Domino Data Lab’s New CEO Targets AI’s High-Consequence Last Mile

Thomas Robinson is betting that enterprise value will shift from general-purpose models to the systems, governance and expertise required to put AI inside consequential business decisions.

Enterprise AI has reached an awkward stage in its development. The technology is moving faster, but the economic returns are not yet self-funding for most organizations. Models are becoming more capable. Token costs continue to decline. Enterprises can generate more content, code and analysis than ever. But those outputs do not, in and of themselves, translate into business value.

This was the premise put forth in aa recent theCUBE interview with newly appointed Domino Data Lab Chief Executive Officer Thomas Robinson. Robinson cited company research indicating that 57% of organizations still struggle to generate returns that outpace their AI spending. His explanation was that Most enterprise AI activity has focused on general employee productivity rather than the core processes where companies actually make money, manage risk and create differentiated products.

We believe Robinson is describing a fundamental shift in the enterprise AI market. The first phase was about access to models and experimentation. The next phase focused on connecting those models to proprietary data, business rules, operating processes and accountable decision-makers to refine workflows and generate new levels of productivity. But many firms struggle to show adequate returns for their efforts.

This is the opportunity Domino is pursuing. The company is positioning itself around a narrower and potentially more defensible market. Specifically, trusted AI systems for a small number of expert users making high-consequence decisions. This is what Robinson calls “the last mile.”

Think drug research, financial risk, underwriting, intelligence and national security. In these environments, a model response is not that valuable. It is only one input into a larger system of evidence and human judgment. Trust in high impact decisions that drive clear business value is the objective.

The research perspective at a glance

The following table summarizes the key takeaways from our discussion.

The CEO change is part of the strategy

Robinson is a new CEO, but he is not new to Domino. He has spent roughly a decade helping build the company’s partnerships, go-to-market motion and solutions business. Under the new structure, Nick Elprin will concentrate on product and engineering while Robinson leads the company and expands its ability to deliver business outcomes.

Robinson described the organizational change as an effort to bridge the gap between technology and enterprise delivery. Domino has already moved toward a vertical strategy in regulated markets such as pharmaceuticals, financial services, national security and intelligence. The next step, according to Robinson, is to place that expertise more directly inside customer transformations.

Our take is that the leadership change reflects where Domino believes it needs more emphasis – i.e. go to market. The company believes it has a strong technical platform. It must now help customers absorb the platform into their operations. That requires process redesign, integration, risk management, implementation talent and proof of measurable results.

In that sense, Domino is separating two distinct responsibilities:

• Product and engineering, where the objective is to extend the platform for increasingly complex but governed AI systems; and
• CEO and go to market execution, where product/platform translates into repeatable customer business within targeted industries.

The logic is reasonable in our view. The challenge Robison is taking on is that the new structure will require clear decision authority. Product direction and customer delivery cannot become separate centers of expertise. The customer experience must feed directly into the product roadmap, while product discipline must prevent the company from becoming a collection of bespoke projects.

Domino has selected a specific approach: High consequence, few experts

One of Robinson’s main messages was between broad employee AI deployment and high-consequence enterprise AI. Domino is going after the latter.

He characterized general-purpose assistants as a form of end-user computing. These tools can help large numbers of employees draft documents, summarize information or perform routine analysis. They have value, but their use alone does not necessarily change the economics of the business…or its core workflows.

Domino is targeting a different model. Its approach is to deploy relatively few expert users making decisions that can have substantial financial, regulatory, scientific or national-security consequences. Robinson cited pharmaceutical research, mortgage underwriting and even the setting of interest rates as examples.

This strategy underscores the trend that customers want to pay for outcomes, not seats. Specifically, most enterprise software is priced by seats or consumption. Domino’s target market may have fewer users, but each workflow can represent more economic value that it can charge for. The relevant measures become:

• The value of the outcome;
• The speed with which the organization can act;
• The degree to which evidence can be reproduced and audited;
• The amount of business risk the system can safely manage and mitigate.

This creates the potential for large and durable customer relationships. It also may create longer sales cycles, higher proof requirements more implementation intensity, but ultimately more stickiness.

The key point is Domino is choosing depth over breadth. We believe that is the correct strategic direction for a company of its maturity and scale. Competing directly with hyperscalers, frontier-model providers and large data platforms for generic enterprise AI workloads would be difficult. Specializing around demanding, regulated and high-value processes may be a smaller TAM, but it makes Domino much more relevant in our view.

The last mile is not just technical deployment

The phrase “last mile” is often used to describe the technical act of moving a system into production. Robinson’s definition is broader. Specifically, in Domino’s view, the last mile includes the full path from a business decision to a production system that can influence that decision safely. According to Domino, iIt requires organizations to:

• Start with a high impact process – i.e. what business outcome will have high leverage?
• Compose the system – i.e. which resources, models, data, processes, etc. are needed?
• Define the interface – i.e. how will experts receive input/evidence and act on it?
• Manage the lifecycle – i.e. what testing and policies must be adhered to?
• Govern at runtime – i.e. how will the organization monitor the systems in real-time and make corrections if necessary?
• Have clear accountability – i.e. who is responsible for what result?

Robinson described an advanced AI system as more than an LLM. It may combine generative AI with predictive machine learning, statistical models, rules engines, computer vision or simulation. Domino’s role is to integrate those components, govern their behavior and expose the result through an application, dashboard or other interface used by the business.

This struck us as a meaningful decision point by the company. There are many industry examples of so-called last mile strategies proving out. Infor has been able to compete with Oracle and SAP with a “micro verticals” strategy, generating billions in revenue. Procore is a $1.4B construction industry project management system that competes effectively with horizontal products like Asana. Toast is a multi-billion dollar company that provides last mile operational capabilities to restaurants.

Applying this concept to high-consequence applications is a reasonable strategy. An LLM should not be the sole decision mechanism. A model may interpret unstructured information or propose an action. A predictive model may estimate probability. A rules engine may enforce policy. A simulation may test potential outcomes. Ultimately, a human expert could approve, reject or modify the recommendation.

We believe Domino is trying to establish a position above the data, infrastructure and model layers but below the final business application. A useful description is a sort of decision control plane – meaning a layer that assembles, governs and connects diverse AI components to high impact workflows.

That is an attractive position and one that will be contested by larger horizontal players. Hyperscalers, data platforms, model providers, enterprise application vendors and process software experts all want to own more of this layer. Domino’s focus on openness is therefore necessary. The company must demonstrate that it can deliver better governance, faster implementation and stronger domain outcomes than customers can assemble from broader platforms or that other general purpose platforms can deliver as a feature.

As models become more ubiquitous, value moves into the surrounding system

Robinson expects models to become increasingly commoditized. He pointed to the growth of forward-deployed engineering (FDE) teams at major AI companies as evidence that model providers also recognize the enterprise implementation gap. They are adding people because a more capable model does not automatically integrate itself into a customer’s operating process.

We agree with the direction of the premise, although “commoditization” requires some inspection.

Frontier-model capability will continue to advance and be fundamental. Different models will retain advantages in reasoning, coding, economics, etc. But enterprises will increasingly treat model selection as a portfolio decision rather than a permanent platform commitment.

As model abundance increases, enterprise differentiation shifts toward:

• Proprietary data and context;
• Workflow integration;
• Evals and testing;
• Governance and auditability;
• Human interaction design;
• Tacit organizational knowledge;
• The ability to route to or change models without rebuilding the entire system.

This shift supports Domino’s strategy. It also raises the standard bar the company must meet.

Model independence cannot be demonstrated only through a list of connectors. It must mean operational sovereignty. A customer should be able to replace a model while preserving policies, evaluations, telemetry, business logic and, importantly, the user experience. In our view, the time and risk involved in changing models will be a better test of openness than the number of models a platform claims to support.

Manual delivery can be a feature, not a failure

One of the more interesting data points discussed in the interview was that 40% of organizations still deliver AI output through scheduled reports or direct requests to data scientists.

The conventional interpretation is that this is due to immature tech. Robinson offered a different take. Specifically, sometimes a report that informs a responsible human is exactly what the business requires. The ability to automate a process does not mean that full automation is desirable.

This leads to an important idea. The objective of last-mile AI should not be maximum autonomy. It should be consequence-adjusted autonomy. The graphic below provides additional color to this concept:

A scheduled report may therefore be an effective last-mile deliverable. The mistake would be to measure the percentage of a process that has been automated– that is not a business outcome. The correct measures are things like decision quality, cycle time, economic outcome and risk.

This is especially relevant as enterprises move toward agents. The industry often treats human involvement as temporary friction that better models will remove. We believe that view is too simplistic, at least as of today. In high-consequence systems, the human is both a fallback mechanism and part of the control architecture.

Trust must be designed into the operating system

Robinson outlined a three-layer model for trusted AI:

1. A governance policy engine that defines requirements, approvals and controls throughout the development lifecycle;
2. Monitoring and tracing that determine whether the system continues to operate within its intended parameters;
3. A responsible human who remains accountable for the business process and its consequences.

As Robinson put it, Domino still believes that “judgment is better than hallucination.”

The significance is that governance cannot be a separate compliance exercise performed after the system has been built. It must become part of the path to production and part of runtime operations.

This is even more important for agentic systems. The discussion cited Domino research indicating that 41% of organizations are piloting or scaling agentic AI without the governance needed to manage it. Robinson warned that the market remains in a “Wild West” phase and has not yet experienced the watershed failure that forces organizations to become more cautious.

Organizations should not wait for that event.

A multi-agent system in which one agent judges another may improve reliability, but it does not create accountability. Enterprises still require traceability, escalation, override mechanisms and a named business owner. The systemic risk is not only that an AI system makes a bad decision. It is that employees gradually stop challenging its recommendations because the output appears credible.

For Domino, trust could become an important differentiator, but only if the company can turn the concept into measurable operating evidence. Customers should expect to see reproducible evaluations, policy enforcement, exception rates, human overrides, rollback procedures and audit-ready records.

Forward-deployed engineering is central to the business model

Robinson also addressed the growing use of forward-deployed engineers (FDEs). He rejected the idea that companies can simply rename traditional consultants and claim to have an AI-native delivery organization.

Domino’s talent strategy combines younger AI-native recruits with its existing platform and enterprise expertise. The company intends to teach these employees how to use Domino’s technology and how to operate inside complex enterprise transformations.

The strategy has some merit. AI-native employees often have greater fluency with new tools, agents and rapid experimentation. But high-consequence environments also require deep domain knowledge, regulatory awareness and an understanding of how enterprises actually change.

The differentiator will come from pairing those capabilities, not choosing one over the other.

Robinson stated that Domino does not perform services-only engagements. The platform is part of every implementation. He also argued that coding time is collapsing because of agents and coding assistants. As a result, value is moving toward specification, testing, validation, governance and rollout. He described the emerging process as a “solution development lifecycle” rather than simply a software development lifecycle.

This was a key strategic point in the interview.

As code generation becomes abundant, the scarce capability is no longer typing code. It is defining the correct problem with enough precision that a machine-generated solution can be tested and trusted. It is also proving that the system behaves as intended when it encounters real enterprise data, policies and users.

Domino’s opportunity is to convert implementation knowledge into reusable software assets, in particular:

• Industry-specific workflow patterns;
• Evaluation suites;
• Policy templates;
• Model and data connectors;
• Monitoring configurations;
• Human-approval designs;
• Repeatable deployment methods.

Every customer engagement should strengthen the platform. If that knowledge remains primarily inside individual employees, Domino risks becoming a high-end services company with software attached. If it is codified and reused, forward-deployed engineering can become a product-development and distribution engine.

A platform that requires experts is not inherently a problem. A platform whose value remains trapped inside those experts is.

Pricing will reveal where value lives

Robinson expressed support for value-based pricing because it aligns Domino’s incentives with customer outcomes. He also acknowledged that many customers prefer predictable enterprise pricing and do not want to share a large portion of the economic upside with a vendor.

Domino has historically avoided pure consumption pricing and has priced around the value of people using the software. But Robinson noted that this model must evolve as coding compresses, technical roles change and business users perform more technical tasks.

This is a broader industry problem.

Consumption pricing can reward vendors when customers spend more, even when the spending does not produce a better outcome. Seat pricing becomes less logical when agents execute work on behalf of people. Pure outcome pricing can be difficult because results may depend on many factors outside the vendor’s control.

Domino’s eventual pricing model will provide an important signal. If the company truly owns part of the decision and outcome layer, its unit of value may need to move toward governed AI systems, production workflows, assurance levels or agreed business-value bands.

The objective should be to preserve customer predictability while reflecting the value and risk of the process. There will be no perfect metric, but the pricing structure must not encourage token consumption, user proliferation or unnecessary automation.

Analyst Angle: Domino has a credible premise, but it must prove leverage

We believe Domino has identified a real and viable market gap. Enterprise AI has no shortage of models. The difficulty is combining models with business context, controls, existing systems and accountable human judgment.

Domino’s history across earlier generations of enterprise data science may be an advantage. The generative AI market is rediscovering disciplines that regulated enterprises already understand: statistical evaluation, model governance, production monitoring, reproducibility and controlled rollout. The new requirement is to extend those disciplines to LLMs, agents and mixed-model systems.

The company’s focus on high-consequence, expert-driven workflows is also more differentiated than a generic enterprise AI platform story. It provides a coherent answer to three questions:

• Where can AI create enough value to justify a high-touch implementation?
• Where is governance important enough to influence platform selection?
• Where does Domino’s accumulated enterprise and domain experience matter?

However, the position is not simply protected by the phrase “last mile.” Every major AI and data platform vendor sees the same movement of value toward workflows, governance and implementation.

Domino’s moat must come from the accumulated knowledge it can codify in software – i.e. evaluation methods, domain-specific controls, deployment patterns, model portability and evidence that the platform improves consequential decisions that drive ROI.

What to watch over the next 12 to 24 months

Robinson said observers should expect new product capabilities for building advanced AI systems that combine multiple types of models. He also emphasized that customer solutions and outcomes will be the primary measure of progress. Talent remains his top organizational priority.

The following scorecard will help determine whether Domino is executing against the thesis.

Action item

CIOs, chief AI officers and line-of-business leaders should not treat AI as only a horizontal adoption program. Select one consequential workflow where the business metric, decision owner and cost of error are clear. Establish the baseline outcome before selecting technology. Then define the appropriate level of autonomy with requisite lifecycle governance and runtime observability. Put in place a human override and test whether models can be changed without rebuilding the entire solution. Vendors, including Domino, should be measured on improvements to the business process—not tokens consumed, copilot adoption, models placed in production or demonstrations completed.

Disclosure: This analysis is based on the August 27, 2026 theCUBE interview with Thomas Robinson. Domino Data Lab is a paid sponsor of theCUBE. Neither Domino Data Lab nor other sponsors have editorial control over this post, or any content on theCUBE, theCUBE Research or SiliconANGLE.