Most MSP shops have figured out by now that ChatGPT or Claude can generate PowerShell scripts and API calls on demand. The developer community calls it “prompt engineering.” It works. Scripts run, problems get solved. But let’s talk about what happens when you want that to scale: automation running across dozens of client environments, handling edge cases, reliably, six months from now. 

I’ve watched MSPs discover prompt engineering and have an immediate reaction: this is incredible. And honestly, I get it. You describe what you want, a script comes out the other end, and it mostly works. The bar for creating automation dropped significantly. That’s real. I’m not going to pretend otherwise. 

But there’s a difference between creating something and being able to run it reliably across your clients. That gap is where I think the conversation about AI tools for MSPs gets interesting. 

Seventy-one percent of MSPs on a dedicated MSP automation platform are using AI as part of how they build and run automation. Among MSPs who aren’t on a platform: 38%. 

What prompt engineering actually gets you 

General-purpose LLMs like ChatGPT or Claude are good at building software. They can write applications, prototype tools, generate APIs, help you learn unfamiliar platforms faster. I use them for that kind of thing every day. 

Where things get harder is moving from something that works to something your business depends on. AI can produce the logic. It doesn't know your customers, your operational policies, or the systems that make your MSP run. 

The limitation shows up when you want to operationalize something. When you want an automation that touches customer environments every night, runs across multiple tenants, handles approval flows, and doesn’t require a human to check whether it worked. That’s a different problem. 

A script doesn’t know your tenant structure. It doesn’t know which clients have approval processes, which have specific business rules, or how your environment is wired together. It knows what it was trained on, and then it guesses. 

What context-aware AI actually gets you 

A standalone chat interface doesn’t know your environment. Prompt engineering can close some of that gap: you feed it your org variables, your tenant structure, your integration setup. But that's significant work to maintain, it resets every session, and I'd argue that time is better spent building the automations themselves. 

Prompt engineering can get you partway there. But your environment is a moving target: integrations get added, tenants get reconfigured, org variables change. Keeping that context current becomes its own job. 

The Rewst RoboRewsty AI Workflow Builder is built inside the platform. When you open it on a workflow, it can read what you give it access to in your live environment: your integrations, your org variables, your execution history, your tenant structure. When I say it “closes the implementation gap,” I mean that specifically. The output is a running automation, already wired to your environment. 

Code generated by a general LLM requires you to figure out where to put it, how to authenticate it, how to adapt it to your specific setup. RoboRewsty does that work inside the Rewst platform where the context already lives. 

The expertise question 

In our State of MSP Automation report, 54% of MSPs cited lack of expertise as their top automation barrier. I get asked whether RoboRewsty solves that. My honest answer: it changes where expertise is required, more than it eliminates the requirement. 

What it takes off a technician’s plate is the how: the syntax, the wiring, the mechanics of connecting an integration to a workflow action. What it still requires is understanding the problem well enough to describe it clearly. You still need to know your environment, know what outcome you want, and be able to recognize whether the output looks right. 

I’d actually argue that’s a better use of a technician’s time. Knowing what to automate and why matters more than knowing exactly how to write a Jinja template. The expertise moves up the stack, which is good. 

When things break 

You run a script manually a few times. It works. You set it on a schedule and move on. Six weeks later, a client tells you something has been wrong for a month. By then you’re doing detective work: digging through logs, trying to remember what the script was supposed to do, figuring out if it was the script or something downstream. 

An automation platform built for MSPs handles this differently. Failures are logged, visible, and tied to specific execution runs. RoboRewsty can read those logs; it can look at actual failed execution history and surface what went wrong. The human isn’t doing the detective work; the system is. 

This matters practically. An MSP running automation at scale across dozens of clients can’t afford silent failures. At scale, observable failures are a requirement for running this use case operationally. 

Compounding vs. starting over 

One thing general-purpose AI does consistently: it starts from zero. You get something useful, and then it’s gone. The Rewst platform accumulates context. Sub-workflows you build in Rewst get reused across dozens of other automations. Forms, variables, integrations: built once, they become the foundation for the next thing. Over time, RoboRewsty’s understanding of your environment deepens because the environment itself grows. Engaging with the Rewst Community can add even more to this: an automation shared during a weekly Community Open Mic session or discussed in one of our Discord channels often gets exported or published to a public Git repository, where others can import it, customize it, and continue building on top of it.

The MSPs treating automation as a revenue stream got there through compounding. Each workflow they built made the next one faster to deploy. Dozens of our customers have already identified processes their clients run, built automations around them, and are now charging clients for those automations on a flat rate or per-run basis. Some are building custom integrations for their clients’ specific tool stacks. MSPs using AI within their automation see 41% significant revenue growth, and they’re 43% more likely to be actively delivering Automation as a Service. That model requires a platform built for MSPs. You can’t build it on a series of scripts and chat sessions that don’t talk to each other. 

Our State of Automation report mapped MSPs across five maturity levels, from those just getting started with automation to those running full programs across their business. At Level 1, 26% are using AI as part of how they automate. At Level 5, it's 90%. The further along the program, the more central AI becomes. 

I kind of get a dopamine rush when I complete a really intricate automation. The good kind, where you can actually trace through the logic and understand every step. Scripts don’t give you that. They give you a result you’re not sure you trust. 

Where generic AI still belongs 

This isn't a capability argument against general-purpose AI. With enough effort, you can build an automation platform on top of a generic LLM. Deployment pipelines, tenant isolation, approval workflows, execution history: all buildable. 

The question is whether that's your actual job. At Rewst, we don't build our own cloud infrastructure or authentication systems. We buy those because we're in the business of automation software, not infrastructure. MSPs face the same decision: invest in building and running your own automation stack, or buy software built for it and spend that energy on customer outcomes. 

The differentiator isn't the AI model anymore. Everyone has access to powerful models now. It's what's built around the AI: the observability, the reliability, the tenant structure, the audit trails. Everything that makes automation maintainable over years, not weeks. 

What’s coming with automation and AI 

Rewst’s next iteration takes it further: an AI-native MSP automation platform where the AI agent is the primary build tool from day one. 

Describe what you need, watch the AI agent build a complete workflow on a canvas you can see and edit, and deploy it the same day. Every action, every conditional, every integration step is visible and editable before anything runs. For MSPs serving regulated clients, that’s how AI gets through the door. 

App Builder extends that capability outward: client-facing portals and dashboards from the same platform, without adding a separate tool to deliver them. 

Beyond that, there’s an agentic layer. Agents that interact with your clients through whichever channels they use: Slack, Teams, email. The agent pulls from your PSA, your monitoring tools, and your ticketing system to handle requests without a technician in the loop. You define the data sources. You set the guardrails. That’s the move from reactive support to proactive service delivery, without proportional headcount. 

Want to see what the next iteration of Rewst looks like in practice?

We’re running a three-part series of free live sessions in July for Rewst customers about the future of automation: Register once for all sessions

Danial Gagne
VP of Product