How often do AI automations break after delivery, and how can we minimize maintenance risk?
Automations do break pretty often, depending on your proficiency as a developer; even the best development process will have bugs and unknown unknowns. I experience systems breaking from time to time. I have a reasonable ask policy: if I think it's my fault (which is most things), I'll fix it free of charge. However, I often use the risk of an unmaintained system as an upsell to get clients onto a retainer, so I now have very few clients whose systems break and I'm not on a retainer to fix. I recommend treating system breakdowns similarly: from the client's perspective, a broken system is scary, so being available to fix it is a strong value proposition. They do break reasonably often. I use error notifications and similar measures to keep issues from becoming too big. Failures happen days later pretty often, weeks later less often, months later very unlikely. You need to have provisions in place for when a delivered system becomes non-functional.
Creator says he documents all systems and makes clear that maintenance is only covered when the client pays a monthly maintenance fee or is on a retainer; otherwise, after a short grace period (e.g., a week), there is no ongoing obligation. He will provide documentation but expects the client to handle fixes themselves or pay for support.
I haven't thought about that in detail, so I don't have a detailed backup plan. I expect the core work of running an automation agency to remain viable for at least three to five years. Over that period the work will evolve—no‑code AI tools like pre‑draft generators will make many tasks more efficient. That doesn't mean you won't be able to run an agency; agents and other services will start to infiltrate the business model soon, but the model itself remains sound because the same shift will happen across the industry.
Yes, there is ongoing maintenance, but instead of offering a binary choice (retainer vs. no retainer), give clients three options: 1. Self-Service: Deliver the system along with a video walkthrough and Google Doc guide so they can maintain and adjust it themselves. 2. Maintenance Retainer (e.g., $985/month): Provide a service level agreement (SLA) to fix any broken integrations or microservice failures within 48 hours, plus an emergency support thread. 3. Growth/Peace-of-Mind Retainer (Higher Ticket): Include full maintenance under a faster SLA, plus build two new systems per month, conduct team training, host weekly calls, and act as a fractional CTO. By offering three options, you shift the decision from 'pay vs. don't pay' to choosing a service tier, reducing the drop-off rate. For standard maintenance retainers, cap your actual time at 3 to 4 hours per month. Minor maintenance issues (like retrying failed executions on Make due to temporary third-party API downtime) take only a few minutes, making a $985/month retainer very profitable. If maintenance requires extensive hours, it's usually an architectural flaw in how you designed the input validation or workflow.
I offer a maintenance period—usually a week or two depending on the system size. If a client gets on a retainer, I’ll maintain the system for life. I frame this as peace of mind: a retainer gives them a 99% SLA, guaranteed uptime, and fixes within 24 hours if anything breaks. The retainer also includes extra benefits like builds, regular meetings, daily Slack availability, and monthly team training, which makes it an attractive sell.