Draft:AI automation

AI automation is the use of artificial intelligence (AI) technologies to automate tasks, workflows, and business processes. It combines artificial intelligence with software-based automation so that systems can process information, support decisions, generate outputs, and execute actions with reduced manual intervention.

Traditional automation generally relies on predefined rules or scripts, while AI-enabled automation may incorporate machine learning, natural language processing, generative artificial intelligence, predictive models, and intelligent agents to process more variable forms of information.

AI automation is used in areas including customer service, finance, human resources, software development, document processing, marketing, data analysis, and enterprise operations.

Background

Software automation has been used for decades to perform repetitive processes such as data entry, form processing, system integration, and transaction handling. Technologies such as robotic process automation (RPA) typically execute predefined actions based on explicit rules.

AI automation extends these approaches by incorporating artificial intelligence into automated workflows. This may allow systems to classify information, interpret natural language, recognize patterns, generate content, recommend actions, or coordinate a sequence of tasks.

The distinction between artificial intelligence and conventional software automation can depend on the technologies and level of autonomy involved. The OECD has developed frameworks for classifying workplace AI according to its capabilities and effects on workers, emphasizing that AI systems should be evaluated according to their context of use rather than treated as a single category of technology.

Technologies

AI automation systems may combine multiple technologies.

Machine learning

Machine learning uses statistical models and algorithms to identify patterns in data and generate predictions or classifications. In automated workflows, machine learning may be used for fraud detection, demand forecasting, document classification, anomaly detection, and recommendation systems.

Natural language processing

Natural language processing (NLP) enables software to process and generate human language.

NLP can be used in automated workflows to classify emails, summarize documents, extract information from text, interpret user requests, or generate responses.

Generative artificial intelligence

Generative artificial intelligence systems can produce new content such as text, software code, structured data, images, and summaries.

When generative AI is integrated with workflow software, it can be used to interpret requests, retrieve information, prepare outputs, and interact with external systems.

The National Institute of Standards and Technology (NIST) has identified generative AI as presenting distinctive risks requiring specific governance, evaluation, monitoring, and risk-management practices.

Robotic process automation

Robotic process automation uses software to carry out repetitive interactions with digital systems.

RPA can perform tasks such as entering information into applications, transferring data between systems, generating reports, or processing standardized forms.

AI technologies can be combined with RPA to process less structured or more variable inputs.

Workflow orchestration

Workflow orchestration coordinates actions across applications, databases, APIs, AI models, and human users.

For example, an automated workflow may receive a request, analyze it using an AI model, retrieve relevant information, update another application, and request human approval if predefined conditions are met.

Intelligent agents

An intelligent agent is a software system designed to perceive information, make decisions, and perform actions toward a defined objective.

Agentic systems can be connected to tools, APIs, databases, and enterprise applications. This allows them to perform multi-step workflows rather than only generating individual responses.

Applications

Customer service

AI automation can be used to classify customer requests, search knowledge sources, generate responses, route tickets, summarize interactions, and assist customer-service workers.

Amazon Web Services describes the use of retrieval-augmented generation in customer support, where generative AI retrieves relevant information from organizational knowledge sources before producing a response.

Human resources

AI automation can support administrative processes such as employee onboarding, document processing, training, internal knowledge retrieval, scheduling, and candidate management.

The use of AI in employment and workplace processes has also raised concerns about fairness, transparency, privacy, accountability, and worker autonomy.

Finance

Applications in finance include fraud detection, invoice processing, reconciliation, expense classification, forecasting, transaction monitoring, and anomaly detection.

AI automation can also be used to extract information from financial documents and route unusual transactions or exceptions for human review.

Sales and marketing

AI automation can be used for customer segmentation, lead classification, campaign support, personalized recommendations, content generation, and workflow coordination.

Software development

AI technologies are increasingly incorporated into software-development workflows.

Potential uses include code generation, testing, debugging, documentation, code review, refactoring, incident investigation, and modernization of legacy software.

Document processing

AI automation can combine optical character recognition, machine learning, and natural language processing to extract and classify information from invoices, contracts, forms, emails, and reports.

Knowledge management

AI systems can search organizational documents and internal information repositories.

Retrieval-augmented generation is commonly used to connect generative AI systems to approved external information sources before generating a response.

Benefits

Reduction of repetitive work

AI automation may reduce the amount of repetitive or administrative work required in some processes.

The OECD reported in 2024 that four in five workers surveyed said AI had improved their performance at work, while three in five said it had increased their enjoyment of work.

Operational efficiency

Automated systems may process high volumes of information more quickly than manual workflows.

The potential benefits depend on the quality of the system, underlying data, level of integration, and suitability of the process being automated.

Data processing

AI systems can process structured information as well as less structured inputs such as documents, emails, images, and natural language.

This makes AI automation applicable to processes that may be difficult to handle entirely through fixed rules.

Scalability

Automated systems can allow organizations to process increasing volumes of work without increasing manual effort at the same rate.

However, scaling AI systems may also increase infrastructure, monitoring, security, and governance requirements.

Human-AI collaboration

AI automation does not necessarily eliminate human involvement.

In many systems, humans continue to supervise results, handle exceptions, approve sensitive actions, and make decisions where legal, financial, ethical, or safety considerations are significant.

Challenges and limitations

Accuracy and reliability

AI models can produce inaccurate, incomplete, or unsupported outputs.

Generative AI systems can also produce information that appears plausible but is factually incorrect. This creates a need for validation, testing, monitoring, and human review in applications where mistakes could have significant consequences.

Data quality

The performance of AI systems depends on the information used for training, retrieval, inference, or decision-making.

Inaccurate, incomplete, outdated, or biased data may reduce system reliability.

Security

AI automation systems may have access to enterprise applications, databases, APIs, customer information, and internal documents.

This can increase the importance of access control, authentication, identity management, system isolation, logging, and monitoring.

NIST's Artificial Intelligence Risk Management Framework recommends systematic risk-management practices for organizations developing, deploying, or using AI systems.

Privacy

AI automation can involve the processing of personal, proprietary, confidential, or regulated information.

Privacy risks can depend on the types of information collected, how data is stored and transferred, which models have access to it, and how outputs are used.

Bias and fairness

AI systems may reproduce or amplify biases contained in training data, evaluation data, or operational processes.

This is particularly relevant where AI contributes to decisions involving employment, credit, healthcare, insurance, or access to services.

Transparency

Some AI systems may be difficult for users or organizations to interpret.

The ability to explain how an automated decision was reached can be important for accountability and regulatory compliance.

Human oversight

Human oversight remains important in many AI-assisted processes.

The OECD has highlighted concerns relating to privacy, fairness, worker agency, transparency, robustness, safety, security, and accountability in workplace AI systems.

Governance and risk management

Governance refers to the policies, controls, responsibilities, and procedures used to manage AI systems.

NIST's AI Risk Management Framework organizes AI risk-management activities around four functions: Govern, Map, Measure, and Manage.

The associated NIST AI RMF Playbook provides suggested actions that organizations can use when incorporating trustworthiness considerations into the design, development, deployment, and use of AI systems.

Governance measures may include:

• defining responsibility for AI systems;
• limiting access to data and external systems;
• maintaining audit logs;
• monitoring system behavior;
• testing models before and after deployment;
• implementing human-approval requirements;
• establishing procedures for incidents and failures;
• evaluating whether systems comply with organizational policies.

Testing and evaluation

AI automation systems may require forms of evaluation beyond conventional software testing.

Traditional software testing often focuses on whether software produces deterministic outputs and remains available under expected conditions. AI systems may additionally need to be evaluated for accuracy, robustness, bias, safety, tool selection, data usage, and behavior under unusual conditions.

In August 2026, NIST released the initial public draft of the TEVV-Athlon Framework, a framework for test, evaluation, verification, and validation of AI systems.

The framework is designed to be applicable to statistical machine-learning systems, large language models, multimodal models, and agentic systems.

Workplace impact

The increased use of AI automation has generated debate about productivity, employment, job redesign, worker surveillance, and inequality.

The OECD reported that occupations considered to be at the highest risk of automation account for about 27 percent of employment across OECD countries, while also noting that many workers report productivity benefits from AI.

AI may replace some tasks while augmenting others. The effect on employment can therefore vary according to occupation, industry, technology, and organizational choices.

Adoption

Business adoption of artificial intelligence increased substantially during the first half of the 2020s.

The Stanford Institute for Human-Centered Artificial Intelligence's 2026 AI Index reported continued growth in corporate investment and organizational AI adoption, with generative AI accounting for a substantial share of private AI investment.

The report also describes increasing use of AI across business functions while noting that the extent of deployment varies between organizations and industries.

Agentic AI and multi-step automation

One emerging area of AI automation is the use of AI agents capable of carrying out sequences of actions.

Rather than responding only to individual prompts, an agentic system may interpret an objective, retrieve information, interact with external software, and carry out multiple steps.

Enterprise use of AI agents increased during 2025 and 2026, although the extent of deployment varied widely.

Microsoft's 2025 Work Trend Index reported that 46 percent of surveyed leaders said their organizations were using agents to fully automate some workstreams or business processes.

Microsoft's 2026 Work Trend Index reported continued adoption of agent-based workflows across industries and emphasized the development of repeatable human-agent workflows, handoff procedures, and quality standards.

Emerging trends toward 2027

Trends toward 2027 are expected to reflect the continued integration of artificial intelligence with enterprise software, workflow orchestration, data systems, and automation platforms.

Because projections about future technology are uncertain, developments toward 2027 are generally better described as emerging trends rather than established outcomes.

Greater use of AI agents

AI agents are increasingly being connected to databases, APIs, software tools, and enterprise systems.

This can enable automated systems to perform multi-step tasks rather than individual isolated actions.

The expansion of agentic workflows is supported by recent enterprise adoption data, although deployment remains uneven across industries and organizations.

Human-agent workflows

One emerging model combines AI agents with human supervision.

In such systems, agents can perform routine activities while humans approve sensitive actions, review uncertain outputs, handle exceptions, and remain responsible for high-impact decisions.

Microsoft's 2025 Work Trend Index emphasized the importance of determining the appropriate division of work between humans and AI agents rather than assuming that all processes should be fully autonomous.

Increased evaluation and observability

As AI systems perform more actions, monitoring may increasingly focus not only on system availability but also on whether an AI system selected appropriate actions, used reliable information, followed required policies, and produced accurate outcomes.

The development of NIST's TEVV-Athlon framework reflects growing attention to systematic evaluation of large language models and agentic AI systems.

Stronger AI governance

The expansion of AI automation is expected to increase the importance of governance, auditing, access control, model evaluation, security, and accountability.

NIST's AI Risk Management Framework and Generative AI Profile both emphasize risk-management practices throughout the AI lifecycle.

Retrieval-augmented generation

Retrieval-augmented generation is expected to remain important in enterprise AI systems because it allows generative models to retrieve information from approved data sources before producing an output.

This can allow organizations to connect AI systems with internal documents, knowledge bases, databases, and other information sources.

Specialized models

Organizations may increasingly use different models for different tasks.

For example, one model may perform classification, another may generate text, and another may validate outputs.

Smaller or specialized models can also be selected when they offer sufficient performance with lower latency or computational requirements.

Voice-based automation

Speech recognition, speech synthesis, generative AI, and workflow automation can be combined to create voice-based automated systems.

Potential applications include customer service, scheduling, appointment management, information retrieval, and automated call handling.

Enterprise integration

AI automation is increasingly being incorporated into existing software rather than operating only as standalone applications.

This includes integrations with customer relationship management systems, enterprise resource planning systems, databases, communication platforms, and internal knowledge systems.

Cybersecurity and access control

As automated agents gain access to software tools and organizational information, cybersecurity controls may become more significant.

Agent permissions, credential management, authentication, audit logs, and restrictions on tool access can affect the security of automated workflows.

Outlook

AI automation represents a shift from software systems that primarily execute predefined instructions toward systems capable of interpreting information, generating outputs, supporting decisions, and coordinating more complex workflows.

Recent evidence indicates rapid growth in organizational AI adoption and investment, while deployment of agentic AI remains uneven across business functions and industries.

By 2027, AI automation may increasingly involve combinations of AI agents, generative models, retrieval systems, workflow orchestration, enterprise applications, and human oversight.

At the same time, increased autonomy is likely to make accuracy, testing, security, governance, transparency, and accountability more significant considerations.

The effects of AI automation will depend on technological development, regulatory requirements, industry practices, organizational infrastructure, and the extent to which automated systems complement or replace individual human tasks.

See also

• Artificial intelligence
• Artificial intelligence in business
• Automation
• Business process automation
• Robotic process automation
• Machine learning
• Natural language processing
• Generative artificial intelligence
• Large language model
• Intelligent agent
• Retrieval-augmented generation
• Workflow
• Business process management
• Human-in-the-loop
• AI safety
• AI alignment
• Algorithmic bias