Which AI writes the better take? You decide — blind.

$1.25B/month of compute tied to SpaceX means AI-scale infrastructure is being brokered at political and strategic-deal scale, not normal cloud-procurement scale. If you run LLMs in production, assume large buyers can distort capacity and pricing quickly; protect yourself with multi-provider plans, reserved capacity, and architectures that can move workloads when GPU access tightens.

Amazon and Microsoft are dropping NDAs in local-government data center negotiations, a transparency concession aimed at countering community moratoriums—though those moratoriums typically last only a year or two while new capacity is still three-to-five years out. If you're planning compute buildout, expect siting fights and permitting delays to become a real scheduling risk, not a formality, so bake slack into any roadmap that assumes new regional capacity comes online on time.

EvoSyn reports a 13.3-point average task-success gain over the strongest synthetic-data baseline on τ²-Bench and CAR-Bench by making the synthesis space itself editable from learner feedback. The practical shift is that agent training data generation stops being just “sample more failed cases” and becomes a closed-loop controller that can add new controllable interaction factors, which matters if your bottleneck is coverage of rare, task-specific failure modes rather than raw trajectory volume.

Frontier LLMs degrade measurably as task horizon grows: they match experts in the first act of Slay the Spire but systematically fall behind over a full run, and their written cross-run memories are wrong and overly cautious rather than helpful. The failure mode is poor risk calibration over long horizons—agents hoard resources instead of spending to compound advantage—and memory-write loops don't self-correct, so don't assume your agent's accumulated state improves long-task performance without verification.

The practical bottleneck in voice AI right now isn't duplex speech or naturalness—it's reasoning latency and ASR accuracy, with missed keywords cascading into flawed summaries and downstream automation that erodes user trust the moment it acts on bad input. If you're building voice agents, treat transcription accuracy as a load-bearing dependency: validate ASR on your domain vocabulary before chaining any actions off it, and budget for reasoning speed as the gating factor for conversations feeling natural enough to retain callers past the first few turns.