Impossible to prompt: How founders should rethink their companies with AI to stay competitive

For the first half of 2025, 64% of all VC money in the US was invested in AI startups. That amount gives every existing founder a stop. What do I have to do to become a relevant business that is not wiped out in today’s world? The good news is that many companies can utilise AI to their advantage; the bad news is that many products aren’t suited and will likely vanish. To prevent being a victim of a prompt, there are three things you need to get right: your strategy, your product, and your organisational efficiency.

Transform your strategy

To help you shape your product strategy, first, we need to define what AI is today in an oversimplified way. AI is yet another way to input data to a computer, so that it gives you an output. We first started giving instructions to computers with punch cards in the 60s, then character-based UI in the 80s, graphical interfaces in the 90s, and now we are using natural language and voice to tell a computer what we need.

What has also improved over the years is the quality of the output from simple algorithms doing math to sophisticated LLMs that have reasoning and problem-solving.

But at its core, AI is yet another way for us to input instructions to a computer, so by definition, the quality of the output will depend on the quality of your input — just Google "prompt engineering" is proof.

Consider the application I can generate using Vibe coding with 18 years of engineering & product background, versus what a gardener can do with the same tool when it comes to user experience, security, scalability, quality, extensibility, interoperability, and so on.

Sure, you usually don't compete with gardeners eating your lunch as a founder, but you get my point — people who understand how to prompt AI and know how to build products and learn how to talk to customers will be able to ship products faster and create internal systems instead of paying for archaic SaaS products.
 
Knowing that the majority of your codebase will become a commodity with people reproducing it with prompts, you need to be very honest about what your real competitive advantage is that will still hold true in 10 years. In other words, what are the unique inputs that make your business model work that are hard to replicate or impossible to prompt that will stay timeless?

Is there anything unique and challenging to build in your business that requires more than AI to succeed, such as regulatory licenses, distribution channels, contracts, relationships, skilled labour, or even a company culture that can adapt to market disruptions?

Knowing and defining this will help you build more defensibility and be less attached to your existing codebase, features, or approach, so that you can fully transform your business. If you don't have a great answer to the above, your product is likely just a system of record that lacks any competitive advantages and will likely be wiped out by prompts with Lovable (👋 Hello to most of the HR systems out there).

Transforming your product

"Coding is fast. Deciding what to code is slow," - Andrew Ng.

Knowing the answers to the above will make it easier to start building the new vision for your product. Since AI is just a new input to your system, you can treat it as a channel to your application next to web and mobile, rather than anything else.

Think about which experiences work better through voice or text instead of clicks and forms. Don't just focus on how you'd build your product today with agentic flows—look at how your existing assets let you create something stronger than someone starting from scratch. And don't fall into the trap of adding just any AI agent. For SaaS companies, having one is already becoming table stakes, much like having a mobile app was 12–15 years ago.

Let's take the following example: at Payhawk, we are building a new travel agent. We are a spend management company that issues cards, but travel has never been a focus for us. As a result, many people rightly ask the question: why yet another travel management system, and why build it with a conversational UI rather than a web interface? Because we have the ingredients to build something truly unique, combining what we have and what didn't exist as an experience before.

Yes, you can build an agent with n8n that searches Expedia or other travel providers and shows you hotel and flight options for a few nights. But building a payments infrastructure takes time — especially the kind that lets the agent pay with your card, or move money efficiently (and secure better prices than Expedia).

Or enterprise SaaS, so the agent can give you suggestions for flights and hotels based on your travel policy defined by your finance teams or review whether the trip request has been approved or give you recommendations based on your preferences for hotels, airlines, and airports and even remember that you like to seat on a plane (the red-eye flight from San Francisco in an aisle seat; but a window seat when you fly in the morning..."); or check you in for the flight and obtain your boarding passes or call you a taxi/Uber/Bolt when you land; or create an expense report for you at the time of the booking so it can then add your trip expenses directly to it.

And doing all of the above with highly profitable unit economics. Without relying on too many counter-parties when building it that eat up your margins, and can disrupt your business if they go down (and things go wrong in payments more often than you think, I have more horror stories to tell than I wish for).

You shouldn't just build AI features for the sake of catch-up and to slap AI on your homepage, so you look relevant or in panic claim you have been "using" AI even before George Orwell wrote 1984. You should solve specific and concrete problems that your customers want. And the demand today is massive.

According to the latest MIT report on AI Adoption, more than 90% of Enterprises are looking to adopt AI, but only 5% of AI trials end up being successful. The main reason is that most AI tools don't learn and don't integrate well into existing enterprise workflows. Why? Because they were built in isolation, creating the boring parts like security, multi-tenancy, policies, audit trails, and workflows doesn't earn you brownie points with VCs, but they are surely needed to pass any serious IT security team.

Once you understand your strategy and how your product fits into the AI world, you can focus on transforming the way you build your product. But this is not about which code generation tool you use, since code generation is already adopted by 84% of developers; instead, it's more about who leads your product.

The role of the classical Product Manager planning for 12 months ahead in a traditional company is gone. The Product Manager is being gradually replaced by the role of a Product Leader, who combines customer knowledge, user experience, product design, and architecture into a single role. Instead of working with a UX designer to determine what needs to happen and then convincing engineers that this is the best direction, they can rapidly develop code prototypes and small-scale features that can accelerate the learning curve for the whole organisation.

And if your PM is not good at vibe coding, then they're not good at building products in the first place. Why would you trust people who cannot tell AI what they want, but you trust them to coordinate a dozen designers and engineers on what to build? Vibe coding skills will be a required task for every PM role at Payhawk from today.

Transforming your organisation

Your entire product is the sum of all products, services, and experiences your users need, and many of those depend on the underlying teams and processes. You can't transform your product unless you transform how your company operates, which is the hardest thing of all.
One of the biggest competitive advantages of the native AI companies is their lean performance. The expectation is that it's now possible to have 20-30 people who build >$500M business within 17 months.

The good news is that you can use off-the-shelf products to improve individual performance and some operational efficiency with AI.

The bad news is that MIT claims that they only work 5% of the time and can lead to a direct P&L impact, despite $30–40 billion in enterprise investment in GenAI. Why the negative outcome? Because AI faces its biggest obstacle: The human and its incarnation, the manager, who knows it all.

The low-hanging fruit: 5%

Many would mistake adopting individual employee tools, like ChatGPT, Cursor, Co-pilot, or Perplexity, that enhance individual employee performance, with transforming your organisation.
In fact, companies are already reporting that 80% of them have tried them, and 40% of them use them actively, but this individual productivity doesn't directly translate into your numbers. And given that every other company can and is adopting them, they can rarely be a source of competitive advantage.

What's needed is a systematic deployment and change of existing processes and skills to achieve a sustainable competitive advantage that will change the trajectory of your business, which requires strong leadership and coordination from the top.

The next step: 10%

There are also excellent off-the-shelf products that can enhance your company's performance and make a significant difference in a particular department. Intercom’s Fin is a good example that can handle 50-60% of your support with AI, and even reach 85-90% if you give it access to your product and customer data to solve more complex problems.

Fin can identify weaknesses in your help articles and enhance them or write new articles based on its knowledge. It also comes with the right handoffs between an agent and a human, along with strong reporting & analytics to ensure that your team isn’t just a watchdog, but an active participant and co-pilot of the AI.

The promise land: 85%

To get to the next stage of transformation, you have to go deep on automation and replace how you work today. There are many products, such as n8n, Glean, Lovable, and Make, that offer the holy grail: Building your own automation, minimising human error, and speeding up work, so that you can do more with less.

But, fully transforming your company and adopting AI internally to boost productivity for everyone would require a lot more coordination, leadership, and effort beyond these tools.
Not least, you also need to identify what needs changing and why. Adopting AI internally to transform how you work is the hardest part for two main reasons: a) organisational complexity is a function of the number of people you have, and b) unlike your codebase, which can be an asset, the existing processes, manuals, and most notably the skillset of employees are likely a liability.

The bigger the company is, the bigger the knowledge base. The data becomes distributed, processes become discrete, and information travels more slowly. An incumbent company that is not AI-native will require significantly more time and resources to adopt AI and transform its processes, people, and tools into a truly native AI company, achieving exceptional operational efficiency with $1M+ ARR per FTE.

Let's illustrate how hard it is to transform an organisation and rebuild its roles and processes to fully adopt AI. To do this, we can use a formula called the IT Core Complexity, which is a generic model for IT transformation. It uses as input the number of unique connections between employees, and the number of roles in an organisation that need transformation. It then adds 2 or 3 as a stress factor, which represents the number of priorities each role handles at any given moment.

Complexity = (Communication Channels + Roles Complexity) X Stress factor
The number of individual communication channels is also key in Conway's Law of Organisational Complexity, representing the communication structure of the organisation. It dictates that the system design will mirror the communication boundaries and friction of the organisation that created it. For example, a company with 10 employees has 45 unique channels/connections between individuals. A company with 100 people has 4,950 unique connections, while a company with 500 employees has 124,750 individual connections.

Below is a breakdown of the inputs and timeline to transform an organisation based on its IT Core complexity. We can even assume that AI will take half the time, but still, we're looking at an average of 12 months to transform a business with just 100 employees if this is a full-time initiative across the board.

And yes, the math shows that having fewer people can be a competitive advantage in a market that is moving so fast, and can significantly impact the business with AI. But it also doesn’t guarantee that the next-gen native AI companies won’t end up building a complex and heavy internal structure that slows them down.

History tells us that such a wide-scale transformation of an industry typically takes 3-5 years to fully realise the impact of AI on the P&L of most businesses. Until then, it’s your decision how you want to shape your strategy, product, and organisation, and make a statement on whether you want to be in the lead, follow, or do nothing and risk being wiped out as irrelevant.