Most Enterprise Agentic Projects Are Doomed, Here's Why — Jess Grogan-Avignon & Jack Wang, Accenture

TL;DR

• 1

  The biggest blocker is enterprise scaffolding, not model quality — Accenture’s teams say AI projects stall on security reviews, data governance, AI gateways, and deployment processes, like a simple two-week build that needed another 12 months to align infrastructure, security, governance, and app teams.

• 2

  Business-case culture kills agentic work before it starts — Traditional enterprise funding assumes scope, value, and delivery timelines are knowable up front, but Grogan-Avignon and Wang argue AI value often emerges through experimentation, citing AI achievers with roughly 50% higher revenue growth and examples from Walmart and JP Morgan.

• 3

  Agentic delivery should look more like data science than classic IT — Because models are non-deterministic and behavior is emergent, teams should run hypothesis-driven loops focused on statistical confidence, not fixed milestones, Jira rituals, or upfront utopian designs.

• 4

  Trust compounds more than features do — They frame deployment as an 'exposure ladder' from shadow mode to advisory mode to controlled autonomy, where each step is earned through evidence and outcomes rather than pass/fail testing or project-plan completion.

• 5

  Your real moat is 'living memory,' not the systems you already own — CRM, ERP, and SOP data are just table stakes; defensibility comes from the feedback signals you collect in real customer interactions—edge cases, corrections, intent, and behavior at your specific scale.

• 6

  The prescription is blunt: bet like a VC, upgrade for machine speed, engineer for trust — Their closing advice is to treat governance speed as the CTO’s core engineering problem, make finance a portfolio partner, and build every feature either to create feedback or to act on what feedback has taught you.