AI Deployment Moves to the Center of Enterprise Spending as Adoption Hurdles Persist

The competitive edge in enterprise AI is shifting toward implementation. OpenAI and Anthropic have both announced businesses built around placing engineers with customers, a sign that the industry’s next commercial test is whether AI can perform reliably inside everyday operations.

OpenAI launched its deployment company in May 2026 with more than $4 billion in backing. Its engineers work inside customer organizations to understand processes and help systems reach routine use. Days earlier, Anthropic announced a similar venture aimed at mid-sized businesses that often lack the technical staff to integrate AI on their own.

These specialists are known as forward-deployed engineers, a model associated with Palantir. Their role goes beyond configuring software: they connect AI tools to a company’s information and workflows, then work with the team to address problems that emerge in practice. Their results are measured by whether the system continues to function amid real transactions, customers and exceptions.

Deployment becomes a commercial priority

For years, AI competition centered on model performance, speed and cost. Those measures still matter, but the spending on customer-facing engineering suggests a broader commercial calculation. A capable model has limited business value if a company cannot connect it to reliable information, fit it into existing processes and keep it useful after a pilot ends.

That changes the economics of selling AI. The traditional software model depends on building a product once and distributing it widely. Embedded engineering requires more direct work with each customer, but it can expose where products deliver measurable operational value—and where they fail. For AI providers, that work can also build a close understanding of customer needs.

The companies’ announcements point to deployment as a growing part of the market’s investment thesis. The question is no longer only how much a model can do in a controlled demonstration. Buyers and sellers also have to establish whether it can handle the uneven data, approvals and exceptions that characterize business operations.

Why pilots struggle to reach routine use

Pilot projects often begin with selected data and cooperative users. Production systems encounter records that are duplicated, outdated or inconsistent across departments. An AI tool may process those records quickly, but speed does not correct a flawed source or resolve disagreement about which version is authoritative.

The source article cites Gartner estimates that sharpen the execution risk. Gartner predicts that through 2026, companies will abandon 60% of AI projects that lack AI-ready data. In the same research, 63% of companies said they either do not manage their data in a way that meets AI’s needs or are unsure whether they do.

Gartner has also warned that more than 40% of agentic AI projects could be canceled by the end of 2027, citing costs, unclear outcomes and weak controls. These figures frame the challenge as an organizational one as much as a technical one. Data quality, ownership of workflows and clarity about what a system may do without human approval all affect whether a project can move from trial to sustained use.

That is where hands-on deployment teams can make a difference. They can identify conflicting records, map how work actually moves between teams and help define what a successful result looks like. They can also clarify which decisions an AI system may handle independently and which require a person’s review. The work is specific to each organization, which makes it harder to scale than selling a standalone model.

Readiness will shape the next phase of adoption

Gartner expects task-specific AI agents to appear in 40% of enterprise applications by the end of 2026, up from less than 5% in 2025. It also forecasts that by 2028, AI will make at least 15% of everyday work decisions autonomously. These are forecasts, rather than guaranteed outcomes, and their realization will depend on companies’ ability to put the underlying systems and controls in place.

For businesses, readiness comes down to practical questions. Is the information accurate and consistent? Can teams describe the outcome they consider acceptable? Are responsibilities clear when an AI system encounters an exception or proposes an action with consequences? Companies that cannot answer those questions may find that adding agents expands operational risk without producing dependable results.

The implications reach beyond technology procurement. Engineers embedded with customers can learn where delays occur, where errors recur and which processes create or protect value. That knowledge can inform product design and future purchasing decisions. It also explains why AI providers are investing in teams that spend time inside customer organizations: the relationship offers a view of demand that model benchmarks alone cannot provide.

Execution could widen the field

For founders and smaller companies, the shift may open paths into specialized markets. A business does not necessarily need to train a frontier model to build a valuable AI company. It may instead focus on adapting existing systems to the data, requirements and workflows of a particular industry or customer group.

That opportunity comes with an operating challenge. Customer-specific implementation can generate insight and support adoption, but it also requires skilled staff and sustained engagement. Companies will need to find ways to turn lessons from individual deployments into repeatable products without assuming that every customer works the same way.

Model releases will continue to attract attention, and better performance can expand what businesses are able to automate. The commercial test, though, is whether systems produce consistent results once they meet ordinary operating conditions. As AI spreads through business software, the companies that understand those conditions—and can make deployment repeatable—look best positioned to turn technical capability into durable use.