, and it’s transforming the best way we live and work. For corporations, this revolution presents a dual opportunity. On one hand, the prospect to resolve previously really complex problems and to construct incredible recent products and features. Then again, the promise to spice up efficiency across a wide selection of tasks.
Many resources cover one opportunity or the opposite, but as a ML/AI Product Manager, I’m actually curious about considering the 2 opportunities at the identical time. What this actually means is: the best way to successfully develop AI-powered solutions, accelerated by AI itself. Based on my experience and learnings, I’ve split the method into five phases: ideation, design & plan, development, deployment, and impact and monitoring. In each phase, we’ll cover “” must occur, but additionally “” to make use of AI to spice up efficiency and quality.
Phase 1: Ideation
The primary and most important step is to do not forget that AI is a tool, not an answer. At all times start with the issue it’s essential to solve. It needs to be directly aligned along with your company’s high-level OKRs and validated with evidence from user research and data.
Once the issue is clearly defined, brainstorm quite a lot of solutions. This could include each traditional non-AI approaches and potential AI-powered features. Prioritize these solutions using a structured method. A framework like RICE (Reach, Impact, Confidence, Effort) means that you can make a data-informed decision by weighing the potential value of every solution against its cost. For AI solutions, do not forget that “Effort” includes coping with AI’s complexity similar to data acquisition, system evaluation, or determining required guardrails.
To make certain this blog post shouldn’t be too abstract, I’ll use my favorite marketplace use case for instance. A standard user pain in marketplaces is the effort and time it takes to list recent items (e.g. determining the fitting price, category, writing the outline…). Data allows to quantify this problem: a high percentage of users who start creating a list but never finish.
To handle this, you would consider traditional non-AI solutions like offering templates, providing suggestions for every field, or making a higher onboarding process. Or, you would explore AI-powered solutions, similar to using a big language model (LLM) to generate a product description or suggest a category. AI is a very cool tool though, as it could be applied to multiple use cases and applications, problems that was hard are actually feasible, and it lowers the entry barriers to predictive models.
⚡️ Accelerating the Ideation Phase with AI
- Tools for Brainstorming: AI chatbots like OpenAI’s ChatGPT, Google’s Gemini, Anthropic’s Claude, or Perplexity can act as an additional brainstorming team member. You possibly can prompt them along with your user’s pain points and ask for a wide selection of potential solutions, each traditional and AI-based. Consider and test different AI chatbot “flavors” available: getting answers from easy LLMs, getting answers with LLM leveraging “reasoning” capabilities (Chain of Thought prompting or “reasoning” model versions, e.g. Gemini 2.5 Pro, OpenAI o3,…), getting answers with LLMs using web search results, and Deep Research functionalities.
- Tools for Knowledge Management: Platforms like Notion AI, Mem, Tettra, or Glean can make it easier to organize your research and concepts, using AI to connect with relevant internal knowledge and data.
Phase 2: Design & Plan solution
With a prioritized GenAI solution in hand, implementation needs to be designed holistically across 4 dimensions:
- Capturing user input and relevant context (moving from prompt to context engineering)
- Choosing and configuring the fitting model (balancing cost, latency, and performance)
- Generating and evaluating outputs for quality and safety
- Delivering results through effective UX/UI that supports user trust and feedback.
Throughout, teams must embed monitoring, evaluation, and risk management practices (addressing bias, compliance, and observability) to make sure reliability, scalability, and trustworthiness. In case your are constructing an AI Product, one other crucial a part of this phase is assessing the 4 big risks of product management: value, usability, feasibility and viability.
With all this in mind, we start further designing and planning for the project. On this step it is essential to grasp the best way to start small (Minimum Priceless Product) to expand and iterate once value has been proven.
For our marketplace example, let’s consider a feature that means an outline and category based on the product title. The flow could appear like this: a user inputs their item’s title, which is then used to construct a prompt for an LLM. The model returns a suggested description and category, that are pre-filled for the user to edit.
The feature’s risks may be broken down by category. For usability and value, the feature is familiar and editable, which is sweet for user experience. The essential risk is AI hallucinating or producing non-relevant suggestions, which should be measured with evaluations through the development phase. For feasibility, generating an outline and category with current LLM capabilities needs to be feasible. And for viability, this includes considering ethical risks, for example, generating biased or discriminatory suggestions (e.g. a cleansing product -> “perfect for ladies”), which also must be specifically evaluated during development.
For each potential risk you discover now, make sure you include it into your future evaluation plan.
⚡️ Accelerating the Phase with AI
- Tools for writing: Speed up the creation of your Product Requirements Document (PRD) with tools like ChatPRD. It’s also possible to improve the clarity and quality of your writing with assistants like Grammarly or Quillbot, and even get specific feedback in your writing with Quarkle.
- Tools for preparing presentations: Different tools like Gamma, Pitch, or beautiful.ai, are offering generation of slides from easy text and other documents.
- Tools for prototyping: AI can make it easier to create every thing from easy front-end mockups to complex, fully functional prototypes. Tools like Figma Make and Uizard are great for design-focused prototypes, and Claude artifacts can also be great to prototype UIs really fast. Platforms like Replit, Lovable, V0, Bolt can generate prototypes closer to totally functional MVPs, by generating the code full stack.
Phase 3: Development
That is where prompt engineering and trying out different models and approaches takes place, with the goal to get the very best possible outputs your use case needs. The bottom line is to arrange a request to an LLM with specific instructions, which can return generated text within the requested format (e.g. JSON object containing the suggested description and category).
A critical and sometimes missed a part of development is evaluation. You might want to make sure that the model’s predictions and generations meet a certain quality bar before they go live, and that the risks identified in phase two are mitigated. This involves defining use-case-specific evals to measure things like hallucination, correctness, bias, and task-specific performance. For a deep dive into this topic you may check my previous post:
For our marketplace example, we’d need to track the proportion of times the output is in the proper format, the accuracy of the category predictions, the relevancy of the generated descriptions, the proportion of times our outputs where biased or discriminatory…
⚡️ Accelerating the Phase with AI
- Tools for Coding: Many software development tools now include Generative AI features that help lower the entry barrier to coding. Assistants like Github Copilot, Cursor, Windsurf, or Claude Code are widely used to suggest code, complete functions, and solve coding problems. Using AI Chatbots can also be widely prolonged amongst programmers to speed up code implementations.
- Tools for Evaluation: LLMs are each used to generate input datasets when real data shouldn’t be available, and to design metrics that scale through the technique LLM as a judge.
Phase 4: Deployment
In our example, deploying the answer would allow that, when a user within the platform publishes a product, this triggers the decision to an LLM to acquire the outline and category from the title, and people are displayed of their corresponding touch-points.
Cloud providers like AWS, Azure, and Google have dedicated tools to speed up the means of integrating LLMs into your platform in a scalable way. On top of using these tools, you will want to care about service metrics like latency to make sure a great user experience.
A serious challenge with Generative AI is the “free input/free output” nature of the technology, which might introduce recent risks. For instance, users might by accident enter personal information and even attempt to attack your system through “prompt injection”. That is where guardrails are available. Guardrails are checks you set in place to make sure the robustness of your solution. They may be used to detect and block unwanted input, and to make sure outputs follow certain predefined rules like avoiding profanity or mentioning competitors.
Don’t just deploy the AI feature: your go-live plan isn’t complete without being ready for what can go improper in production and due to this fact ensuring observability (service performance, security, quality…).
⚡️ Accelerating the Phase with AI
- Tools for Guardrails: You possibly can implement safety checks using specific open-source libraries like Guardrails AI and LangChain, or use managed services from cloud providers like Microsoft Azure AI Content Safety. These tools, similarly to evals, repeatedly include LLM calls to automate the guardrail check.
Phase 5: Impact and Monitoring
This involves a mixture of:
- Service monitoring, where you employ tools like Datadog or specialized platforms like WhyLabs and Arize to trace the operational health and quality of your AI system in production.
- Quantitative data with analytics dashboards to measure the feature’s impact on key product metrics like user retention and engagement. For our marketplace example, you’d need to see if the brand new feature results in a decrease within the variety of users who abandon the listing process.
- Qualitative feedback from users to grasp further their experience and discover areas for improvement.
⚡️ Accelerating the Phase with AI
- Tools for Qualitative Evaluation: Many vendors that help collect user feedback, similar to Typeform and Canny, are actually incorporating AI features to robotically analyze and categorize responses. It’s also possible to leverage LLMs directly to research large volumes of qualitative feedback. As an alternative of manually reading 1000’s of comments, you should utilize an LLM to summarize themes, classify feedback by sentiment or topic (e.g. “inaccurate suggestion,” “UI feedback,” “latency issue”) , and discover emerging issues.
Wrapping it up
Developing an AI-powered solution is a journey from a user problem to a measurable impact. By moving through these five phases, you may navigate complexity and risks, while significantly improving the chances of constructing something of value.
In a meta twist, AI itself has grow to be your creative partner on this journey, able to make it easier to and your team brainstorm, code, and analyze feedback faster than ever before.
