Authors: Bryan Barton and Kassandra Svoboda, Precisely Platform Engineering
Key takeaways
- AI agents are most effective in platform engineering when they improve decision-making context.
- A two-phase workflow (planning + guided implementation) helps teams reduce manual cross-referencing and catch gaps before changes execute.
- Human review and approval remain central; AI handles information synthesis, not execution authority.
Better Decisions Start With Better Context
Complex platform changes are rarely difficult because engineers lack skill or effort. More often, they’re difficult because the information needed to make good decisions is spread across many places: release notes, infrastructure code, deployment workflows, cloud state, operational runbooks, and the history of previous changes.
That creates a familiar challenge for platform teams. Even when a change is well understood in principle, it can still require careful coordination across systems, dependencies, and review steps. The cost of missing a detail isn’t usually a lack of automation. It’s incomplete context at the moment a decision needs to be made.
That was the problem we set out to solve.
At Precisely, we built an internal AI-assisted workflow designed to help engineers gather and organize the information required for complex platform work before implementation begins. The goal was not to remove human judgment from the process, it was to support it with a more complete picture of the work.
Designing for information completeness
Our team approached the problem with a simple principle: for high-consequence operational work, AI is most useful when it helps engineers reason with more context, not when it acts independently.
Instead of building a system that executes changes on its own, we built an agent-based workflow that helps with three things:
- Identifying relevant change requirements across multiple sources
- Checking infrastructure definitions against the current environment
- Turning findings into a structured implementation and review plan
The result is an internal platform workflow that helps engineers move from scattered inputs to a more reliable plan of action.
What does an AI-assisted platform change workflow look like?
The workflow is built around a set of focused agent skills. Each skill is responsible for one part of the process, such as:
- Analyzing change documentation and release guidance
- Reviewing infrastructure configuration for required updates
- Validating assumptions against the live environment in read-only mode
- Generating structured implementation tasks and verification steps
We use these skills in two broad phases.
Phase 1: AI-assisted planning
In the planning phase, the workflow gathers information from the sources an engineer would normally inspect manually. It reviews change guidance, examines relevant configuration, and checks the current environment to identify dependencies, required updates, and areas where assumptions should be verified before execution.
This phase produces a concrete plan rather than a generic summary. Engineers get a checklist of recommended actions, validation points, and follow-up items that can be reviewed before any change proceeds.
Phase 2: Guided implementation with human oversight
In the implementation phase, the workflow supports engineers step by step. It helps translate the plan into executable changes, proposes updates, and surfaces the reasoning behind them.
Human review remains in place throughout the process, and any mutating action still follows the normal approval and delivery controls used by the team.
This makes the workflow useful not just for drafting changes, but for improving confidence in the path from planning to execution.
The impact of AI-assisted change management on platform teams
The biggest benefit was not speed alone. It was consistency.
By using AI to gather context across multiple systems and present it in a structured way, the team was able to reduce the amount of manual cross-referencing required during complex platform work. Engineers spent less time reconstructing state from scattered sources and more time reviewing decisions with the right information in front of them.
That produced several practical improvements:
- Clearer pre-work before implementation begins
- Better visibility into dependencies and environmental drift
- Stronger review artifacts for human approvers
- A more repeatable process for similar classes of operational change
In other words, the workflow helped us improve the quality of preparation, which in turn improved the quality of execution.
Where does AI fit in operational engineering?
There’s a lot of discussion in the industry about using AI to automate engineering work. Our experience has led us to a narrower and more practical view.
For complex infrastructure and platform tasks, the most valuable role for AI is often not autonomous execution. It is disciplined assistance.
AI can be especially effective with platform engineering when it helps teams:
- Synthesize information from multiple technical sources
- Surface gaps between intended and actual state
- Organize work into a more reviewable sequence
- Make implicit operational knowledge more explicit and reusable
Those benefits matter because platform work often depends on details that are easy to miss when they live across documents, code, and live systems.
Building on open and familiar patterns
We also wanted this workflow to fit naturally into the tools engineers already use. The system is built on common development patterns, standard interfaces, and human-readable instructions so that the logic behind each step can be reviewed and improved over time.
That was important for adoption. Engineers are more likely to trust an AI-assisted workflow when they can inspect how it works, understand its role, and keep humans in control of the final decisions.
What we learned about using AI agents for complex platform changes
The main lesson from this work is that high-quality execution starts well before implementation. In complex platform changes, the real challenge is often not writing the change itself. It is collecting the right context early enough, and presenting it clearly enough, for engineers to make sound decisions.
AI can help with that.
Used carefully, it can reduce the effort required to gather and organize technical context, improve consistency in planning, and strengthen the quality of review before changes move forward.
That’s the role we believe AI is best suited to play in this kind of engineering work: not replacing human ownership, but helping teams operate with a more complete and more reliable understanding of the systems they manage.
