The Great Unbundling: Why Your SaaS Is a Ticking Time Bomb
The age of Software-as-a-Service is over. It just doesn't know it yet.
The model that defined the last two decades of tech — paying a monthly fee for access to a dashboard you have to log into and operate yourself — is fundamentally broken. It mistakes access for outcomes. It sells you a shovel and wishes you good luck with the mountain you need to move.
Agent-as-a-Service (AaaS) is the successor. It’s not a tool; it’s a digital employee. An autonomous agent that doesn’t need you to log in. It takes the objective, uses the tools itself, and reports back on the outcome. This is the only model that matters for the next decade. If your business is built on users clicking buttons in a dashboard, you are a dinosaur. The meteor is coming.
This is my most controversial take, but the one I stand by most firmly: deep experience in the marketing industry can be a liability when building an AI-first company.
Marketing is filled with dogma, received wisdom, and frameworks that ossify into unbreakable rules. AI requires first-principles thinking. It demands that you break a problem down to its absolute fundamentals, ignoring the way it has always been done. My best hires are not seasoned marketing directors; they are physicists, mathematicians, and systems engineers.
Our head of agentic development, for instance, holds a PhD in computational physics. He has never read Kotler, Ogilvy or the latest HubSpot blog post. He sees marketing not as a creative art but as a complex system of incentives, probabilities, and outcomes. He isn’t trying to optimise the old ways; he is trying to discover new ones. We have a team of five in our core agent group, and not a single one has a marketing degree. This is not an accident.
Founders are obsessed with their tech stack. Their algorithm. Their unique code. None of it matters in the long run.
LLMs are becoming a commoditised utility, like electricity. You can plug into OpenAI, Anthropic, or an open-source model with a few lines of code. Your unique algorithm will be replicated or surpassed within months. The only durable, defensible moat in the age of AI is proprietary data. A unique, high-quality, and constantly refreshing dataset that your competitors cannot buy or replicate.
Consider the tragic missed opportunity for UK retail. Imagine if Tesco had started building an AI model in 2015, feeding it two decades of Clubcard data. It could have built an agent that could predict, with terrifying accuracy, the weekly shop of every single customer. It could have automated its entire supply chain, personalised every offer, and become an unassailable fortress. Instead, it sends me generic coupons for cat food. I don’t own a cat. The data is the moat; they just forgot to build the castle.
When you approach a new problem, do not ask, “What tool could we build for this?” Ask, “What autonomous agent could solve this?”
This is a fundamental shift in product development. It forces you away from building dashboards and towards building outcomes. A tool helps a human do a job. An agent is the one doing the job.