From caution to confidence: Tackling AI obstacles with education (www.techradar.com)

Generative AI, agentic systems and automation are moving from experimental perks to strategic necessities, but many enterprise projects stall: Gartner predicts 40% of agentic AI projects will be abandoned by 2027 and UK leaders cite governance, privacy/security and cost as top blockers. The article argues these obstacles aren’t insurmountable — rather, skipped foundational steps and poor integration often doom pilots. The author urges business leaders to adopt disciplined, education-driven approaches so AI projects progress beyond proof-of-concept and deliver sustained ROI, noting that compute costs are expected to fall dramatically (Sam Altman predicts ~10x yearly reductions), but timing alone isn’t a substitute for strategy. Technically, the remedy centers on lifecycle governance and orchestration: treat agents as system users with defined identities, scoped access, owners, least-privilege policies and non-repudiation; version, review and retire models like software; and constrain behavior via structured prompts, contextual grounding and restricted output scopes. Security best practices include zero-trust architectures, granular access controls, continuous auditing and observable action logs, plus strict data separation and hourly-rotated encryption keys for customer data. Orchestration—connecting AI tools to existing business logic, data and people—ensures experiments scale into dependable, auditable automation that aligns with compliance, reduces risk and maximizes long-term value.