# Minutes of Meeting — Enterprise Brain Platform Skeleton and Scope Control
**Date:** March 8, 2026  
**Source transcripts:** [Mar 8, 2026.txt](Mar 8, 2026.txt), [Mar 8, 2026-2.txt](Mar 8, 2026-2.txt)

**Attendees:** Yeshwanth Reddy Yerraguntla, Vara Kumar Jagarapu, Manisha Gundapuneedi, Gopal Gottumukkala, Rajashekar G, Satyasri Prabhakar Mantripragada

## Act I — 00:00 to 00:09, the first visible problems

The meeting opened with the practical issues still on the table: timezone chatter, flickering in the current UI, and the question of what can actually be shown in the experience center. Yeshwanth and Vara were already talking about what the user sees when the response starts streaming.

> *"Unless we go to switch to react completely we can't fix them."*
> — Yeshwanth Reddy Yerraguntla _(~00:05:35)_

That made the first part of the discussion a reality check. Some problems were architectural, some were just the current technology stack showing its limits.

## Act II — 00:09 to 00:15, the demo needs a real ask

The next chunk of the call was about whether the current demo is strong enough for the experience center pitch. Yeshwanth said the strongest thing they had was still a basic dynamic shell, and Satyasri pushed the team to ask leadership what they actually expect before spending more time polishing the wrong thing.

> *"What they are saying is we need a basic prototype to showcase experience center concept."*
> — Yeshwanth Reddy Yerraguntla _(~00:09:39)_

> *"Let us ask Rakkesh or PMA for what are their expectations from that experience center."*
> — Satyasri Prabhakar Mantripragada _(~00:46:54)_

That set a constraint for everything else in the meeting: do not overbuild a demo until the target behavior is agreed.

## Act III — 00:15 to 00:25, A2A and the task flow

Yeshwanth then moved the call into implementation shape. He explained the A2A model, the `message/send` and `task/get` flow, and how the agents should talk to each other without inventing a different transport pattern.

> *"You will use message/send to send a message, you will use task/get to get the response back."*
> — Yeshwanth Reddy Yerraguntla _(~00:38:00)_

> *"The way we have to communicate with agents is via two main methods."*
> — Yeshwanth Reddy Yerraguntla _(~00:36:46)_

The key point was that the user interface should wait for the result instead of trying to manually orchestrate a special case every time a request moves through the stack.

## Act IV — 00:25 to 00:35, repo skeleton and task splitting

The discussion then moved to how the work should be split for the team. Yeshwanth wanted a starter repo with a stable structure so people can branch from the same baseline rather than invent their own layouts.

> *"Every agent will have its own endpoint."*
> — Yeshwanth Reddy Yerraguntla _(~00:34:50)_

Gopal’s guidance was to pick one end-to-end flow and make it solid before stretching to the rest. That way, the team gets a believable backbone rather than a wide but shallow architecture.

> *"Let's pick only one."*
> — Gopal Gottumukkala _(~00:32:13)_

## Act V — 00:35 to 00:57, what counts as solid

The final stretch was about scope discipline. The current HTMX experience is flickering because the page refreshes too aggressively, so the team discussed a temporary bandage and the possibility of switching to React later.

> *"The whole page is refreshing for every word."*
> — Yeshwanth Reddy Yerraguntla _(~00:44:07)_

The team also agreed that once the platform skeleton is in place, the next cycle can focus on either the visualization layer or the alerting layer, depending on what gives the strongest proof of platform value.

## Act VI — 00:57 to 00:59, close on patience

The meeting closed on the same note it opened with: the team should keep correcting the shape as they learn, but not jump between contradictory approaches every day. The goal is a stable first cut, not an illusion of completeness.

## Todos

<todo>
  Create the shared agent skeleton repo structure and commit the standard `message/send` and `task/get` flow for each service.<br/>
  <span class="owner">Yeshwanth / platform team</span>
  <span class="deadline">Next build cycle</span>
</todo>

<todo>
  Implement one end-to-end ad hoc Q&A path that moves from UI to core agent to data agent and back.<br/>
  <span class="owner">Core architecture team</span>
  <span class="deadline">First implementation pass</span>
</todo>

<todo>
  Add telemetry, logging, and request idempotency to every agent before broadening the scope.<br/>
  <span class="owner">Platform engineering</span>
  <span class="deadline">Before the next demo</span>
</todo>

<todo>
  Decide which lifecycle comes next after ad hoc Q&A, with alerts and training as the leading candidates.<br/>
  <span class="owner">Gopal / Yeshwanth / Manisha</span>
  <span class="deadline">This week</span>
</todo>