As part of The State of the AI Frontier 2026, AI Frontier Network invited leaders building and deploying AI in the real world to share where the frontier is actually moving. In this contribution, Monisha Athi Kesavan Premalatha, Lead AI PM at Microsoft, gives a candid read on what changes in 2026 — and what to watch.
On the capability that will reach the mainstream
Autonomous workflow agents that can reliably execute multi‑step operational tasks end‑to‑end. We are moving past copilots/agents that “assist” and into agentic workflows eg: triaging exceptions, coordinating across systems, triggering actions, and closing loops with minimum human intervention. In large-scale supply chain environments, we’re already seeing agents that monitor risk signals, update systems of record, communicate with teams, and escalate only when needed. Over the next year, this shift from “AI that answers” to AI that acts will become mainstream across enterprise operations.
On AI and decision-making
By 2027, decision-making will become continuous, simulation‑driven, and largely AI agent initiated. Instead of leaders reviewing dashboards, AI agents will run real‑time scenario simulation, quantify tradeoffs, and recommend (or execute) the optimal action. Human decision-makers will shift from “pulling data” to auditing and approving agent‑generated decisions. In high‑velocity environments like cloud infrastructure, this will compress cycle times by more than 70% and fundamentally change how organizations plan, prioritize, and respond to risk.
On the most underestimated risk
Operational brittleness caused by silent data & business documentation quality drift. Most discussions focus on model hallucinations or safety, but in operational production environments, the biggest failures come from subtle, compounding data & business inconsistencies, eg: schema changes, supplier feed delays, missing timestamps, or misaligned business logic. These issues quietly degrade agent reliability and can cause cascading operational errors. The industry underestimates how much data & business governance, lineage, and real-time validation will determine whether autonomous systems can be trusted at scale.
On the advice that matters now
Start with one high-friction workflow, automate it end‑to‑end with an agent, and measure real business impact. Avoid broad “AI transformation” programs. Instead, pick a workflow where delays or manual effort materially affect outcomes like order management, customer onboarding, risk triage and build a closed-loop agent that can observe, decide, act, and learn. Success in one workflow creates the internal credibility, data foundations, and architectural patterns needed to scale AI across the organization.
Also I have written an article in Harvard journal - https://hdsr.mitpress.mit.edu/pub/fdzqkh85/release/1 - sharing AI maturity framework and also how AI Agents are transforming decision making with what leaders should know
On the trend worth watching
I’m most excited about agentic systems that blend with operational execution like agents that don’t just detect risk but proactively resolve it. In my work supporting Microsoft’s global AI datacenter supply chain, we’ve seen how combining forecasting, anomaly detection, and autonomous action can eliminate thousands of hours of manual coordination and materially improve infrastructure readiness. The convergence of prediction + action is where AI becomes a true operational teammate for me.
