Inc42 AI Summit 2026: The Playbook For Building Agentic AI Systems That Deliver ROI
At Inc42 AI Summit 2026, panelists said agentic AI needs clear boundaries, context, and security to deliver business value. The discussion argued companies should work backwards from workflows, not chase every new model.
Intelligence analysis by GPT-5.4 Mini
A panel at Inc42 AI Summit 2026 argued that agentic AI is still hard to scale, with only a small share of organisations seeing measurable returns. The practical answer, speakers said, is to protect sensitive data, define strict operating limits, and design around customer outcomes rather than hype.
The article says smart AI helpers only work well when they are given the right job, the right rules, and the right information. It is like giving a delivery worker a map and a safe route instead of locking them out of the building.
Analysis
What the panel argued
At the Inc42 AI Summit 2026, a session titled “The Operator Playbook For Building And Scaling Agentic AI Systems That Deliver Real Business ROI” focused on why agentic AI has not yet produced broad business returns. The article cites a Deloitte study saying only 10% of organisations are seeing meaningful, measurable returns from their deployments.
The core problem: trust and context
The panel, moderated by Ideaspring Capital founder and managing partner Naganand Doraswamy, included leaders from Wingify, Razorpay, Skyflow, Shipsy, and Avalara. Their shared view was that AI tools become useful only when they have enough business context to act well. At the same time, handing agents access to internal data makes many companies nervous, especially when sensitive information is involved.
Skyflow’s Amruta Moktali argued that fear of data leakage can stop teams from getting value at all. Her point was that companies often block data so aggressively that the agent cannot do anything useful. The article says her view was that organisations should protect sensitive information inside documents while still allowing systems to use the context they need.
Working backwards from the workflow
Another recurring theme was that companies should not treat AI adoption as a race to use the newest model. Avalara’s Job Sam Koshy said the focus should be on what brings ease, speed, and accuracy for customers, and then work backwards from the model’s strengths.
Shipsy’s Dhruv Agrawal pushed a similar idea: leadership teams should cut through external noise and focus on execution fundamentals. The article frames this as a broader lesson for the agentic era: successful teams will be the ones that build predictable, secure environments with clear operational boundaries, not the ones that simply chase the latest trend.
The bottom line
The panel’s playbook is practical rather than flashy. Build around real workflows, protect sensitive data carefully, and give agents enough room and context to create value without turning them loose on everything.
Key points
- A Deloitte study cited in the article says only 10% of organisations are seeing meaningful, measurable returns from agentic AI deployments.
- Panelists said AI agents need business context to be useful, but access to internal data raises security concerns.
- The article argues companies should protect sensitive information without blocking the context agents need to work effectively.
- Speakers said organisations should work backwards from customer needs such as ease, speed, and accuracy instead of chasing every new model.
- The main message was that real ROI depends on execution, boundaries, and predictable operating environments.
If companies follow the panel’s advice, agentic AI could move from experiments to tools that save time and improve customer service. The article suggests that clear boundaries and better context could help more organisations join the small group already seeing measurable returns.
If companies keep blocking data too broadly, the article warns they may never get useful results from agentic AI. Chasing every new model also risks endless experimentation without clear business value or ROI.
