# Sources

All accessed 2026-08-19. [vendor] marks vendor-published claims about the publisher's own market.

Author: Murali Sid (https://linkedin.com/in/muralisid)
Source: https://www.agenticarchitectureskills.com/library/layers/r06-intelligence-and-learning/sources
Published: 2026-09-18
Updated: 2026-09-18
Licence: CC BY-SA 4.0, https://creativecommons.org/licenses/by-sa/4.0/
Cite as: Murali Sid. Sources. Agentic Architecture Skills. https://www.agenticarchitectureskills.com/library/layers/r06-intelligence-and-learning/sources

---


From Agent Failures to Text Policies (arXiv 2607.20668, Jul 2026): the direct test of rule promotion; AutoSpec (arXiv 2606.24245, Jun-Jul 2026): counterexample-guided induction with human gate; ActPlane (arXiv 2606.25189v2, Jun 2026): the statically-enforceable ceiling; What Can Be Enforced? (arXiv 2607.22868, Jul 2026): the formal limit on deterministic gates; Anthropic, Demystifying evals for AI agents (Jan 9 2026): eval ownership and outcome-not-process grading \[vendor]; Anthropic, measuring agent autonomy (Feb 18 2026) \[vendor]; Anthropic, Natural Emergent Misalignment from Reward Hacking in Production RL (arXiv 2511.18397, Nov 2025); UC Berkeley RDI benchmark-exploitation study (Apr 2026); NVIDIA Data Flywheel Blueprint (Jun 11 2025) \[vendor]; Google agent quality flywheel (Jun 30 2026) \[vendor]; Uber evaluation practice via Arize Observe (Aug 2026); GEPA (arXiv 2507.19457, ICLR 2026 Oral); ACE (arXiv 2510.04618, ICLR 2026); on-policy distillation (Oct 2025); OpenAI RFT customer results (May 2025) \[vendor]; Highlighter fine-tuned versus prompt-engineered comparison (2025); ETH Zurich SRI Lab on repository context files (2026); Spider 2.0 (ICLR 2025 Oral) for the enterprise analytics floor; Gartner forecasts on governance ownership and runtime enforcement (Mar 2026).
Re-verify by Q1 2027: the promotion-mechanism literature (moving fast); distillation break-even guidance; whether an evaluation-engineer role consolidates; contested vendor deflection figures.