Adding AI agents to old processes rarely pays off
In 2026, AI agents are moving out of pilots and into real business processes, but the profit has not caught up. Among large companies, 40% now report scaling AI agents, up from 27% a year earlier, according to McKinsey's State of AI 2026, while the share reporting any EBIT impact from AI stayed flat. The evidence points to a common cause: companies add agents to old processes instead of redesigning the process around them.
How far have AI agents moved into business processes?
Further than a year ago, and mostly in large companies. In the McKinsey survey of 1,719 people (fielded May to June 2026), the share of large organizations scaling agents in at least one function rose from 27% to 40%, while smaller organizations stayed flat at 22%.
Share of respondents whose organization is scaling AI agents
- Large organizations, 2025
- 27%
- Large organizations, 2026
- 40%
- Smaller organizations, 2026 (flat vs 2025)
- 22%
Source: McKinsey, The state of AI in 2026 (n=1,719, May to June 2026)
Scaling is still narrow. In the 2025 edition, 23% of respondents were scaling an agentic system somewhere and another 39% were experimenting, but no more than 10% were scaling agents in any single business function. Agents also multiply without a plan: organizations run 12 agents on average and 50% of those agents work in isolated silos, according to the Salesforce 2026 Connectivity Benchmark, a vendor survey of 1,050 IT leaders.
Are companies getting a return on AI agents yet?
Individuals are, companies mostly are not. Eight in ten respondents told McKinsey that AI improved their own productivity, yet only 37% attribute any EBIT impact to AI, about the same as in 2025. The group McKinsey calls AI high performers (at least 5% of EBIT from AI and "significant" value) stayed at about 6% of respondents.
Cost is starting to bite too. About one in five organizations say AI operating costs now constrain their AI use, per the same survey. Personal time savings do not show up in the profit and loss statement unless the process itself changes: fewer handoffs, fewer manual checks, shorter cycle times.
What are the main challenges in agentic process transformation?
The data points to five recurring problems.
- Projects without clear value get canceled. Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. That is a forecast, not a measurement, but it matches the flat EBIT numbers above.
- Agent washing. In the same release, Gartner estimates that only about 130 of the thousands of agentic AI vendors are real; the rest rebrand assistants, chatbots and RPA.
- Risk, skills and legacy systems. In the Salesforce 2026 survey, the top hurdles were risk, compliance and legal concerns (42%), lack of internal expertise in agent design (41%), legacy infrastructure (37%) and siloed apps and data (35%).
- Integration debt. In the same survey, 86% of IT leaders worry that agents will add more complexity than value without proper integration.
- Quality. Among teams that build agents, 57.3% have them in production and quality is the top barrier at 32%, according to LangChain's State of Agent Engineering (a tooling vendor's survey of 1,340 practitioners).
Top hurdles to agentic transformation, share of IT leaders
- Risk, compliance, security or legal
- 42%
- Lack of internal expertise in AI and agent design
- 41%
- Legacy infrastructure or system incompatibility
- 37%
- Integrating siloed apps and data
- 35%
Source: Salesforce (MuleSoft), 2026 Connectivity Benchmark Report, vendor survey (n=1,050, Oct to Nov 2025)
Why does process redesign matter more than the model?
Because the companies that get value from AI redesign the work. In McKinsey's 2025 survey, high performers were nearly three times as likely as others to have fundamentally redesigned individual workflows, and workflow redesign had one of the strongest links to business impact of all the factors tested. In 2026, high performers were more than three times as likely to be scaling agents in most business functions.
Gartner reaches the same conclusion from the failure side. Its June 2025 release warns that integrating agents into legacy systems is technically complex and that rethinking workflows from the ground up is often the ideal path. An agent placed on top of a ten-step approval chain still waits for ten approvals. An agent placed in a process redesigned around what it can check, draft and route removes steps.
What does good governance for AI agents look like?
Most organizations are not there yet. A Q2 2026 Gartner survey of over 2,000 CIOs and technology executives found that 71% expect to deploy agentic AI within 12 months, but only 21% have agentic AI governance capabilities in place, per a Gartner release from October 2026. Only 54% have a centralized governance framework for their agentic capabilities, according to Salesforce.
Practitioners already keep agents supervised: 60% block agents from making unapproved system changes, according to a Stack Overflow pulse survey of about 1,100 technologists. High performers in the McKinsey 2025 survey are also more likely to define when model outputs need human validation. For processes in the EU, the AI Act high-risk obligations for Annex III use cases now apply from 2 December 2027.
Good governance in practice means:
- an owner for every agent and every process it touches
- scoped permissions, so an agent can only read and write what its step requires
- human approval on actions that are hard to undo, such as payments, contract changes and customer messages
- logs of every agent decision, kept long enough to audit
- automated tests and evaluations that run before any change to an agent's prompt, tools or model
How should you start an agentic process transformation?
Start with one process, not a platform.
- Pick a process with a measurable cost. Invoice matching, ticket triage, onboarding or quote preparation all have clear cycle times and error rates.
- Map it before you automate it. Count steps, handoffs and waiting time. Remove steps that exist only because humans were slow at them.
- Decide what the agent does and what a person decides. Write it down as part of the process spec.
- Connect the data properly. Agents fail on missing context; that is integration work, not prompt work.
- Test like software. Build evaluation sets from real cases, run them on every change, and track the business metric you picked in step 1.
- Scale only what pays. Gartner's advice is to pursue agentic AI only where it delivers clear value or ROI. Kill the rest early.
Ready to redesign a process around AI agents?
SDCLabs has built software since 2011. Our way of working: AI agents draft the specs, code, tests and reviews, senior engineers review and decide what ships, and automated tests verify every change. We are the founders of the open-source testing frameworks Codeception and CodeceptJS. If you have a process that is slow, manual or error-prone, see our agentic process transformation service or tell us about it; engagements start in 1 to 2 weeks.
For the engineering side, read the state of AI engineering in 2026 or see how we build AI agents.
Sources checked on 2026-11-17.
Frequently asked questions
How many companies are scaling AI agents in 2026?
In McKinsey's State of AI 2026, 40% of respondents from large organizations reported scaling AI agents, up from 27% a year earlier. Smaller organizations stayed flat at 22%.
Do AI agents deliver a return on investment?
Not yet for most companies. Only 37% of respondents attribute any EBIT impact to AI, about the same as in 2025, according to McKinsey, even though eight in ten say AI improved their own productivity.
Why do agentic AI projects get canceled?
Gartner predicts that over 40% of agentic AI projects will be canceled by the end of 2027 because of escalating costs, unclear business value or inadequate risk controls. It recommends rethinking workflows from the ground up rather than bolting agents onto legacy systems.
Do companies have governance in place for AI agents?
Mostly not. A Q2 2026 Gartner survey found that 71% of organizations expect to deploy agentic AI within 12 months, but only 21% have agentic AI governance capabilities in place (Gartner, October 2026).
