--- meeting: Enterprise Brain Scrum date: 2026-04-16 participants: - Rajashekar G - Rakkesh (UI/Design) - Sai Senthil - Vasa (Vara Kumar) - Navin - Dante duration: ~1:20:17 transcript: 20260416-enterprise-brain-scrum.txt --- # Enterprise Brain Scrum — April 16, 2026 A high-intensity scrum that cut straight to a live demo, then escalated into a strategic product direction session. Rajashekar used the Tata Steel UK NC conversion rate query as a stress-test, exposing the absence of confidence scores, the failure to distinguish fact from interpretation, and the team's drift toward shipping a thin Claude wrapper rather than a genuinely differentiated platform. The meeting established five non-negotiable mandates—confidence scoring, last-line-first answers, visual-first rendering, multi-lens perspectives, and intent-driven intelligence—and closed with concrete research and architecture assignments due before the next day's demo. --- ## Todos | # | Action | Owner | |---|--------|-------| | 1 | Modify the architecture to grasp Perspectives files provided by the team; agent should act accordingly | Rajashekar | | 2 | Add confidence/estimation scores to all responses (DB = 100%, internet = explicit %) | Rajashekar | | 3 | Remove standard industry term explanations for expert users; add beginner/expert persona config | Rajashekar | | 4 | Implement "last line first" pattern — one-liner conclusion upfront, full narrative below | Rajashekar | | 5 | Eliminate text–graph repetition; every KPI must be paired with a graph, never standalone text | Rajashekar | | 6 | Build persona/perspective files for Tata Steel UK (finance, PM, quality, risk, contractual lenses) | Sai Senthil | | 7 | Share Tata Steel data access via Vasa; confirm Sai Senthil has access before Saturday | Sai Senthil | | 8 | Research Tata Steel UK personas & benchmarks (internal / furnace-replacement peers / general steel market) | Sai Senthil | | 9 | Design visual-first UI pattern library with animations and micro-interactions | Rakkesh | | 10 | Build dynamic multi-perspective infrastructure and UI lens-switching control | Team | --- ## Act I – Live Demo, No Warm-Up (~00:00 – ~00:03) Rajashekar skipped pleasantries and fired the system with a real query mid-sentence: > *"Compare our NC conversion rates against industry benchmarks for UK steel profits."* > — Rajashekar G _(~00:00:00)_ The system returned a response that pulled data and surfaced some context about Tata Steel UK. The network was slow entering the metered connection, and the room waited. What came back was a lengthy text narrative—no confidence markers, no estimation signals, and no explicit delineation between raw data and model interpretation. The stage was set for everything that followed. --- ## Act II – Confidence Scores: Non-Negotiable (~00:03 – ~00:08) Rajashekar went straight to his primary objection: the absence of confidence and estimation numbers in all responses. > *"Where are your estimation and confidence numbers? So right now we are not sure. Why you not sure? That is important then. System is not whole if you don't give me confidence."* > — Rajashekar G _(~00:03:00)_ He outlined a clear tiering model: if a figure comes from the internal database it should carry 100% confidence. If it comes from an internet search it should carry a lower score—perhaps 50–80%—based on source reliability. The system currently makes no distinction. Everything looks equally authoritative, which means nothing can be trusted. > *"Whatever it's responding it's based on your true data. We are not going to intermittent searching for all the facts. I don't want that. System is not whole if you don't give me confidence."* > — Rajashekar G _(~00:05:26)_ This is not a polish item. It is foundational credibility. A CXO who cannot distinguish a database-backed number from a hallucinated benchmark will not stay on Enterprise Brain. --- ## Act III – The Product Thesis: Hidden Insights and Intent-Driven Intelligence (~00:08 – ~00:13) Rajashekar stepped back from the demo and articulated the core product thesis that he has been repeating for 18 months. The team nodded along, but the demo had just proved the gap was still wide open. > *"Regular data analyst will only give you regular insights. We will pull out the hidden insights from the data. And the hidden insights are always risky. They are questionable insights."* > — Rajashekar G _(~00:08:09)_ The second pillar he named is intent-driven intelligence. Enterprise Brain is not Q&A. When a CXO asks a narrow question, the system should infer what they *meant* to ask—not just what they typed—and surface the why, how, what, when, and where that underlies it. > *"You may ask one small question but we don't just stop at answering that question. We dig dig dig and give you as to why you would have asked this question. Not what you asked. We understand why you ask this question and try to answer the why, how, what, when, where and all of this—the five W's H."* > — Rajashekar G _(~00:08:09)_ He also introduced the **Challenger** concept: a companion agent that continuously cross-examines the main agent's conclusions. It functions as a built-in adversary—probing for gaps, surfacing counter-evidence, and offering the user a mechanism to invoke doubt constructively rather than passively accepting the output. This is distinct from the board-of-directors model of multiple lenses; the Challenger is explicitly adversarial. --- ## Act IV – The NC Language Problem: Fact vs Interpretation (~00:13 – ~00:18) The demo response had called Tata Steel UK's NC conversion rate "very strong." Rajashekar objected forcefully—not to the number, but to the editorial colour around it. > *"74% NCES are approved. That is a boolean variable. It is the truth. It is the fact. I can't question it. But anything else is a perspective."* > — Rajashekar G _(~01:41:37)_ The room debated whether to frame the number positively or negatively. Rajashekar's point was sharper than that: the system should not be making the framing decision at all. A CFO looking at a high NC conversion rate might see reckless scope-creep spend. An administrative manager might see efficient dispute resolution. The same fact looks opposite depending on the lens. > *"Pro or con—because if a CFO is looking at it, it might be a con, but if an administrative person is looking at it, it is a pro."* > — Rajashekar G _(~00:18:38)_ The engineering response was that the LLM can be instructed to present only facts without colour commentary. Rajashekar rejected that too: pure fact-dumping is not what he wants either. He wants *multiple lenses*, each making the editorial call appropriate to that perspective. Neutral silence is not a substitute for multi-perspectival analysis. A team member proposed showing confidence levels on specific claims: > *"When you look at broader UK market these internal figures put us in the leading position... what is your confidence level for this prediction? It can say: my confidence level is 60%."* > — _(~00:11:52)_ Rajashekar confirmed that is the right direction, and asked the team to wire that immediately. --- ## Act V – 360° View, Board of Directors, and the Challenger (~00:18 – ~00:26) Rajashekar articulated the architecture of forces he uses personally and wants Enterprise Brain to replicate. He described a board of eleven internal advisors—each embodying a different thinking framework—that he invokes in his own reasoning process. The system currently has no equivalent. > *"I have 11 forces working for me. Those 11 forces are something very different. Also from a business perspective: the positioner is saying something else is important, the challenger is saying something else is important. You guys are understanding it. And based on that I take my decisions. So it gives me a holistic view."* > — Rajashekar G _(~00:23:34)_ He then articulated why this matters commercially: Enterprise Brain cannot compete with Claude on raw capability. The only defensible differentiation is the human-intelligence layer—the curated perspectives, the frameworks layered on top, the institutional knowledge encoded through persona files, the challenger loop. > *"Don't throw me what the LLM is throwing. Then why would I come to Enterprise Brain? I'll use Claude. This is something I'm shouting and screaming for a long time now."* > — Rajashekar G _(~00:18:38)_ On the echo chamber risk: the system should not reinforce the user's existing worldview by defaulting to their own lens. If Rajat is the logged-in user, the *default* analysis should still surface perspectives from the finance head, the PM, the risk officer—not just "what Rajat would think." The system must proactively show views the user hasn't asked for, not just mirror back their preferred frame. --- ## Act VI – AI-First Design vs UX/UI Thinking, and the ODC Framework (~00:26 – ~00:32) A harder critique followed. Rajashekar challenged the entire design approach. > *"Don't still be in the UX and UI phase. What you are doing right now is UX and UI. It is not AI-first design. You're understanding the difference."* > — Rajashekar G _(~00:24:55)_ He introduced the **ODC framework** as a lens for product thinking: - **Solve** the problem as stated - **Resolve** the deeper friction behind the problem - **Dissolve** the problem entirely so it stops existing He used Amazon as the example: they first solved retail logistics, then resolved the friction of physical retail, then dissolved the problem by becoming infrastructure. iPhone dissolved the separate camera/phone/map/player problem class entirely. Enterprise Brain needs to aim at dissolution, not iteration. > *"Extraordinariness comes not by solving or resolving—it comes from dissolving."* > — Rajashekar G _(~00:28:28)_ The check he ran on Enterprise Brain at the current state: it is somewhere between solve and resolve. It does not yet dissolve anything. That is the target for the next layer of product investment. He also called for Enterprise Brain to carry **ten decision-making frameworks**—drawn from the body of work already defined by professors and strategists worldwide. The task is not to invent new frameworks but to map when each one should be invoked given the nature of the query and the persona of the user. --- ## Act VII – Tata Steel UK: Persona Research Assigned (~00:32 – ~00:42) With the strategic framing established, the conversation turned to the immediate Tata Steel UK engagement. The context: Rajit (next in line to Shankar) manages the capex project of replacing blast furnaces with electric arc furnaces—one of a very small number of such transitions globally. The system needs to be able to serve this use case with multiple comparison contexts: 1. **Internal**: within Tata Steel UK itself across projects 2. **Furnace-replacement peers**: other steel companies that have made the blast-to-electric transition (very few globally) 3. **General steel market**: all steel manufacturers regardless of furnace type Sai Senthil was assigned the research and persona-building work: > *"What you need to understand is in every environment—who are the different... for that problem statement, for that data statement—what are the different perspectives that are needed for somebody to hear? What is the legal perspective? What is the contractual perspective?"* > — Rajashekar G _(~00:26:06)_ Vasa was asked to point Sai Senthil to the existing Tata Steel data link so he has access to the source data before doing comparative analysis. The loop needs to close by Saturday. The deliverable: perspective files for each lens (finance, PM, quality, risk, contractual) that the system can consume and use to generate multi-perspective analysis automatically for Tata Steel UK queries. --- ## Act VIII – Visual-First Mandate and Last-Line-First Pattern (~00:42 – ~01:07) Rajashekar spent significant time on the UI problem. His frustration: the system returns walls of text that explain things the domain expert already knows. Two mandates came out of this block. **Last line first**: Every response must open with a one-line conclusion. The full narrative may follow. But a user who is impatient—whose temperament that day means they have two minutes—must be able to read the top line and get the answer. > *"Say your last line first. There are different kinds of temperamental people. There are situations where I'm becoming restless. There will be situations where I'm extremely patient. Knowing that—every person—we have to first give a last line... followed by: here is your narration story."* > — Rajashekar G _(~01:09:11)_ A team member raised a conflict: the system was instructed yesterday to build up to the answer in a storytelling fashion. Rajashekar was clear the two modes are not mutually exclusive: the one-liner is always present at the top, and the full narrative follows. The one-liner is the headline; the story is the article. **Visual-first, no text repetition**: If a number appears in a chart it must not also appear in the key-highlights section as text. If a visualization conveys the insight, the text summary must strip the number and carry only editorial context. > *"I don't want this this this this this. It is repetitive information. You're giving me numbers, then you're showing me a graph, then you're putting me this line. Why? All of this could have been in one line."* > — Rajashekar G _(~01:11:12)_ He also set a non-negotiable visual standard: every KPI must be paired with a graph. Standalone text KPI indicators are not acceptable. Micro-interactions, animations, and high-quality rendering are mandatory—not polish. He directed the team to step outside regular tooling patterns and draw inspiration from gaming environments and 3D visualization spaces. > *"Visual-first driven. Either that visual—if it is graphs it has to have beautiful animations, micro interactions, whatever all those gimmicks you guys do—either that, or take me as close to realistic objects as possible."* > — Rajashekar G _(~01:14:28)_ --- ## Act IX – Accountability and Zero-Progress Confrontation (~01:03 – ~01:17) The session's temperature peaked here. Rajashekar named directly that in his assessment there has been zero visible progress from what was shown the previous week. > *"In your eyes, it might feel like progress, but in my eyes, truth be told, there's zero progress from what you showed me last week."* > — Rajashekar G _(~01:17:38)_ He escalated to the structural diagnosis: team members are operating in silos, treating each product direction as an isolated task, and not synthesizing requirements into a coherent vision. The result is incremental improvements that do not accumulate into differentiation. He also called out a recurring communication failure in the room: the team signals understanding in meetings but returns having implemented something different. His explicit ask going forward—when he finishes making a point, someone in the room should play it back before the meeting moves on. > *"What is the solution? Because for me in these meetings—I took out one hour or one and a half hour of my time to just reiterate certain things. I need all of us also take accountability."* > — Rajashekar G _(~01:05:43)_ The immediate confrontation was over the "key highlights" pattern: the agent generated a text block with three metrics, then rendered the same three metrics in a chart below it, then added a narrative that repeated the same numbers again. All three outputs were present simultaneously. Rajashekar asked the team to define a UI pattern library that prevents this—a system of templates where the agent picks the correct visual template and is prohibited from repeating data across modes. --- ## Act X – Closing Assignments (~01:17 – ~01:20) Rajashekar closed with targeted direction. Rakkesh was assigned a solo architecture session: identify the decision-making frameworks and personification models that should be encoded into Enterprise Brain, then propose how the architecture ingests perspective files and routes queries through them. If he wants to walk through the frameworks with Rajashekar afterwards, that offer is open. The goal is that by tonight, there is a design—not code—for how the system will dynamically generate multiple perspective analyses and offer the user lens-switching without overwhelming them. > *"Whatever you have, give it off to Yashan and them. And if you have to build some architecture—I don't know whether you have to or not—don't wait till tomorrow morning. You go and figure out your engineering and AI side of it. Me and Dante, we puzzle pieces, we will come and fill the puzzle pieces."* > — Rajashekar G _(~01:17:38)_ Tomorrow's demo is the hard deadline. All deliverables above are aligned to that gate.