--- title: "What small teams actually do with AI agents" excerpt: "Not the demo reel: the use cases that show up when real small teams put AI teammates to work, roughly in the order they show up, with the playbook for each." author: naomi category: Use Cases date: "2026-08-14" image: "/cover-samples/x-monitors.png" --- Ask what AI agents are for and you'll get the demo reel: book a flight, plan a party, browse the web in a jerky little video. Watch what small teams actually do with them and the list looks different: quieter, more repetitive, and much closer to the org chart. These are the use cases we see when [AI teammates](/blog/who-is-an-ai-teammate) join real companies, roughly in the order they tend to show up. ## The plumbing comes first The least glamorous pattern is the most common: before an agent is anyone's assistant, it's the person who finally connects the tools. Watching the error tracker and filing the real issues. Moving what's in the repo into the doc where the team looks. Being the one place where the CRM, the calendar, and the codebase are all reachable at once. It's unglamorous precisely because it was everyone's job and no one's. [Wiring one up takes minutes](/blog/connect-your-tools), and it's usually the first "oh, this is real" moment. ## Coding, but not the kind in the demos Engineers reach for agents early, and mostly not for greenfield features. The demand is chores: the dependency bump, the flaky test, the refactor that's been on the list all quarter, the PR review pass before a human looks. [The terminal is where a lot of teams meet their first agent](/blog/your-terminal-into-dock), and the pattern that sticks is an agent that owns the backlog's boring tail, not one that writes the product. ## Content, on a schedule instead of a burst Content is where consistency beats brilliance, and consistency is exactly what a standing role provides: [the social calendar that actually happens](/blog/run-your-social-content-with-dock-agents), drafts arriving on a routine, one voice held across channels. Humans stay on taste and final say; the agent makes sure Tuesday's post exists to have taste about. ## Sales, spread across the whole funnel Sales usage spreads across the whole funnel rather than concentrating in one step: teams put agents on [list building and enrichment, outreach drafting, reply triage, and pipeline hygiene](/blog/run-your-sales-outreach-with-dock-agents) without leaning hard on any single one. Taken together that's a new job description: [one person running a revenue machine by managing teammates instead of operating tools](/blog/gtm-engineer). What never transfers: a human decides what gets sent and what a deal is worth. ## Research: getting smart, cheaply More learning than spying: the standing brief on a market, the pre-read before every call, the "get me smart on this by Thursday" hand-off. [A researcher that never stops reading](/blog/ai-research-agent) is the quietest first hire and the one that sells the second. ## The assistant jobs: inbox, calendar, standing load The executive-assistant cluster ([triage the inbox](/blog/ai-email-management), run the calendar, carry the recurring briefs and follow-ups) is the use case people search for first and adopt slightly later, usually right after the first agent proves the pattern. [It's the first hire we'd argue for](/blog/ai-executive-assistant), because the hours it returns fund attention for everything else on this list. ## Hiring and support: the pipelines Two more queues that reward a standing role: [the hiring pipeline](/blog/run-your-hiring-pipeline-with-dock-agents) (sourcing, screening notes, scheduling loops, candidate follow-ups) and [customer support](/blog/run-your-customer-support-with-dock-agents) (triage, drafts, escalation with context attached). Same shape as the inbox: the queue gets processed the moment things land, and humans spend their time only where judgment lives. ## The record-keeping nobody misses doing The sleeper category: [the weekly report that writes itself](/blog/run-your-weekly-reporting-with-dock-agents), the docs that stay current, [the company that remembers what it decided](/blog/a-company-that-remembers). No one buys software for this. Everyone keeps the agent that does it. ## What the list adds up to Two things are true across every category. First, the wins are roles, not tasks: the agent that owns a lane beats the tool that performs a trick, every time. Second, almost nobody stops at one: the first working use case makes the adjacent one visible, because the wiring, the memory, and the trust carry over. Which is why the practical question isn't "which of these is best." It's which queue is eating your week *right now*. [Start there, today](/blog/launch-your-ai-agent-in-30-seconds), and let [the rest of the org chart follow](/blog/run-your-startup-without-hiring-more-humans).