By Tom Chapman
Faisal Hoque on why structural integrity, governance and human oversight – not speed – are the CIO’s real test in the agentic enterprise
Faisal Hoque has spent a career building things, as an engineer, a founder and an advisor to enterprises and defence agencies. When he says the last twelve months changed how he works more than the previous decade did, it is worth listening to. Analysis, design, prototype and production – once sequential – have collapsed into hours.
Then the sentence CIOs should pin above their desks: “You can spawn an agent in ten minutes, but that does not mean you should shoot it into somebody’s network,” said Faisal.
Speed is now the cheap part
Faisal’s phrase for what is missing is “structural integrity”, a civil-engineering term applied to software, data and organisations: how data is defined, how it is governed, whose network it sits on, who may act on it. Agents raise the stakes because they take the human out of the loop. Input, output, a person decides — becomes input, agent, action. That is the whole point and the whole risk. An agent has to stand on solid ground before it is allowed to stand alone.
Most enterprises are not there yet. In a September 2026 Thinker Media survey of 600 enterprise technology leaders, only 24% have governance, security and compliance controls built into AI and data workloads by default, and only 24% apply the same framework wherever a workload runs. Yet 55% already run AI in production for predictive analytics and 53% for cybersecurity and fraud. Production is ahead of the guardrails.
Five principles emerge from the conversation:
1. Never build what you cannot override
Asked what the agentic equivalent of a parachute is, Faisal says most are not technical. The technical layer is necessary: contain an agent, revoke its access, shut down a network. The harder governance is philosophical, board-level and leadership-level: deciding what is enough, what will not be handed over, and keeping the capability to wind things back. Just because we can does not mean we should is, for him, a design requirement, not a slogan.
2. Governance by default, wherever the workload runs
The pattern enterprises keep falling into is the opposite of that: controls bolted on after the fact, and a framework that quietly changes when the workload moves. The principle is that policy travels with the data and the agent, on-premises, in a sovereign region or in a hyperscaler, and an exception is a decision someone signs, not a default.
3. Sovereignty by design
Sovereignty in the engineering sense means knowing where the data lives, whose network it crosses and which rules travel with it. It has climbed the agenda fast, driven less by any single factor than by a stack of reinforcing ones: regulation, data privacy and residency, geopolitical uncertainty, dependence on technology providers in other countries, cybersecurity and business continuity. “Sovereignty has become a property of the platform, not a clause in a contract,” said Kevin Dallas, CEO of EDB.
4. The humans are part of the structure
The software, hardware and data, Faisal says, are just the stuff; structural integrity has to include the people who run it. That is where the most common transformation mistake sits. Companies cut institutional knowledge because they had a capable chatbot, then rehired the same people when the knowledge turned out to be the product. Supplement and amplify, do not substitute. If there is nobody there, who uses it?
Two observations sharpen the point. The most enthusiastic adopters are not the digital-native young; Faisal’s own interns rejected AI avatars and AI-generated answers outright in favor of human interaction. The adopters are middle managers, who know how the work runs and fear for their jobs. Enterprises removing that layer are removing the people best placed to make agents work.
5. Think in decades, not quarters
Faisal assesses a leadership team in ten minutes: do they live on sound bites or ask complex, brain-hurting questions? Quarter-to-quarter thinkers want instant answers. Leaders who last ask what must be true across people, culture and execution for a strategy to hold. Strategy, he says, is only as good as its execution.
What the successful ones are building
Leaders already know the target architecture. Most describe their data, AI and agents as fragmented today, but expect that to consolidate fast, with many aiming to be unified or embedded in daily operations within a few years.
The idea gaining ground fastest is the AI operating system. “What leaders mean by it is structural integrity made concrete,” says Kevin, “deployable and governed wherever the data must reside, including on-premises and air-gapped, built on open standards, with many vendors’ agents sitting under one set of controls.” Some already say their foundation is so embedded that moving would mean re-architecture. which is exactly why the window to choose deliberately is now.
Worse before better
Faisal’s forecast is philosophical: things get worse before they get better, because the people shaping this are few and the awareness is young. His hope is the generation asking who is this for before how fast can we ship. CIOs need not wait. The five principles are the parachutes; build them now, with humans in the loop as conductors, not musicians. Technology should elevate humanity, not eliminate it. The enterprise that acts on that will still be standing when the rest are hiring their people back.
Drawn from Faisal Hoque‘s conversation with Michael Gale on the AI & Data Horizons podcast. Faisal is the author of Transcend and founder of Shadoka.
Original article @ AI Magazine.
