Notes from the
review queue.
What we’re learning as we build the team-context layer in the open. Engineering depth, product thinking, and the occasional field note from a founding customer.
Five founding customers. One founder. The opening.
I am taking on five founding customers in 2026. I will sit with each of you through the first pass personally. Here is what that means, who it is for, and how to be one of the five.
Lasse Nørby3 MINSempleo speaks MCP both ways. Here’s why that matters.
The Model Context Protocol is the first serious open standard for how LLMs read and write to external systems. Sempleo is an MCP server and an MCP client. That choice is load-bearing.
Lasse Nørby5 MINFive layers are the shape of a team. Here’s why Sempleo modelled them.
Most “AI context” products flatten everything into a single pile. Teams do not work that way. Sempleo’s five-layer model matches how ownership and knowledge actually sit inside an organisation.
Lasse Nørby6 MINCapability is not the bottleneck. Context is.
Every six months a new frontier model. Every six months the median enterprise AI project still fails to land. The bottleneck moved a year ago, and most of us missed it.
Lasse Nørby4 MINHaiku is the reviewer. Here’s why a smaller model does this job.
Every entry in your context gets a quality dot — empty, thin, or sharp — scored by a small model running constantly in the background. Why Haiku, and why not the frontier model, is the entire trick.
Lasse Nørby4 MINPending, not silent. The review queue is the product.
When an agent learns something new about your team, it does not write it to your context. It proposes. A human approves. That single design choice is what makes Sempleo safe to trust.
Lasse Nørby5 MINThe chasm between a working demo and a working agent.
Every enterprise AI project I have watched has the same shape: an electric demo, a polite retro, nothing in production. The chasm between those two states is where the category actually lives.
Lasse Nørby5 MINWhy Sempleo’s context layer runs on Postgres.
A new category does not need new infrastructure. It needs boring, predictable, well-understood infrastructure so the novel thing can fail safely. Postgres is the right answer.
Lasse Nørby5 MINThe week I decided to build Sempleo.
For eighteen months I watched good teams fail at the same thing. In October the shape became impossible to ignore, and I stopped being able to work on anything else.
Lasse Nørby4 MIN