How Generative AI Informs Platform Engineering Strategy
Editor’s note: This article is an excerpt from the Manning Early Access Program (MEAP) book Effective Platform Engineering by Ajay Chankramath, Nic Cheneweth, Bryan Oliver and Sean Alvarez. In MEAP, you read a book chapter by chapter while it’s being written and get the final ebook as soon as it’s finished.
Given the tremendous growth in interest and innovation around large-language models (LLMs) and other forms of generative AI (GenAI), it is important to note how it is impacting approaches for platform engineering and building platform experiences.
How AI Can Help
We think of AI’s purpose as complementing and augmenting the proposed goals of platform engineering (reducing developer friction and improving the developer experience). AI helps platform engineering solutions scale with increased usability, reliability and extensibility that take into account security, observability and sustainability considerations.
Two AI Approaches
Predictive AI (leveraging AI techniques to answer various what-if scenarios in platform engineering to make the decision-making process more straightforward) and applied AI (leveraging AI techniques to solve multiple software development life cycle [SDLC] challenges) have always been part of the platform engineering space.
A critical approach to predictive AI that is becoming more prevalent is the increased usage of decision trees and models to select a specific approach when there are many choices. An example of using predictive AI helpfully is when you are trying to look at your application pattern and figure out the appropriate runtime configuration.
As of the writing of Effective Platform Engineering, from which this article is excerpted, major public cloud providers offer a wide variety of compute options — including virtual machines (VMs), serverless and Platform as a Service (PaaS) — spanning over a dozen distinct services to run your applications, tailored to different workload needs, Kubernetes, batch and step functions, and more. This number will increase as technologies evolve, which will require a codified approach to making the right decisions — most probably through your engineering platform. These can be highly complicated for the developers and architects to recommend.
Moreover, inconsistent usage of architectural patterns and associated solutions can create more operational challenges for all the downstream teams.
Applied AI, on the other hand, has always had a significant impact on any data-driven decision-making process. This is where you can apply your learnings from prior executions of similar solution patterns to a problem to a current scenario. The critical concept of self-healing is an example of using applied AI.
Both predictive and applied AI should be reconsidered under the new interest in GenAI.
Generative AI Emerging
GenAI can already generate complete or partial solutions to platform engineering problems with sufficiently clear prompts. The challenge is the availability of appropriate LLMs to help generate appropriate and contextually aware solutions for a given organization.
We defined the product strategy for a platform in “Foundational Concepts in Platform Engineering,” a previous excerpt from our book. While describing the need to understand the overall product vision and strategy, we also clarified that strategy creation would not be the primary task of many readers.
However, executing a specific roadmap and actions around it require a clear understanding of the underlying strategy. In this context, we must consider the impacts of GenAI and agentic AI, as several tools in the market assist in creating a platform engineering strategy.
Identifying Areas of Focus
The best starting point to identify the areas of focus for LLM-orchestrated control planes will be to create a first-pass strategy, followed by a detailed path-to-production analysis to identify the areas within the SDLC where you have the most inefficiencies. Once you have that, you should look at the places you plan to invest in within the following phases under the general categories.
Focus areas for GenAI in platform strategy. This is a rapidly evolving space, and focus areas should be identified using the same product strategies discussed throughout this chapter.
Evolving Strategy With GenAI
From a strategic point of view, you should first identify the focus areas and then break down the problem into multiple steps. Let’s look at this for each focus area.
1. Plan
Typically you would begin by conducting empathy interviews, which allow product managers to gain a deep understanding of motivations, goals and challenges. This means examining the available data exposed using a detailed path-to-production analysis, as well as examining some lagging indicators, like DORA metrics, to better understand the problem space.
GenAI can be effective in analyzing the planning process’ output to identify priorities. One critical element of prioritization that GenAI can solve is understanding potential conflicts or inconsistencies with requirements.
Agentic AI can also turn planning and acceptance criteria analysis into actual user stories. Textual assistance can accelerate (if not completely replace) the effort of business analysis and product owners by reducing some of the toil necessary to package prior work into various forms based on the audience.
2. Design
GenAI provides numerous tools to create architectural diagrams, data models and deeply descriptive visualizations for your designs. It translates its knowledge of design patterns similar to the models used to improve the design experience and the product’s usability and quality.
For example, there are tools to automatically create entity relationship diagrams (ERDs) based on your design specifications. They can also provide you with options to select from, driven by either the design criteria set forth at the outset or enabling an iterative process. Options generated by the AI models can tell you the potential cost and performance implications of implementation to support your decision-making process.
When putting together a strategy, the biggest challenge is figuring out how soon you can prove your hypothesis. Addressing this challenge requires you to build proofs-of-concept that can give your end users and decision-makers a better feel for what you are proposing to develop. Simply showing them a document or a spreadsheet with numbers goes only so far in aligning them with your thinking. GenAI can be a potent tool by providing a streamlined way to generate your prototype quickly.
3. Develop
Since the release of ChatGPT-4 and the rapid evolution of advanced GenAI systems like Claude, Gemini and open source agentic frameworks, some of the most debated — and often alarming — narratives have centered on whether these technologies will eventually replace software developers.
While we think this is still a far-fetched idea, the fact is that using AI will make good developers better and more productive. Specifically, this includes:
- Facilitated code generation based on the design with all the necessary documentation.
- Smart code reviews that are more timely for faster feedback.
- Debugging to ensure functional equivalence.
Another bane of developers is developing unit tests that can help continuous integration go through seamlessly. Some of the most popular assisted coding tools provide developers with great real-time insights to make the development process more efficient and enjoyable with clear outcomes.
If you are not a startup or a scaleup, you are bound to have a significant amount of legacy, suboptimally architected code. While the models still need improvements to take a tightly coupled architecture and break it down into manageable modules, many new tools can help with tasks that are difficult to find developers for, such as migrating mainframes and codebases. Considering these in your platform strategy will help you set proper investment and focus levels.
GenAI can help improve development velocity in the following ways:
- Analyzing complex coding issues: LLM-generated solutions can help resolve difficult-to-solve coding challenges.
- Improving test coverage: GenAI can generate certain categories of tests, reducing coding time.
- Feedback on static code analysis: Code complexity and maintainable issues can be discovered sooner and possible solutions implemented more quickly.
3. Tests
Testing has the highest potential of all the opportunities for GenAI to improve the SDLC. The scope of this work typically centers around informing test creation, evaluating test data and testing metrics against goals..
Test evolution from a GenAI point of view. GenAI techniques can help perform efficient root cause analysis by quickly evaluating observability data.
Testing can produce a more holistic impact by focusing on test-case generation and automation. Once the tests are executed, GenAI techniques can improve root cause analysis by allowing you to review your metrics, logs and traces through a heavy lens of observability to detect anomalies.
4. Deploy
GenAI will play a major role in releasing software across various user environments and improving customer experience. These include security checks based on anomalies in usage patterns and compliance changes.
5. Maintenance
GenAI can improve software support and maintenance. Here are some areas to consider when developing a software maintenance strategy within your platform buildout.
- Improving end-user experience: Consider how you might use natural language processing (NLP) and deep learning chatbots to create real-time responses to the most common questions your system users might have.
- Alerting and resolution: Continuously identifying drifts against sensible defaults and acceptable ranges within your system can lead to informed alerting and insights that can result in automated remediation.
- Feedback cycles: GenAI can rapidly discover usage patterns that can guide roadmap prioritization decisions.
Wrapping Up
You must evaluate the impact of GenAI — including cutting-edge tools like ChatGPT-4, Claude, Gemini and emerging agentic frameworks — as a core component of your evolving platform strategy, just as you would assess any product innovation.
Despite advancements, LLMs and machine learning (ML)-generated outputs can still be inconsistent, often producing inaccuracies. Yet, they’re also capable of delivering significant value through accelerated ideation, prototyping and decision support. By leveraging platform value measurement techniques, you can continuously assess whether the return on investment from integrating GenAI tools and agents meaningfully advances your platform’s outcomes.
