wenzel orlandevery foundation model is converging. GPT, claude, gemini, mistral. they're all getting better at the same things. reasoning, coding, analysis. the gap between them is shrinking every quarter.
so if the model isn't the moat, what is?
expertise. the accumulated taste, judgment, and decision-making that lives in people's heads. the thing a senior engineer knows about your codebase that no documentation captures. the way a founder evaluates deals after 15 years. the quality standard a creative director holds that makes work good vs great.
this expertise has always been the most valuable thing in any organization. but it's rarely been capturable. it lived in people, and when people left, it left with them.
that's changing. AI can now begin to learn expertise, store it, and apply it. the key is memory.
most AI systems are stateless. every conversation starts from zero. they have no context about your domain, your preferences, your past decisions. they're smart, but they're strangers.
a system with persistent, structured memory works differently. it holds what happened (episodic). it can surface what it means (narrative). it can guide what to do next (strategic). three layers, each building on the last.
this is why we built doobls. your doobls are agents that hold your expertise as something captured and yours. the episodic layer is live today; the deeper narrative and strategic layers are maturing.
as those layers mature, what you capture stays yours: portable, deletable, and running on eu infrastructure you can leave with.