills these startups use to develop their software; the tech is rendering their unique understanding a commodity and leaving such startups with a diminished ability to dominate a market. 

The whole concept of what B2B SaaS sells is changing. They’ve relied on selling code, but this product will soon have little value as anyone can build it themselves. With the evolution of the tech, software-only startups may find themselves in a precarious position, as their once-vaunted skills, experience, and knowledge become generalised and widely available.

This leaves us with a very promising outcome. Previously, code acted as a constraint for businesses; they were limited in how far they could write code and build products. Now, however, there’s an abundance of code, thanks to AI. We can leverage this to our advantage and get back to basics: what are the customer pain points and how do we solve them? And how can we build barriers for businesses – so they’re not taken out by AI?

Would you trust a robot with your keys?

This is where startups grounded in something tangible are able to distinguish themselves because they are not purely reliant on software and code. 

Dwelly is one such type of startup: at its core is the management of physical properties, linking clients to real world houses and flats. It is strongly grounded in the real world, taking care of keys, doors, and leaky sinks in the process. 

The offline world can be messy and problematic in a way that the online one isn’t: keys get lost, doors broken, sinks leak. But this complexity is exactly what protects this group of startups.

AI will happily write you some code and a robot can give you a lift in a car. But neither of them can hand out keys or fix a leaky sink. My take is that, until robots can reliably manage such intricate operations, these startups are provided with their very own barrier, a defensible layer that can’t be automated away.

Defending against disruption

That’s not to say that we can rule out disruption in this space, so we shouldn’t get lulled into a false sense of security. Take the drive-hailing industry, for example. It would once have been inconceivable for robots to drive cars. Now look at San Francisco: robotisation is already killing Uber and Lyft, as Waymo’s driverless fleet saps them of users. 

If technological advancement teaches us anything, it’s that we can’t predict how it will reshape industries.

Our industry has a defensible layer from automation and LLMs, for now, but there’s a case to be made that disruption will arrive sooner or later. To successfully mitigate that and meet the needs of an evolving business environment, we need to build barriers and secure our product. The model needs to be fundamentally defensible no matter how much AI-generated free code is out there: basic network effects and some offline presence will make sure you are one of the winners.

This is something I’m thinking about a lot. It’s no mean feat, but it comes with the territory in the startup world – remaining reactive and flexible to change. So, how do you build barriers before disruption arrives? That deserves to be the subject of another article entirely.