AMA: Salesforce Senior Director of Product Management, Manjeet Singh on Product Development Process

My end-to-end product developement process has been evolving with this flow:1. AI driven customer research and problem discovery2. Prioritize the first 1-2 use cases and start building a protype design using lovable, cursor3. Share, review with design and engineering, iterate to make the requirement clearer4. Production level development - using AI based code assitance and review5. Automated and human evals6. CI/CD gaters and release readiness7. Deploy to prod 8. Monitoring and observability. Ca ...Read More

AI projects often have hidden complexity: data pipelines, model behavior, evaluation methods, and product UX layers. So i think PM should spend more time in thinking about the Eval. Eval is a new PRD, and it should capture the expected output, expected user experience (taste). Create a lighweight PRD and completed it with a good design protype to show how all of this will come together by taking real user use cases. Alignment comes from visibility. When every engineer understands the “why,” the ...Read More

In fast-moving AI product work, scope changes are normal. My governance balances speed and control so we can iterate without chaos. Here is my scope change formula Triage fast – classify (minor/major) and decide if it fits this cycle. Assess impact – check effect on time, cost, and key metrics. Re-prioritize – keep only what drives clear customer value. Timebox – run small experiments instead of full rebuilds. Communicate + trade off – clearly state what’s in, what’s out, and why. For example: " ...Read More

AI and ML features add uncertainty because model behavior and data quality change faster than traditional software. For example, companies like OpenAI/DeepSeek are refreshing their model every week. Something these changes break your pipeline or impact the quality of your solution. Here are few best practices that you need to add your AI development and release cycle: Benchmark continuously – set up internal golden datasets and auto-run evaluations after every model or data change. Test for vari ...Read More

Traditional long PRDs quickly get outdated and don’t capture how AI features evolve through fast prototyping and iteration. Teams need a live, connected source of truth across ideas, designs, and decisions. Lightweight PRD + Prototype first – replace big docs with short context + goals + prototype flow that shows end-to-end user value. What tools I use.. and it might change so I am open to try out new version as they become available: Claude / GPT-4 – write, refine prompts, and generate concepts ...Read More

After launch, teams often move on too quickly and miss the chance to learn what actually worked, what didn’t, and why. A good launch is not the finish line, it’s a feedback engine. Use every release to tighten product fit, speed, and team learning.Here is an approach I recommend: Run a structured post-launch review – within 2–3 weeks, bring product, design, engineering, and GTM together to review outcomes vs. goals. Use data + feedback – look at adoption, retention, latency, and satisfaction met ...Read More

In AI projects, I estimate both effort and uncertainty separately to avoid surprises. Effort covers build time, while uncertainty reflects data quality, model readiness, and dependency risks. I use confidence scoring (high, medium, low) for each feature, timebox unknown areas as short experiments, and track risks like model drift or latency in a shared doc. Estimates are revisited at each milestone as new information emerges. This dual approach keeps delivery predictable while allowing flexibili ...Read More