The News
HeyDonto, the company behind data harmonization products Axiomera and DFT Labs, is making a pointed argument about the structural economics of enterprise AI: that agentic reasoning models cost at least five times more in provider inference than basic chatbot interactions, and that this cost gap widens as task complexity increases. CEO Rivers Morrell contends that the industry’s parallel focus on cheaper tokens and more sophisticated agents creates a self-defeating paradox. His proposed remedy is real-time data harmonization across silos and clouds, applied before AI or analytics ever touch the data, which he describes as establishing the “underlying law of intelligence” rather than continuing to scale by brute force.
Analyst Take
The Cost Paradox Is Real, but the Framing Deserves Scrutiny
Morrell’s core observation is worth taking seriously. Agentic AI architectures do consume dramatically more inference compute per task than simple prompt-response interactions. Multi-step reasoning, tool calls, memory retrieval, and chain-of-thought loops multiply token consumption in ways that flat per-token pricing tables obscure entirely. Enterprise buyers signing AI contracts based on chatbot-era cost assumptions are going to get a very unpleasant surprise when they graduate to production agentic workflows.
Where the argument gets more speculative is in the implied solution. Morrell frames semantic interoperability as the missing “law of aerodynamics” for AI, suggesting that once data is truly harmonized, the cost equation reverses. That’s an elegant metaphor. It’s also doing a lot of work. The inference cost of agentic systems is driven by model architecture, task decomposition, and the number of reasoning steps required, not primarily by data fragmentation. Clean, interoperable data can reduce the number of corrective reasoning loops an agent needs to run, which does trim costs at the margin. But it’s not a fivefold cost reversal on its own.
What the Semantic Layer Actually Solves
The more defensible version of HeyDonto’s argument is narrower and, frankly, more interesting. When an agent operates against fragmented, poorly labeled, or semantically inconsistent data sources, it compensates through additional reasoning cycles: re-querying, reconciling conflicts, and validating outputs against multiple schemas. Every one of those extra steps is a billable inference event. A well-constructed semantic layer doesn’t eliminate the cost of agency; it reduces the error rate and the number of corrective passes, which can meaningfully shrink per-task inference costs in data-intensive enterprise workflows.
That’s a real business case, and it’s most acute in environments where AI is being asked to orchestrate across heterogeneous data estates. This is not a niche scenario. Kubernetes environments, which increasingly host both the AI inference layer and the data pipelines that feed it, sit at exactly this intersection of complexity. According to ECI Research’s 2024 Nutanix Kubernetes Operations Benchmark Study, 47.1% of respondents described their AI training data governance as “semi-manual (managed independently by each project).” That fragmentation is precisely the kind of environment where an agent will burn extra inference cycles reconciling inconsistent metadata before it can do anything useful. HeyDonto is right that this is a cost driver. Whether their specific products are the solution is a separate question.
What ITDMs and Developers Should Actually Do
For IT decision-makers, the immediate takeaway is not to buy a semantic middleware platform on the strength of a metaphor. It’s to audit where your agentic AI pilots are consuming inference budget and why. If your agents are spending significant cycles on data reconciliation rather than on the reasoning tasks they were deployed to perform, that’s a signal that data architecture debt is becoming an AI cost problem. That’s a business case for investment in data governance infrastructure, whether from HeyDonto or from any number of competing approaches.
For developers building on Kubernetes-hosted AI stacks, the operational picture compounds the economics. ECI Research’s Nutanix Kubernetes Operations Benchmark Study found that 41.8% of respondents identified “complexity of orchestrating data pipelines with container infrastructure” as the primary obstacle preventing their organization from scaling AI infrastructure on Kubernetes. That stat maps directly to Morrell’s argument: the harder it is to orchestrate clean data into an AI pipeline, the more the agent compensates through inference, and the higher the bill. Addressing pipeline orchestration complexity is, in practice, an AI cost optimization strategy, not just a platform engineering concern.
Looking Ahead
The broader market dynamic HeyDonto is pointing at will intensify over the next 12 to 24 months. As enterprises move from chatbot pilots to multi-agent production deployments, inference cost management will become a board-level concern, not an engineering footnote. Vendors that can credibly demonstrate reduced per-task inference costs through better data architecture will find a receptive audience among CFOs who approved AI budgets based on a very different cost model. HeyDonto’s positioning is early but the timing is not wrong.
The competitive question is whether semantic interoperability becomes a standalone product category or gets absorbed into the data fabric, data mesh, and AI observability platforms already vying for enterprise budget. HeyDonto’s differentiation will depend on whether Axiomera and DFT Labs can demonstrate measurable inference cost reductions in production environments, not just in architecture diagrams. The aerodynamics analogy is memorable. Production proof points are what will move the market.
Stay Ahead of Application Development Trends
Get weekly analyst insights, research notes, event coverage, and AppDevANGLE updates delivered directly to your inbox.
Subscribe for Weekly Insights
Join technology leaders, practitioners, and GTM teams following the trends shaping modern software delivery.
Looking for deeper research access?
Explore ECI Research reports, survey insights, and market analysis through the ECI Research Portal.
