/ AIApril 20267 min read
From Generative to Agentic: What 'AI Agents' Actually Mean for a 10-Person Business
Two years ago the AI conversation was about generation: tools that write, draw and edit when a human presses the button. The 2026 conversation — McKinsey calls it the defining shift in AI-powered marketing — is about agency: systems that carry out multi-step tasks on your behalf. Generative AI makes things. Agentic AI does things.
If you run a small business, you've probably heard 'AI agents' used to describe everything from a chatbot to the end of civilisation. The buzzword fog is a shame, because underneath it is the most practically useful AI development yet for companies without big teams — and a few real limits you should understand before trusting one with your customers.
Here's the plain-English version.
/ The difference, in one scenario
A generative tool, asked about a customer enquiry, will draft you a reply — then wait for you to read it, edit it, send it, log it and remember any follow-up. Useful, but you're still the operator; the bottleneck is still your attention.
An agent handles the enquiry. It reads the message, answers the question from your approved knowledge, asks the qualifying questions you defined, books the caller into a free calendar slot, writes the record into your CRM, and messages you only if something needs human judgement. Same underlying intelligence — but wired to your tools and trusted with a defined job, start to finish.
That wiring is the whole trick. An agent is a language model plus your calendar, your CRM, your WhatsApp and a set of rules about what it may and may not do. No single piece is magic; the composition is.
/ What agents genuinely do well today
Speed-critical, high-volume conversation is the sweet spot: answering enquiries in seconds around the clock, qualifying leads with consistent questions, chasing quotes that went quiet, sending appointment reminders, handling the twenty questions your team answers forty times a week. These tasks share a profile — repetitive, rule-describable, punishing to leave undone — and they're exactly where a 10-person business bleeds opportunity, because nobody's job is to sit by the inbox at 9pm.
The gains aren't hypothetical. Speed-to-lead research shows first-minute responses lifting conversions by triple digits, and across our own client book, instant AI follow-up is the single change most responsible for a 217% increase in qualified meetings booked. Not because the AI is a brilliant salesperson — because it's never late, never busy and never forgets.
/ Where humans stay in charge
Agents fail when the task needs judgement they weren't given: negotiating price, handling a genuinely upset customer, making promises about edge cases, anything with legal or clinical weight. The businesses that get burned are the ones that deploy an agent as a fire-and-forget employee rather than a well-briefed junior with strict escalation rules.
The design principle that works: agents handle the sprint, humans handle the judgement. Define exactly when the agent hands over — by topic, by deal size, by customer request — and make the handoff carry full context so nobody repeats themselves. Review transcripts monthly and tighten the rules. Treat it like managing a new team member, because operationally, that's what it is.