Sustained human–AI collaboration is not a single exchange — it is a relationship. Over months of working together, a shared framework forms: implicit commitments, habits of framing, the moments when a concept crystallizes and quietly governs every decision that follows.
That framework evolves. Perspective Engineering is the discipline of seeing that evolution and managing where it goes.
The PSS reads the collaboration corpus — transcripts, decisions, documents — and surfaces what has shifted. Quiet drift away from founding values. Genuine enrichment that has outrun the stated brief. Both matter.
The haul-out moment. Scrape the accumulated drift. Take stock of what has genuinely improved. Update the Foundation Document and the plans that govern the collaboration going forward.
A 2×2 metric separating silent decay from genuine enrichment — across both mission (present purpose) and vision (future end-state). A collaboration you can read, not just run.
Not the agent. The relationship between
the humans and their AI colleagues —
over time, through change,
toward a better version of itself.
Below the waterline: barnacles accumulate unseen. A founding commitment quietly drops out. A value that was central six months ago now goes unmentioned. The work continues to look the same, but the collaboration has drifted from what it was for.
Erosion is invisible from inside the collaboration. That is exactly what makes it dangerous.
Above the waterline: a practice the team invented on the fly turns out to be better than the plan. The collaboration discovered something real — a sharper framing, a more capable approach — that the founding documents have not yet caught up to.
Emergence is the upside that goes uncaptured when there is no instrument to see it.