For a startup founder, AI raises a product question: what can we build that customers will pay for? For an enterprise technology leader, it raises an operational question: how can we improve performance across existing systems? For a business running older software, the question is more immediate: can we adopt AI without rebuilding everything?
These questions meet at the same point. AI creates business value when it connects to real work, reliable data, and measurable outcomes.
As AI capabilities advance in 2026, US businesses have opportunities to build smarter products, improve everyday operations, and modernize existing applications. Making those opportunities useful requires thoughtful implementation.
What the latest evidence says about AI in the US
The scale of investment is substantial. Stanford’s 2026 AI Index reports that US private AI investment reached $285.9 billion in 2025. That investment signals intense competition around AI development and commercialization. Source: Stanford HAI
Business adoption, however, remains uneven. For the survey period ending May 3, 2026, the US Census Bureau reported AI use among 19.8% of businesses overall, compared with 39.7% in the Information sector, which includes many technology businesses. These figures measure reported use in any business function during the preceding two weeks; they do not mean every company has deployed advanced automation. Source: US Census Bureau
For business leaders, this suggests an opportunity to compete through effective implementation. Buying access to AI is one step. Connecting it to customer needs and daily operations is where the work begins.
1. AI agents are expanding what business software can do
An AI assistant can help someone find information or draft a response. An AI agent can also use connected tools to carry out steps in a workflow within defined permissions.
Consider a customer support process. A carefully designed system could categorize an incoming request, retrieve relevant account information, suggest a response, and route the case to the right employee. More consequential actions, such as approving a refund, can remain subject to human approval.
The technology is improving, but reliability still varies by task. Stanford’s 2026 findings describe advances in agent benchmarks alongside continuing weaknesses in planning across multiple steps. Benchmark performance should therefore inform evaluation, rather than serve as a guarantee of business readiness. Source: Stanford HAI
For US software companies, the practical implication is to design around complete workflows: what the system can access, which actions it can take, and how an employee intervenes when something goes wrong.
2. AI is changing software development and the skills teams need
AI coding tools can help developers draft code, explain unfamiliar components, and prepare tests. Their business impact depends on the engineering practices surrounding them.
The 2025 DORA research found a positive relationship between AI adoption and software delivery throughput, while emphasizing that AI amplifies an organization’s existing strengths and weaknesses. Faster code generation alone does not establish that a product is reliable or useful. Source: Google Cloud’s DORA report announcement
For technology teams, this places greater importance on clear requirements, architecture, code review, testing, and understanding the customer’s problem.
A startup can use AI assistance to explore ideas sooner. An enterprise can apply it to maintenance and modernization. Both still need engineers who can judge whether the resulting software behaves correctly in production.
3. Company knowledge can make AI more useful
A general-purpose model does not automatically know a company’s current policies, product documentation, or internal procedures.
One implementation approach is retrieval-augmented generation, or RAG. The application retrieves relevant material from an approved information source and supplies it to the model when answering a question.
A useful starting point might be an internal assistant that helps support employees find answers across product manuals and troubleshooting guides. It should identify its sources, respect existing access permissions, and make uncertainty visible.
The business opportunity is faster access to usable information. The implementation challenge is keeping that information accurate, current, and available only to authorized users.
4. Older software presents a practical AI opportunity
Many established businesses depend on applications built years ago. Those systems may contain valuable transaction histories, business rules, and operational knowledge, even when their interfaces and integration options are limited.
Adopting AI does not always require replacing the entire application.
Depending on the system’s condition, a phased approach could involve:
-
Connecting approved data through secure interfaces.
-
Adding document search or summarization.
-
Automating a repetitive task around the existing application.
-
Updating components that prevent reliable integration.
-
Replacing unsupported or fragile modules in stages.
For example, a distributor using an older order management system could introduce a tool that prepares an order-status summary for staff. A later phase might allow approved updates after integration testing and operational review.
This is an illustrative use case, not a reported customer result. Its value would need to be measured against the distributor’s actual workflow.
AI cannot repair unreliable data or unsupported infrastructure simply by sitting on top of it. An assessment should establish what can be retained, what needs improvement, and what should be replaced.
What should different businesses prioritize?
The right starting point depends on the organization’s stage and constraints.
|
Business type |
Practical starting point |
Useful success measure |
|
Emerging startup |
One AI feature that solves a specific customer problem |
Adoption, retention, and cost per completed task |
|
Growing business |
A repetitive workflow such as document intake or support triage |
Processing time, rework, and employee effort |
|
Enterprise |
A controlled pilot connected to approved systems and data |
Task accuracy, exception rates, and operating cost |
|
Business using legacy software |
An integration assessment and one contained enhancement |
Reliability, time saved, and continuity of operations |
These measures help teams decide whether to expand a pilot, revise it, or stop it before spending more.
How Mobiloitte USA can help
Mobiloitte USA’s AI application modernization services cover existing and legacy applications, with capabilities spanning AI, cloud, APIs, modern interfaces, RAG, and agents. Its published delivery approach includes discovery, integration planning, implementation, testing, rollout, and ongoing optimization.
For startups, the engagement can begin by defining a focused product use case. For growing businesses, it can begin with a workflow that creates avoidable manual effort. For enterprises, the starting point can be an assessment of system dependencies, data access, and operational requirements.
For organizations running older applications, an initial modernization assessment can establish where AI fits and which underlying improvements must come first.
The first phase should have a clear scope and measurable checkpoints, giving the business evidence to guide its next investment.
Start with one business problem
A useful AI initiative begins with a specific outcome: shortening document processing, improving access to internal knowledge, helping employees resolve requests, or making an existing application easier to use.
Choose one workflow. Establish its current performance. Test an improvement with real users. Expand when the results justify it.
