AI Agents in 2025 and 2026: How Many Companies Use Them and What's Next

AI agents—software that uses large language models to plan, use tools, and complete multi-step tasks—have moved from pilots to production at many companies. In 2025, 57% of organizations already deploy AI agents for multi-stage workflows, and 16% use them across teams and functions. More than 68% expect to have integrated AI agents by 2026, and about 80% report measurable ROI from their agent investments. IDC projects agent use will surge roughly 10x by 2027, with token and API usage rising by orders of magnitude. Protiviti, Google Cloud, and analyst reports have all documented the shift from experimentation to scaled deployment.

Background: From Chatbots to Agents

Early generative AI was mostly question-answer and drafting. Agents add reasoning, planning, memory, and tool use: they can call APIs, run code, search the web, and chain steps without a human in the loop for each one. Enterprises use them for coding assistance, data analysis, process automation, and research. The 2026 State of AI Agents report and Google Cloud's agent-trends material describe adoption by industry and use case; barriers include integration, data quality, and change management.

Key Adoption Numbers

• 57% of organizations use AI agents for multi-stage workflows; 16% for cross-functional processes.
• 68%+ expect integrated AI agents by 2026.
• 80% report measurable ROI from AI agent investments.
• 90% use AI for coding assistance; 60% for data analysis and reports; 48% for internal process automation; 56% plan to add research and reporting in 2026.
• IDC: agent use could grow ~10x by 2027; actively deployed agents may exceed 1 billion by 2029 (up from tens of millions in 2025), with hundreds of billions of actions per day.

What Agents Do Well

Software development leads: code generation, review, refactoring, and tests. Data analysis and report generation, workflow automation, and research are next. Design trade-offs involve latency vs accuracy, autonomy vs control, and capability vs reliability. Architectures typically combine foundation models with deliberation, planning, tool-calling, and sometimes multi-agent orchestration.

Barriers to Adoption

Organizations cite system integration (46%), data quality and access (42%), and change management (39%) as main obstacles. Letting agents act on sensitive systems requires guardrails, audit trails, and clear ownership. Talent and process redesign matter as much as technology.

Impact on Work

Agents are being treated as a "silicon-based workforce" that complements humans rather than only automating old workflows. That implies new roles (agent design, oversight, prompt and tool curation) and shifts in how work is divided. Industries from healthcare and retail to finance and legal are piloting or deploying agents for document review, customer support, and internal operations.

What's Next

Expect more domain-specific agents, better tool use and reliability, and clearer ROI metrics. Regulation and liability for agent decisions will get more attention. Re-evaluating agents every 6–12 months is sensible as models and platforms improve.