ع
The June 9 launch moves industrial AI from advisory software to automated decisions inside a running petrochemical complex. The figures are company-reported, the full release has not shipped, and Yokogawa holds a longer autonomous track record.
On June 9, Honeywell introduced Experion Cognition, an AI-enabled control system platform that makes recommendations and executes automated decisions in industrial facilities, demonstrated in a live proof of concept at Borouge Group International’s Ruwais facility in Abu Dhabi. Across multiple pilot tests, the platform anticipated incidents that subsequently triggered alarms, with an average advance window of five to ten minutes. The claim is narrow and commercially material. An anomaly corrected before alarm conditions develop is handled by adjustment. The same anomaly caught at the alarm stage can end in lost production or a unit trip.
The System
Cognition operates as part of the Experion PKS distributed control system network, which allows deployment into existing control rooms rather than their replacement. It runs real-time predictive simulations of plant state and monitors for what Honeywell terms pre-alarm signatures, small cross-variable deviations that indicate a developing anomaly. Reasoning engines compare live data against historical steady-state parameters and digitized standard operating procedures, then present the affected variables as one situational analysis rather than isolated sensor breaches. Reporting from the Honeywell Users Group conference describes analytics, generative human-machine interfaces and small language models running on-premise, with the cognitive layer containerized separately from the deterministic control layer because AI develops faster than control-system lifecycles. Jim Masso, president and chief executive of Honeywell Process Automation, said the platform “brings autonomous control rooms into production”.
The Borouge Program
The demonstration sits inside a collaboration announced in June 2025. Borouge and Honeywell committed to deliver the petrochemical industry’s first AI-driven control room designed for full-scale, real-time operation at Ruwais, which the company states will become the largest single petrochemical site in the world. The project belongs to Borouge’s AI, Digitalisation and Technology program, a portfolio of more than 200 initiatives that generated 573 million dollars in measured value in 2024 against a 575 million dollar target for 2025. Hasan Karam, chief operating officer of Borouge International, describes the work as the petrochemical sector’s first AI autonomous operations. The next phase assesses scaling the technology across Ruwais and Borouge’s global facilities.
The Demand Case
Honeywell’s stated rationale is workforce replacement: as long-serving operators retire, delegating situation management to autonomous agents is intended to let less experienced staff run plants with knowledge comparable to a veteran’s. The commercial demand is visible in Honeywell’s own filings. In the first quarter of 2026, its Process Automation and Technology segment recorded orders up 3 percent with double-digit growth in process technology, and segment margin expansion of 200 basis points to 23.7 percent.
Competing Deployments
The production-autonomy claim enters a field with prior results. Yokogawa’s reinforcement-learning algorithm FKDPP controlled a distillation column at an ENEOS Materials plant for 35 consecutive days, 840 hours, in early 2022, and in 2023 became the first reinforcement-learning AI formally adopted for direct control of a plant. Yokogawa has since deployed multiple coordinated autonomous control AI agents at a major Aramco gas plant. Emerson introduced Project Beyond in May 2025, a software-defined operations platform built from its DeltaV control system and AspenTech software, designed in part to manage AI agents within a hybrid human and AI workforce. The architectures differ. Yokogawa’s adopted system governs a specific process. Emerson’s platform is an infrastructure rebuild. Honeywell’s approach embeds cognition into the control layer already installed across its customer base.
The Verification Gap
One limitation should be stated plainly. Every performance figure above, including the five-to-ten-minute window, is the vendors’ own account of their pilots. No independent benchmark of Experion Cognition exists, and trade coverage from the Honeywell Users Group conference indicated the full release was still slated for later this year. Adoption in safety-critical plants also requires acceptance from operators, insurers and regulators, none of which a proof of concept establishes. The test ahead is defined, however. Borouge’s assessment of scaled deployment across Ruwais will generate operating data at a size no single-site pilot can, and the five-to-ten-minute figure will either hold at that scale or it will not.
