The Agentic AI Bet: How to Win Without Blowing Up Your Business

Every executive is being told the same thing right now: bet big on agentic AI or get left behind. Gartner says the window to act is three to six months. Accenture is pitching a "10x transformation." But here's what those forecasts leave out — over 40% of agentic AI projects are expected to be canceled by the end of 2027. The companies that will win aren't the ones who move fastest. They're the ones who move smartest.

Writing for ZDNET's special feature on agentic AI, senior contributing editor David Gewirtz breaks down the real failure modes, and the strategies that actually drive returns.

Key Takeaways

• Agent washing is everywhere: Gartner estimates fewer than 13% of self-described "agentic AI" vendors are shipping genuinely autonomous products. Most are rebranding chatbots, RPA tools, and script-based automation. Pilots built on these products are designed to fail.
• Costs compound fast: Unlike generative AI, which consumes tokens in discrete bursts, agentic AI runs continuously — multiple agents at once, voraciously consuming API capacity. Organizations scaling up agentic deployments are finding cloud bills ballooning faster than value accrues.
• Non-determinism breaks standard testing: The same prompt can produce different results across runs. As Salesforce EVP Madhav Thattai noted, "Software used to be solely deterministic. AI agents break that model." Governance and context controls must be designed in from day one, not bolted on post-deployment.
• The right starting point matters more than ambition: High-frequency, predictable workflows with measurable outcomes are where agentic AI earns its keep. Complex, edge-case-heavy processes — those with ambiguity and shifting rules — are where it causes damage.
• Guardrails before scale: Start with limited pilots. Keep humans in the loop on approvals. Increase autonomy only as confidence builds. Mudit Garg of Qventus put it plainly: "Organizations need adaptable governance that evolves as AI advances."

The path isn't a binary choice between 10% savings and 10x transformation. It's targeted wins that pay for themselves, scaled carefully over time.

Read the full article on ZDNET