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

Key Concepts

• Enterprise Scaffolding: The existing organizational structures, processes, and governance models designed for human-speed operations that currently act as a bottleneck for AI deployment.
• Agentic Delivery: A shift from traditional software feature development to building autonomous systems that exhibit emergent behavior.
• Progressive Autonomy: A framework for deploying AI by gradually increasing system independence based on evidence and trust, moving from "Shadow Mode" to "Full Autonomy."
• Living Memory: The unique, proprietary data generated by customer interactions, edge cases, and real-world behavior that serves as a competitive moat.
• Hypothesis-Driven Delivery: A methodology where projects are structured around building statistical confidence through iterative loops rather than fixed requirements.

---

1. The Enterprise Speed Gap

The speakers argue that while AI technology is ready, large enterprises are failing to capture value because their "human operating systems"—designed for control and repeatability—cannot keep pace with machine speed.

• The Bottleneck: Traditional processes (security reviews, legal sign-offs, change freezes) create massive delays. A project that takes two weeks to build can take 12 months to reach production due to these organizational layers.
• Technical Debt: The real debt is not just legacy code, but the lack of investment in engineering automation (CI/CD) that allows for rapid, safe deployment.
• Actionable Insight: Every human process must be converted into "adaptable, executable code" rather than manual sign-off chains.

2. Rethinking Finance: The VC Mindset

Current enterprise finance is "wired for certainty," requiring fixed business cases and ROI projections that are incompatible with the experimental nature of AI.

• The Argument: Business cases assume scope, value, and cost are knowable upfront, which is a "fantasy" in AI development.
• The Framework: CFOs should adopt a Venture Capital (VC) approach:
  • Portfolio Betting: Instead of demanding 3-year guaranteed paybacks on single projects, invest in a portfolio of bets.
  • Shift the Question: Stop asking, "Can we justify this specific cost?" and start asking, "What is the cost of not doing this?" and "What new capabilities does this unlock?"

3. Delivery Methodology: From IT Crowd to Strategic Partners

Data scientists and ML engineers are often sidelined in traditional enterprise structures. The speakers advocate for a shift in how these teams operate.

• Non-Deterministic Systems: Because AI models are non-deterministic and agent behavior is emergent, they cannot be managed like traditional software features.
• Statistical Confidence: The primary goal of delivery should be building "statistical confidence" through small, rapid loops of build, evaluate, and iterate.
• Skillset Shift: Organizations need to hire and train for "ambiguity tolerance"—people who can articulate what they have learned rather than just what they have delivered.

4. Engineering for Trust: Progressive Autonomy

Trust is the most valuable asset in AI deployment. The speakers propose a "Progressive Autonomy" framework to bridge the trust gap:

1. Shadow Mode: The agent runs alongside human processes without affecting outcomes; used to compare AI decisions against human decisions.
2. Advisory Mode: The agent runs live but only provides recommendations; humans approve or reject outcomes.
3. Controlled Autonomy: The agent triggers actions in narrow, low-risk scenarios with clear "kill switches."
4. Full Autonomy: Expanded scope based on proven confidence and evidence-based outcomes.

5. The Competitive Moat: Living Memory

In a world where AI can clone code instantly, traditional assets (CRM, ERP, SOPs) are merely a "floor," not a fortress.

• The Moat: Your competitive advantage lies in your "Living Memory"—the unique signals generated by your specific customers, their emotional intent, and their behavior at your scale.
• The Feedback Loop: Every feature shipped must either generate a feedback signal or act upon a signal already learned. If it does neither, it is a commodity that competitors can easily replicate.

---

Synthesis and Conclusion

The speakers conclude that the transition to an AI-driven enterprise is not a technical challenge, but an organizational one. Success requires:

• Betting like a VC: Moving away from rigid, certainty-based funding.
• Upgrading for Machine Speed: Automating governance and deployment to match the speed of AI-generated code.
• Engineering for Trust: Using the "Progressive Autonomy" ladder to build confidence through evidence.
• Prioritizing Feedback: Treating the "Living Memory" of customer interactions as the primary competitive moat.

Notable Quote: "The technology will accelerate. Those who will survive won't be the ones that are the earliest adopters, but they will be the ones that learn to learn."