How I AI: Anjan Panneer Selvam and a new model of B2B product management

Anjan Panneer Selvam gets from a stakeholder's rough idea to an interactive prototype in under 30 minutes. The result is a concrete artifact that sales, product, engineering, and even customers can react to before the team commits to a single line of production code.

In this episode of How I AI, Anjan shows how his team at Acolyte Health captures an idea, turns it into a prototype that used to take months, researches the market, and keeps a library of demos for early customer conversations. The prototypes accelerate discussion and alignment; they do not replace validation or engineering estimates.

The three workflows he demonstrates create concrete artifacts early: an interactive flow, a research brief, and a shared prototype library. Each artifact gives the entire team something specific to question, debate, and revise.

From a meeting transcript to an interactive prototype

Anjan starts with a common scenario: the CEO has an idea for a new feature and wants to know how fast it can ship. Instead of writing a complete specification, he moves from that rough idea to a clickable prototype. The prototype helps expose missing requirements and disagreements about the intended behavior before weeks are spent on documentation and debate.

The process starts in the meeting itself. Anjan records the initial conversation with a Limitless pendant and uses the live transcript as the raw input for his first prompt.

Capturing the idea with a Limitless pendant

Anjan finds the pendant less intrusive than a phone or conventional meeting recorder, which helps keep the conversation natural. Of course, recording still requires appropriate notice and consent. Any transcript, especially one from a discussion that might include customer or health-related information, needs strict access, retention, and vendor-use controls.

Refining the prompt with ChatGPT

He takes the raw meeting transcript into ChatGPT to turn the conversational back-and-forth into a structured prompt for a prototyping tool. Anjan finds that the model is good at translating a PM's unique product sense into the specific, structured information that tools like Lovable or v0 need. A typical prompt might look something like this:

"A single page canvas, user facing tool, map out personalized workflows."

The prompt describes the user, the core workflow, in this case, a single-page canvas for mapping personalized user journeys, and the desired interaction. It provides enough specific language to guide the AI without attempting to specify every single screen or state.

Prototyping with Lovable

Anjan gives the refined prompt to Lovable to generate an interactive interface. He even added a suggestion to use a library called React Flow for the canvas builder and was surprised that Lovable understood the instruction and implemented the pattern correctly. This ability to build on existing libraries and patterns is a key part of the speed.

The generated prototype gives stakeholders an immediate interaction to review. The stakeholder-to-prototype workflow keeps the transcript, refined prompt, generated flow, and feedback loop connected while the artifact is still cheap to change. Stakeholders can disagree about the same visible behavior instead of separate interpretations.

A concrete artifact for everyone to debate

The example was an interactive drag-and-drop workflow builder for mapping a user journey, a feature Anjan says he has built at five startups over 17 years. What once took six months now takes under an hour.

With a shared prototype, customers and internal teams could react to the exact same flow and point to specific behavior that needed to change. This feedback sharpens the hypothesis, but it does not replace a sound research plan to prove demand or validate usability.

From prototype prompt to market-research brief

Once the prototype shows promise, Anjan uses Perplexity for an initial scan of competitors, market context, and key questions the team should investigate. The goal is to quickly build conviction that the idea is worth pursuing. The output is a research starting point, not evidence by itself.

He starts with the same refined prompt from the prototype workflow, reusing it as a structured description of the product, its users, and the core use case.

Using Perplexity for deep research

He gives Perplexity the product and customer context and asks for competitors, market signals, and sources. The rapid market-research workflow separates the first scan from the team's responsibility to open citations, prefer primary evidence, and verify every consequential claim.

Creating presentation-ready slides with Gamma

He then uses Gamma to turn the reviewed research into a draft stakeholder deck. The polished presentation format makes the material easier to discuss, but it is important to remember that a pretty slide does not increase the reliability of the underlying claims.

Playing devil's advocate with AI

Crucially, Anjan does not just ask the model to support an idea. He prompts it to play devil's advocate and critique the concept, asking if it is really worth building. This helps surface reasons not to build something, which he says has helped him say no to more features than yes. The team can then separate evidence, assumptions, risks, and unresolved questions before deciding whether to invest further.

Building a living library of prototypes

Anjan calls his shared collection of demos a "living product library." Sales and customer-success teams can use these interactive prototypes in early conversations, gathering feedback long before production code exists.

This only works if expectations are managed carefully. Every prototype needs a clear label and a script explaining to customers that it is exploratory, may contain simulated data, and is not a committed roadmap item or a functioning product.

Breaking an engineering deadlock with a mobile prototype

When a mobile idea deadlocked a team without mobile developers, Anjan used Rork to make the interaction concrete. The functional mobile-prototype workflow shows how to use that artifact to restart a feasibility discussion without presenting it as production proof.

The result was a functional prototype that could be tested on a phone, but it was not production-ready software. The goal was to prove what was possible and start a more informed conversation, not to prove that the underlying technical approach would work at scale.

The demo included selfie capture and facial-analysis behavior. Any prototypes involving biometric or health-adjacent data require extreme care. They should only use synthetic or explicitly authorized test data and need rigorous privacy, security, legal, fairness, and accessibility reviews before any real-user deployment.

A prototype is an alignment tool, not a product promise

Anjan's workflows use prototypes to make ideas visible and debatable much earlier in the process, connecting product, engineering, sales, and customer conversations around a shared artifact. This method works only if everyone understands the prototype's limits and the team remains committed to validating feasibility, value, and usability.

The fundamental operating change is a shift toward transparency. Instead of maintaining separate, abstract interpretations of an idea in documents and meetings, teams can now discuss the same interactive hypothesis.

Anjan's process is worth copying for any B2B team struggling with slow alignment between executives, product, and engineering. The living product library is a particularly powerful, if culturally challenging, idea for getting early feedback from sales and customers. The key is starting small: pick one disputed workflow, create a clearly labeled prototype with non-sensitive sample data, and see what the team learns. The AI makes the artifact cheap to produce, but human judgment is still required to manage expectations, interpret feedback, and decide whether the shiny new prototype is actually worth building.