Hardware vendor: tens of logical qubits at 1e-4 error are now demonstrated, not projected.1
It did not move alone. Agentic Memory System, Eval Harnesses and Personal Wiki all took a step this fortnight, and they are at three different stages — which is the more useful thing about them than the fact that they are busy.23
Agentic Memory System is the most settled of them: tested and citable, inside a year. Long-running agents need a memory layer that separates what happened from what was learned; raw vector recall over transcripts is the wrong abstraction and will be replaced by structured episodic stores with summarised recall. The lab's own run: above ~50k tokens of history, retrieval latency — not storage — is the dominant cost of an agent memory layer.45
Eval Harnesses sits between them — tested and citable, inside a year. An eval harness is five instruments, not one — capability, task, release canary, judge calibration, cost and latency — and the one that pays first is the canary; a judge without a human agreement score is not an eval, it is an opinion with a decimal point. The lab's own run: judge reliability is a property of the rubric and references, not the judge model, and uncalibrated judges favour their own family.26
Personal Wiki is the least settled: in somebody's harness, not yet a finding, one to three years out. The durable output of an agent's experience is a human-readable wiki of skills and knowledge that the next agent — and the next person — can read and revise, not a vector store or a weight update; compiling experience into that form is the consolidation step the memory field left hand-rolled. Anthropic: skill storage and retrieval are becoming a first-party platform primitive.37
Something also stopped being true. Vector-store memory as the agent's long-term memory is in the graveyard: the memory bake-off answered the question with a different design: structured episodic store with summarised recall beat raw vector recall on precision and latency above 50k tokens of…8
The next twelve months in Agentic Memory System turn on one thing: A consolidation step that a delivery team can configure without a research engineer — currently the step we hand-roll every time. Nothing in this fortnight moved it.
