Every automated system leans on outside tools: an email API, a payment processor, a CRM, a website it reads from or posts to. Those tools change on their own schedule, not the automation's. An API adds a required field, a login page adds a new verification step, a site redesigns its layout, a rate limit tightens. None of that is unusual. It is just what happens when software runs against other people's software for months and years.
The reason this matters less than it sounds is how an agentic system is built: it checks the result of what it just did before moving on. A failed check does not get waved through as a success, it gets logged and, if it keeps failing, escalated to a person instead of quietly producing bad output or going silent. Because we operate the systems we build rather than handing them off after launch, that escalation reaches us directly. Running 110+ scheduled jobs across 40+ live integrations day to day means this pattern comes up regularly, and most fixes are small: update a field name, swap an authentication step, point a selector at the new layout.
What we cannot do is stop the change from happening, or promise a fix lands before the next scheduled run. There is normally a gap between a tool changing and the integration catching up, and during that gap the affected piece pauses rather than doing something wrong. A tool with no API and no stable way to be reached at all will always be the weakest link in a system like this, and that is worth weighing before automating around it.