Vibe coding was fun. Then you closed the laptop.
Most people’s first experience of Cursor, or any AI coding agent, is a burst of energy. You describe what you want in plain English, the agent writes the code, and for an afternoon you feel like you have a developer on staff. Then you close the laptop and the agent stops existing. Nothing runs while you sleep. Nothing watches your GitHub repo, your Slack channel or your inbox. The moment you walk away, the automation walks away with you.
Cursor’s 3.9 update, and the automations work that led up to it, changes that. We have been testing it at Kaizen because it is the clearest sign yet that “AI coding assistant” is turning into “AI agent that runs your business processes”, not just your codebase. Cursor is not a niche tool either: a recent JetBrains developer survey put it at 18% workplace usage among developers, level with Claude Code and closing fast on the market leaders, which is why what it ships next tends to set the direction other tools follow.
What actually changed in Cursor 3.9
Cursor 3.9 pulls plugins, skills, MCP connections, subagents, rules, commands and hooks into a single Customize page, with controls at the user, team and workspace level. On a team account you can see a leaderboard of the plugins and automations your colleagues already use and add any of them with one click, which matters more than it sounds: most small teams never discover the useful automation someone else on the team already built, because it lived in one person’s local settings.
The bigger shift happened just before 3.9, with the /automate skill. You describe the task you want handled in plain language inside a normal Cursor session, and the agent configures the trigger, the instructions and the tools itself. Automations can now fire from a GitHub event, a Slack message, a schedule, or a “computer use” trigger that lets the agent operate outside the codebase entirely, as Cursor’s own changelog for the automations update sets out. That is the actual headline: the same agent you vibe-code with during the day can now stay switched on after you have gone home, and it does not need you to write the plumbing.
What “always-on” looks like for a small business
Strip away the developer framing and the pattern is one that any small business owner will recognise, because it is exactly the “if this happens, do that” logic behind most useful automation. A GitHub issue gets filed by a client and a draft pull request is waiting by morning. A message lands in a Slack channel and the agent turns it into a task, a document or a first-draft reply. A schedule fires every Monday and a status report gets compiled from three different systems without anyone opening them by hand.
None of this requires the business to be a software company. We have used the same pattern for internal work: routine repository maintenance, first-pass content checks, and pulling structured data out of documents that used to take someone twenty minutes a day. The point of an always-on agent is not that it replaces the twenty minutes of judgement. It is that it removes the twenty minutes of typing that came before the judgement.
The uncomfortable number behind the excitement
It is worth being honest about how far ahead of most businesses this actually puts you. The British Chambers of Commerce, working with Atos and the University of Essex, found that 54% of UK SMEs are now actively using AI in some form, up sharply from 35% in last year’s survey and 25% the year before that, a figure also picked up by trade press covering the report. That is a genuinely fast climb, and it is the headline most coverage led with.
The less comfortable detail sits underneath it: the same research found that 95% of SMEs using AI say it has had no impact on workforce size, and 86% say job roles have stayed the same. Most of that 54% is using AI as a tool that a person still operates by hand, prompt by prompt, not as something that runs a process unattended. An always-on agent that fires from a GitHub issue or a Slack message on its own is a different category of adoption to asking a chatbot a question, and right now very few small businesses have made that jump. That gap is the opportunity. It closes for everyone eventually, and the businesses that automate the boring, repeatable trigger-and-response tasks now get more runway before it does.
What it actually costs to try
This is a tool cost, not a Kaizen fee, and it is worth stating plainly. Cursor’s individual Pro plan runs $20 a month with a matching monthly credit pool, as the current breakdown of Cursor’s plans lays out, which is enough to experiment with a couple of automations without committing to anything. Team automation, the shared Customize page and the plugin leaderboard sit on the Business plan, priced per seat. If you are the sort of business already paying for two or three disconnected SaaS tools to move information between systems, replacing one of them with a single automation is often the cheaper route, not an added cost on top.
The honest caveat: automations still need someone to specify the trigger and check the first few runs. “Describe the task in plain language” genuinely works, but the first version of any automation is rarely the version you keep. Budget an afternoon to set one up properly, not five minutes.
Where we would start
Pick one recurring, low-judgement task that currently involves a person copying information from one place to another: a form submission that needs turning into a task, a weekly report that gets assembled from the same three sources, a repository or document that needs a first pass before a human looks at it. Set up a single automation for that one task, watch it run for a week, then decide whether it earns a second one. Always-on agents are genuinely useful, but they are most useful when they replace a specific, boring, well-understood step rather than an entire job.
We build these automations for clients directly inside tools like Cursor and n8n, matched to the systems a business already runs on rather than asking anyone to switch platforms first.