AI agents can make the technology feel as though it has crossed a threshold: instead of merely producing an answer, a system may plan steps and take actions. For a small business, that makes the conversation about AI more operational. The question is no longer only what a model can generate, but what authority it should receive inside a real workflow and whether that workflow needs redesign before automation can create value.
What McKinsey's 2025 survey actually found
McKinsey's State of AI in 2025 survey, published in November 2025, reports broad organisational AI use while describing enterprise-wide scaling as much less mature. It also reports substantial experimentation with AI agents and links stronger reported AI performance with more ambitious transformation and workflow redesign.
The survey covered organisations of different sizes, but it is not a small-business instruction manual. Its findings are more useful to SMEs as signals about direction than as promises about the outcome of a particular investment.
Agents change the unit of automation
A conventional assistant may draft a response for a person to review. An agentic workflow can potentially gather information, decide the next step, call a connected system and continue towards a goal. That creates opportunities to automate complete pieces of work rather than isolated text generation.
It also increases consequences. A weak draft can be rejected; an agent with broad permissions can propagate a mistaken assumption through several actions. Small firms should therefore think about authority and recovery at the same time as capability.
Innovation does not have to mean a grand transformation
McKinsey's findings emphasise innovation and transformation among organisations reporting stronger value, but a smaller company can translate that lesson proportionately. Redesign might mean removing repeated data entry from one enquiry route or changing how information reaches the employee who makes a decision.
The important shift is from adding AI to an unchanged process towards asking whether the process itself should work differently. That can produce a modest but meaningful operational innovation without pretending the company is undertaking an enterprise transformation.
Workflow redesign starts with decisions
Map a candidate workflow and mark where information enters, where a rule is applied, where judgement is required and where an external action occurs. Then ask what an AI agent would need to see and what it should be permitted to do.
This often reveals that the technology is not the first problem. Conflicting records, unclear ownership or undocumented exceptions need attention before a more autonomous system is allowed to move work through them.
Small businesses need a narrower risk envelope
A compact business may have fewer layers between an automated mistake and a customer. Start with limited permissions, reversible actions and visible monitoring. Keep a human approval point around consequential commitments until the business has evidence that a broader boundary is justified.
The NCSC's guidance on managing agentic AI cyber risk reinforces the importance of monitoring, constrained access and the ability to halt autonomous activity. Those principles are particularly practical for teams without a large specialist risk function.
Measure transformation at workflow level first
Enterprise financial impact is a difficult benchmark for a small trial. Start closer to the work: did the process reach correct outcomes, did staff intervene less often, did customers avoid repetition, and did the new system introduce checking or recovery work?
This prevents a business from declaring success because an AI component is heavily used. Adoption is not the outcome. A redesigned workflow should be measurably easier to operate or better at serving its purpose.
Human capability remains part of agentic capability
More autonomous tools do not remove the need for employees who understand the process. They increase the importance of people who can define boundaries, recognise abnormal behaviour, maintain knowledge and decide when automation should stop.
Managers also need enough understanding to challenge a proposed deployment. If nobody can explain the agent's authority, source information, failure route and owner, the organisation is not ready to treat it as an operational system.
The small-business lesson is transformation with boundaries
McKinsey's 2025 picture is not simply that more AI produces more value. It points towards workflow redesign, innovation and deliberate organisational change while also showing that scaling remains difficult. That nuance matters.
For SMEs, agents may make previously impractical automation possible, but the sensible path is selective. Choose a workflow worth changing, improve the information and ownership around it, give the agent bounded authority and measure the complete result. Transformation can begin with one well-redesigned process. It does not require handing an emerging technology the keys to the entire business.