The series, in order.

Self-improvement stopped being a training-time event and became an operating property — a loop distributed across four substrates: harness, skills, memory, and evaluators.

The agency/automation line is a definable engineering boundary — independence of decision, not tool-calling — and where you draw it fixes architecture, evaluation, and what to fear.

Harness design is model-specific and iterable, so it is becoming something agents do to themselves — weakness mining in, human authorship out.

Skills — versioned, composed, evolved procedural knowledge — are the capability capital that survives model swaps and compounds across tasks.

Agent memory is crossing from retrieval plumbing to trained behavior — metamemory, memory-as-action, learned context representations, and governed persistent state.

Next-state prediction is consolidating perception, prediction, and control — world models are becoming agent infrastructure rather than a research aesthetic.

Learning from self-generated experience has named, measurable failure modes — collapse, addiction, hacking — and the guardrails are becoming standard equipment.

Verification is the binding constraint on agent autonomy — it no longer comes free with generation, and building it is the actual moat.

The measurement layer of the agent economy fails basic measurement science — reliability is not validity, saturation hides regressions, and eval-portfolio design is now a discipline.

Under self-improvement, static evaluators get consumed — evaluation must co-evolve with the agent or the loop optimizes the test.

Inference stopped being a cost center — speculators train on live traffic and search primitives are trained in; the serving stack is inside the learning loop.

Parallel and any-order generation is the first credible bid against left-to-right — simultaneously a training story, a transparency story, and a serving story.