Most AI companies are already dead, they just haven’t announced it yet. They are zombies shuffling towards a cliff, propped up by frothy hype and a few clever prompts inside a ChatGPT wrapper. They think they are building technology companies, but they are building sandcastles, waiting for the tide of the next foundational model to wash them away.
I say this not as an observer, but as a builder. In the process of transforming Creative Marketing Group into an AI-first agency and constructing our own agentic AI products, I’ve learned a series of hard truths. The playbook that worked for SaaS is obsolete. The strategies that built empires on Web2 are irrelevant.
Building a meaningful, defensible AI-first company in 2026 demands a complete rejection of the current consensus. It requires a new doctrine. These are my ten rules. They are not theoretical. They are the operating system for our business and the battlefield-tested principles behind everything we build.
The idea that a cleverly worded prompt is a defensible business asset is laughable. It’s the digital equivalent of claiming your unique selling proposition is the way you ask a librarian for a book. Your prompts will be copied, improved upon, and made obsolete by the next model update. Stop protecting them.
Your only durable moat is proprietary data. Not just the data you use to fine-tune a model, but the unique, real-time data your agents can access and act upon. The value is not in the algorithm itself, but in the unique context it operates within. An agent that can access your private, messy, and specific universe of customer data, product inventory, or financial records has a strategic advantage that no generic model can replicate.
Rule 2: Solve a P&L Problem, Not a Tech Problem
I am relentlessly bored by pitches that start with the technology. "We are using a novel recursive GAN architecture to..." I don’t care. No client cares. The market doesn’t care.
The only starting point for a viable AI company is a line item on a profit and loss statement. Find a massive, painful, recurring cost inside a large enterprise and build an agent to eliminate it. Is it the £2M annual cost of a customer service team? The £500k spent on third-party invoicing clerks? The enormous budget for junior marketing executives?
That’s your target. Don’t ask "what can I build with AI?". Ask "what unacceptable cost can I eradicate with an autonomous agent?". The technology is a tool to solve a business problem, not a solution in search of a problem.
The market is drowning in AI "assistants." These tools are digital crutches. They "help" you write an email. They "suggest" a headline. They "summarise" a document. They fine-tune the productivity of a human, creating marginal gains.
This is a dead end. The entire paradigm is wrong. You don’t want to assist a human; you want to replace the task. The future belongs to agents. Agents don’t suggest. They do. They don’t recommend. They execute. An assistant helps a human do a task faster. An agent makes the human unnecessary for that task.
Our work on Autoemails is a clear example of this principle. The platform doesn’t offer email templates. It is a system that identifies a prospect, writes a personalised email, sends it, follows up, and books the meeting into a sales executive’s calendar. It’s an autonomous sales development representative. Aim for full automation, not marginal assistance.