Honeywell CTO: AI hits 85%, plants need 99.9999%

Both executives spoke at Fortune's AIQ Summit at the New York Stock Exchange, and both framed their limits as practical rather than philosophical. Honeywell's Suresh Venkatarayalu pointed to accuracy: customers demand 99.9999% while frontier models could be at 85%. Ecolab's AJ Wijesinghe pointed to cost: putting the best model on high-volume work sent tokenomics off the roof, and it was sometimes more expensive than having humans.

Ecolab responded by optimizing its models and cutting token costs by about 70% to 80%, using both frontier and open-source models including Anthropic's Claude and OpenAI's. Honeywell is fine-tuning open-source models partly because customers want sovereignty over their data and models, hand-picking them and working with Nvidia's Nemotron team, since an unsupported open-source model without guardrails would be dangerous. Venkatarayalu also noted over-the-air upgrades are coming to commercial buildings, but with an operator in the loop, because a building cannot be shut down for one and a half hours the way a Tesla can.

On where value shows up, Wijesinghe said individual AI adds little measurable value at the enterprise level, vertical AI within a function adds some, and horizontal AI across functions such as sales, finance and supply chain adds the most. Ecolab expects $325 million in annual run-rate savings by 2027, with significant amounts already in hand. The test, in his framing, is whether data foundation, process readiness and cost discipline stay balanced, since if one is heavier than the other, the value does not materialize. Honeywell's customers who got 7% energy savings from existing controls are now asking whether AI can deliver 30% or 40% more.