Waiting on their resolve-by dates.
Predictions
Claims from the essays, pulled into one dated ledger before resolution. Misses stay visible; ambiguous calls do not count toward the score.
0 ambiguous excluded.
Lower is better; starts once a claim resolves.
Open claims
Pre-registered checks
- Open68%
Local or fine-tuned open models will handle most high-volume narrow LLM calls in cost-sensitive production stacks.
From Local models won the long tail
- Stated
- Resolve by
- Resolution criterion
- By the resolve date, at least three credible production writeups show local or fine-tuned open models serving a majority of narrow classification, extraction, formatting, reranking, or summarisation calls after frontier comparison.
- Open72%
Output-token volume will become a named margin-control metric for agent products, separate from input-token volume.
From The output token tax
- Stated
- Resolve by
- Resolution criterion
- At least three agent-product teams or vendors publicly report output-token reduction, output-token budgets, or output-token caps as a distinct cost-control practice.
- Open62%
AI-assisted coding will keep showing a higher serious-defect rate unless teams narrow it to well-tested or low-blast-radius work.
From Vibe-coded code has more bugs. Price it in.
- Stated
- Resolve by
- Resolution criterion
- New controlled studies or large team retrospectives still show elevated serious-defect or security-defect rates for broad AI-assisted coding, while safer results cluster around test-heavy, low-blast-radius, or prototype code.
- Open70%
Production agent teams will treat MCP servers as governed dependencies with pinning, scoped credentials, and health checks.
From Your MCP server is a prod dependency
- Stated
- Resolve by
- Resolution criterion
- Public MCP deployment guides from at least three serious teams or vendors recommend version pinning, scoped credentials, and real-call health checks for MCP servers used in production agent paths.
- Open66%
Postgres-backed queues will remain the better default than Kafka for most internal queues under roughly 1,000 events per second.
From Postgres is a queue. Stop reaching for Kafka.
- Stated
- Resolve by
- Resolution criterion
- New small-team architecture writeups and incident reports still favor Postgres or similar database-backed queues for sub-1,000 event-per-second internal workloads unless replay fan-out or very high throughput is required.
- Open64%
Teams will increasingly treat LLM model changes as product changes, not implementation details.
From The model switch is a product change
- Stated
- Resolve by
- Resolution criterion
- At least three public release processes, vendor guides, or postmortems describe model changes as requiring product review for tone, latency, quality, cost, or trust changes.
- Open58%
State-changing AI or ops CLIs will be expected to emit receipts that make later audit possible.
- Stated
- Resolve by
- Resolution criterion
- At least three developer tools or internal-platform writeups add explicit command receipts, run ledgers, or audit logs for CLI actions that change state.
- Open55%
Teams shipping AI features will move model cards or model-behavior notes into the repository alongside code.
From The model card belongs in the repo
- Stated
- Resolve by
- Resolution criterion
- At least three public engineering examples keep model behavior, limits, eval notes, or prompt contracts versioned in the same repo as the feature using them.
- Open57%
AI-product teams will treat cache behavior as a product contract because cache misses change latency, cost, and quality.
- Stated
- Resolve by
- Resolution criterion
- At least three public AI-product or infrastructure writeups define cache hit behavior, invalidation, or fallback behavior as user-facing contract surface rather than a pure optimization.
- Open60%
Early-stage teams will keep underinvesting in observability until incidents force the first boring dashboard.
From Observability is a tax you pay before you owe it
- Stated
- Resolve by
- Resolution criterion
- Multiple public startup postmortems or retrospectives still cite missing baseline dashboards, logs, or traces as a contributor to avoidable debugging time or incident duration.
Graded claims
No grades yet
The first extracted claims resolve in 2027. This section is intentionally empty until a claim reaches its date and gets scored.