Artificial intelligence companies are putting more resources into helping businesses use their technology, a shift that has gained momentum during 2026 as companies struggle to turn promising demonstrations into dependable everyday systems. OpenAI and Anthropic have each launched ventures that place technical specialists alongside customers, making implementation a bigger part of the competition for business clients.
The change does not mean that the race to develop more capable AI models has ended. Instead, leading developers are expanding beyond selling access to their software, recognizing that customers often need help connecting it to existing databases, applications, employees and approval processes.
OpenAI Invests Billions in AI Deployment
In May, OpenAI announced the OpenAI Deployment Company, a majority-owned venture backed by more than $4 billion in initial investment. The company is designed to embed engineers inside organizations to identify valuable uses for AI, build working systems and help them operate reliably.
The announcement also included an agreement to acquire Tomoro, an AI engineering and consulting firm. OpenAI said the transaction would bring approximately 150 experienced engineers and deployment specialists into the venture, giving it an established team rather than requiring it to build one entirely from scratch.
Reuters reported that the venture brings together OpenAI and 19 investment, consulting and technology partners. TPG leads the partnership, while Advent, Bain Capital and Brookfield are among the co-lead founding partners.
Anthropic moved in a similar direction earlier that month. On May 4, it announced plans for an enterprise AI services company with Blackstone, Hellman & Friedman and Goldman Sachs, supported by other investment firms.
Anthropic said the operation would focus on midsize businesses and place applied AI engineers alongside the new firm's teams. Their assignment is to learn how customers work, identify promising uses of Anthropic's Claude models and develop solutions that can be supported over time.
Why Forward-Deployed Engineers Are Gaining Importance
Both initiatives reflect a growing market for forward-deployed engineers, often called FDEs. Rather than building products at a distance and handing them over to customers, these specialists work directly with client teams to adapt technology to the requirements of a particular workplace.
The approach has long been associated with Palantir, whose engineers work closely with customers to configure software around operational problems. What has changed is the prominence of this model among companies best known for developing general-purpose AI systems.
A business may find that an AI assistant answers routine questions convincingly during a trial but encounters problems when connected to actual company records. Customer names might differ across systems, product information may be outdated and permissions may determine which employees are allowed to see sensitive details.
Those complications can turn a seemingly straightforward project into a much larger effort involving data cleanup, software connections and organizational decisions. Engineers may need to test not only whether an answer sounds correct but also whether the system used current information, followed company rules and recorded its actions.
The difference is particularly important for AI agents, systems designed to complete tasks or initiate actions rather than simply respond to questions. An agent that drafts a refund recommendation presents a different risk from one authorized to issue the payment without review.
Companies therefore need to determine which tasks can run automatically, which decisions require approval and how errors will be detected. They also need a clear way to measure whether a deployed system saves time, reduces costs or improves service enough to justify its operating expenses.
Gartner Warns That Many AI Projects May Fail
Research from Gartner illustrates the difficulty. In February 2025, the firm predicted that through 2026 organizations would abandon 60% of AI projects that lacked AI-ready data. That forecast applies specifically to projects without suitable data, not to 60% of all AI projects.
In January 2026, Gartner reported that at least half of generative AI projects had been abandoned after the proof-of-concept stage by the end of 2025. It identified poor data quality, inadequate risk controls, rising costs and unclear business value among the reasons that experiments failed to progress.
The challenge extends beyond data. In June 2025, Gartner predicted that more than 40% of agentic AI projects would be canceled by the end of 2027 because of costs, uncertain business value or inadequate safeguards. These are industry forecasts, not confirmed outcomes for every company adopting AI.
Yet Gartner also expects significant expansion. The research firm forecast that as many as 40% of enterprise applications would include task-specific AI agents by the end of 2026, compared with fewer than 5% in 2025. It separately predicted that AI agents could make at least 15% of everyday work decisions without human involvement by 2028.
Those estimates point to an expanding role for AI in office software, customer service, finance and other operations. They do not establish that every deployment will be useful or that greater automation will always produce better results.
Outside Engineers Bring New Risks
There is a further risk in relying heavily on outside specialists. In September 2026, Gartner predicted that by 2028, 70% of enterprises would abandon agentic AI systems built through vendor forward-deployed engineering because of high costs and difficulties maintaining or improving those systems themselves.
The warning complicates the idea that embedding engineers automatically solves the deployment problem. Gartner advised businesses to establish expectations for ownership, knowledge transfer, governance and an eventual handover before signing contracts, so the customer is not left dependent on the vendor indefinitely.
For smaller companies, outside engineering help could provide access to expertise that would otherwise be difficult to hire. But such arrangements still require staff who understand the organization's records, can assess the accuracy of outputs and can take responsibility when an automated process makes a mistake.
Large enterprises face a related problem at a different scale. Connecting AI to systems used by several departments can expose conflicting records and procedures, making coordination among technology, legal, security and operational teams essential.
AI Success Will Depend on Long-Term Reliability
The growing market for deployment services also creates a new way for technology providers to compete. Model capabilities remain important, but clients may increasingly judge vendors by how quickly a system reaches production, how reliably it performs and whether employees can maintain it after the specialists leave.
OpenAI and Anthropic have made major commitments to this approach, but their announcements do not prove that every customer will see a financial return. The more meaningful test will come after the initial rollout, when systems must keep working with changing data, evolving rules and real employees who depend on them.
For businesses weighing AI investments in late 2026, the practical questions extend well beyond choosing a model. They include who controls the data, who approves consequential actions, who maintains the system and whether its measured benefits justify the cost.