At Transcend Houston 2026, leaders from refining, chemicals, cement and other process industries spent a day comparing how they run AI in their plants. They came with different operating models, different levels of digital maturity and different approaches to adoption, and the friction points they described were close to universal:
- Operators working with systems that do not show why they act.
- Engineers working across fragmented workflows and disconnected tools.
- Teams trying to sustain optimization gains across shifts and organizational boundaries.
- Experienced people retiring while the operation grows more complex.
Attendees treated these as structural problems that are appearing across the process industries at the same time. In the roundtable sessions, people kept recognizing their own plants in stories from other sectors. As one attendee said, "I'm in a completely different sector, and we're dealing with the exact same issue."
Some of the most useful contributions came from operators and plant leaders already running these systems. They described what works, where friction remains and how their organizations have changed, and they learned from each other as much as from the presenters. At the end of an eleven-hour day, the rooms were still full.
From proving value to running it
The industry is largely past the question of whether AI can create value in a plant. The harder question is how to run it across people, workflows, systems and decisions, in an environment where trust, continuity and coordination matter as much as the optimization itself.
The coordination problem
Planning, economics, operations, engineering and plant leadership usually agree on the outcomes they want. In practice they work through different systems and workflows, on different KPIs and at different rhythms, and they often hold different readings of what the plant is doing. One team optimizes for throughput, another for reliability, another for energy and another for economics.
None of these goals conflict in principle. Without coordination across systems and disciplines, though, a local optimum in one place creates friction somewhere else, and the plant behaves inconsistently even when everyone does their job well. Leaders feel it as a gap between the plan and the plant.
More dashboards, copilots and standalone models add information, but on their own they do not close that gap. That takes keeping teams and systems grounded in the same picture of the plant and pointed at the same objectives, with people still making the calls that are theirs to make.
What the Hackathon showed
In the Hackathon sessions, participants worked with live operating information directly, instead of assembling context by hand from hundreds of files, dashboards and reports. That changed the questions they asked. They stopped asking where data lived or which system owned a workflow, and asked operating questions:
- What is happening right now?
- Why is it happening?
- What changed?
- Which operating context matters?
- What actions are available?
Most plants have more data than they can use, and too little connected context. Critical knowledge still sits with individual shifts, in spreadsheets, historians, reports and dashboards, and in people's experience, and much of it is hard to preserve. Engineers still fight many of the problems they fought ten years ago, because the lessons from each episode were never captured and connected to the next decision. That knowledge leaks out of the plant, and the ability to capture and use it in real time is now an operating requirement.
How trust gets built
Operators do not need every mathematical detail behind an optimization system. They do need to see why it recommends or makes a move, which operating conditions drove it, and how it connects to the physical plant in front of them. That is why explainability matters so much in industrial settings. Trust comes from consistency and transparency, reinforced over time by a system that behaves reliably under real operating conditions.
Attendees described adoption moving from experiments toward integration with daily operations. What matters now is whether the systems a plant runs work from the same picture of the plant, and whether operators and engineers can see their own role in the workflow.
Where Imubit fits
Imubit works on this problem in refining, chemicals and polymers, cement, and mining and minerals. The approach is Map, Model, Control: map the plan-to-plant gaps to the recurring decisions behind them, model the causality with learning process models, and take the critical decisions to autonomous execution, inside limits set by plant engineers, on top of the existing APC and DCS, with the operator on the loop. Imubit has over 100 closed-loop controllers deployed worldwide.
