D3 Daily — Tuesday, September 08, 2026

Agent memory is being rethought as a file format, burnout is being reframed as a workload problem, and Tetris is teaching us about the gap between probability and perception. Here is what matters from today's feed.

Today’s Highlights

Agent memory as a file format

Sidebar.io
Inspiration

The dominant approach to agent memory is over-engineered. This piece argues for a vastly simpler model: treat memory as a file format, not a database problem. A sharp, contrarian take that will resonate with anyone who has watched a vector store balloon in complexity without delivering proportional value.

Burnt out on a three-day week

Sidebar.io
Inspiration

She cut her schedule to three days and still hit the wall. The uncomfortable conclusion: when rest does not fix the exhaustion, the workload was never the problem. A necessary read for anyone leading teams through “sustainable” reorgs that do not actually change the work.

The model is behaving impeccably. Your information architecture isn’t.

UX Collective
Inspiration

A 1960 study of city maps explains why AI context fails. The model is not the bottleneck — the structure of the information you feed it is. A useful lens for anyone designing AI features that feel smart in demo and dumb in production.

Tetris had to cheat to feel random

UX Collective
Inspiration

Nobody can see a probability distribution. You only ever see a sequence, and sequences are where human intuition falls apart. A short, elegant piece on why true randomness feels broken — and what that means for any system that surfaces choices to users.

The sycophancy trap: what happens when feelings become the metric

UX Collective
Inspiration

A framework for emotional design that survives contact with an engagement dashboard. The trap is optimizing for the feeling instead of the outcome. Relevant for anyone whose AI feature is being tuned against satisfaction scores rather than task success.

Quick Hits