There’s a pattern I’m seeing.
As more companies use general purpose AI to rewrite and run workflows, they’re risking becoming the same.
Why? Partly because most frontier models are trained on overlapping content, so they naturally tend toward similar outputs and ways of solving problems.
We see it mostly in content, but in time it’ll show in decision-making, operations and execution.
AI is already influencing how we think, work and behave. The extent of that influence depends on what and how much we’re choosing to offload to it.
This homogeneity will compound when that offload is heavy, chained and has minimal human intervention - which is becoming increasingly likely as models become more agentic. We’ll naturally tend to accelerate automation as models become more robust, more capable and better at orchestration.
Whether we like it or not, AI possesses powerful automation capability and so will be capitalised on by opportunists. Of course there are cons to this, but the tech is here to stay. We need to instead find ways to harness it more effectively so we can sustain healthy competition.
Competition is good and originality is key for its survival, but the drift to sameness is like boiling a frog. You don’t realise it until it’s already happened.
So how do we sustain originality? Where does differentiation in an AI age come from?
Part of it is the data, but more of it is the system wrapped around the model. Who I am at work is different to who I am at home, or different to who I am at my old company. The environment I interact with shapes my behaviour.
AI models aren’t any different, they’re just more literal in how they respond to that influence. The system we shape around them can have immediate and more tangible impact on their behaviour.
For instance, where does your AI live? what software, company policies, workflows, databases, other AI endpoints is it connected to? What system prompts or tools does it work with? What other agents can it speak to?
When these components are configured together, what emerges is embedded operational context. Yet we can take this further and evolve embedded operational context into something more powerful: adaptive operational context.
The catalyst here is what is commonly known as Skills.md files - and it is the fuel for differentiation. While skills have existed for a while, they’re becoming increasingly important in how AI-enabled businesses operate within growing complexity.
They are essentially a way of packaging hard-won business methodology into operational context an AI system can follow, reason through and execute against. How they make operational systems adaptive is through the simplicity of how they are created and updated. Run a successful workflow multiple times? create a skill so the AI can repeat it next time with more efficiency. Workflow requirements change? Redeliver the workflow, adjust for variations and then update the skill again. Its a kind of pseudo-learning for the model or a procedural externalised memory for the system.
What makes skill so powerful is that it’s system cognition built off a set of files.
So what does this tell us?
As general purpose models grow stronger, businesses can start skipping expensive model training or fine-tuning and instead adapt at the much cheaper system context layer.
A lot of modern AI systems are quietly moving in this direction.
Prompt engineering evolved into context engineering.
Context engineering is now evolving into persistent operational memory architectures.
This likely changes how businesses should think about AI adoption entirely.
The advantage likely won’t come from simply deploying the latest model. It may come from systematically engineering workflows, decision heuristics, operational memory, retrieval layers, playbooks, feedback loops and human coordination into adaptive systems that continuously evolve alongside the business.