Are Installers Dead?

Exasol Personal is a free-for-individuals edition of Exasol that users deploy to their own AWS account. Before they can even run the installer, they may need to set up IAM policies, create a user with access keys, and configure the AWS CLI locally. This is a chicken-and-egg problem: you can’t automate AWS setup if the permissions to do so don’t exist yet. Even users with solid AWS experience struggled with this prerequisite phase.

We tried solving it with documentation, but the product evolves fast and instructions went stale. We considered generating better docs with AI, but users still had to follow multi-step guides to the letter, and that alone was enough to put people off.

So we built an AI skill instead. It’s a structured set of instructions that an AI coding assistant (in our case, Claude Code) follows interactively, walking through the same steps the installer covers, but conversationally. It asks questions, runs commands, inspects the output, and adapts based on what it finds.

We expected a modest UX improvement, but the actual benefits were larger than that. First, the skill stays current in a way that a script cannot. The model brings broad, reasonably up-to-date knowledge of AWS, so when a CLI flag changes or an API response looks different, it adjusts. Our project-specific instructions (path locations or CLI syntax) are a thin layer on top of that general knowledge.

Second, and more importantly, troubleshooting happens inline. When a command fails, the AI reads the error and either fixes the problem or explains what went wrong. Users resolved IAM permission gaps, PATH issues, and region mismatches without leaving the chat window. None of these recovery paths were explicitly scripted, the model handles them because it understands what the commands are supposed to do.

We didn’t throw away the installer. The AI skill calls it for the actual infrastructure provisioning, because deploying and tearing down cloud resources needs deterministic, repeatable code. The division of labor turns out to be clean: the installer handles what must be exactly right every time (infrastructure state, credentials, teardown), and the AI handles what varies between environments (prerequisites, configuration, debugging).

This raises a practical question about where installers are headed. Much of what a traditional installer does is guided UX: checking prerequisites, collecting input, validating state. That’s precisely what an AI assistant handles well. As AI adoption grows, installers can probably get thinner, focusing on the bare mechanical operations and leaving the interactive guidance to the model. They’re not dead, but they might not need to do as much as they used to.

Skill Repository: GitHub – exasol-labs/exasol-agent-skills: Exasol skills for AI agents. Optimized for Claude Code and OpenAI Codex 🤖

Exasol Personal: GitHub – exasol/exasol-personal: The High-Performance Analytics Engine — Free for Personal Use