I promised myself I would document failures as carefully as successes. Tessera has clear limitations, and understanding them is as important as understanding her strengths.
Missing Context
When an email chain omits a constraint I would normally infer implicitly, political context, contractual nuance, or a known personality dynamic, Tessera frames the right decision but misses a social landmine. In those cases, I add a sentence or soften a recommendation. I do not reverse it.
This is the most common failure mode and the most benign. Tessera is working from the signals she has. When signals are missing, the analysis is incomplete but not wrong. The fix is richer input, not better algorithms.
Novel Situations
When a situation has no close precedent in the lattice, Tessera undercommits by design. Her recommendations are safe but lack the sharpness I would bring after deeper reflection. This is the correct failure mode. I would rather have a system that hedges on novelty than one that fabricates confidence.
Emotional and Political Dimensions
Tessera does not model emotion well. She can identify situations where emotional dynamics historically influenced my decisions, but she cannot read the emotional state of current participants. Political maneuvering is similarly opaque. She can flag when a situation resembles past political problems, but she cannot assess current political dynamics in real time.
What Tessera Does Not Save Time On
To be explicit: novel first-principles strategy, high-stakes people decisions, situations where politics dominate facts, and decisions requiring moral judgment rather than technical judgment. In those cases, Tessera supports framing, not resolution. The human remains essential.
These are not temporary limitations awaiting a better model. They are permanent boundaries that reflect the difference between a decision support system and a decision-making system. Tessera is the former by design.