@Gokul & @Nikhil prediction about Systems of Record vs. AI Agents is the one I keep coming back to. But I think there's a missing layer in that battle.
Systems of Record capture what happened. AI Agents execute what should happen. But neither captures why we decided.
That's the gap. When an LP asks "how did you make that call?" — Affinity can't answer it. ChatGPT can't answer it. The reasoning lives in IC discussions, Slack threads, and institutional memory that walks out the door when a senior partner retires.
Jeff Morris and Mercedes are right that memory becomes the moat. But for institutions, it's not just personal memory — it's decision memory. The audit trail of judgment that compounds over decades.
This is exactly what we're building at @Quantum Mosaic — decision infrastructure for institutional capital. We call it the "Judgment OS": encode how decisions get made, enforce consistency across teams, and evidence outcomes for LPs and regulators. The companies that own the decision graph — not just the relationship graph or the deal analytics — will own the infrastructure layer between SoRs and AI agents.
Foundation Capital published their "Context Graphs" thesis in December describing this exact gap. a16z's Big Ideas 2026 called for "systems of coordination" to manage multi-agent interactions. The category is emerging in real-time.
Excited to see how 2026 plays out — and to be building in the middle of it.
unlimited creativity by those who love to build and are curious about using this new ghost for their weird, niche interests. I for myself think about doing a tv series on napoleonic era just because it's so interesting. Also maybe building a mini neural net to see if it can guess european football club games. and personalized games for my nephews based on our family context (one goes to fishing, so a fishing game where he is the player). So much to do and so exciting at the same time.
They’re terrible at telling operators what to do next.
That gap is where AI agents actually matter.
For restaurants, they need agents that live inside the system of record, watches the data in real time, and behaves like an always on GM. It flags profit leaks, labor drift, and menu issues, then tells teams what to fix before service starts. Not weeks later in a P&L.
What’s been striking is how fast this shows up in practice.
Operators don’t want another dashboard.
They want an agent that understands their business, works inside their existing stack, and drives action.
The idea that systems of record will eventually bundle agents is right. But the winning agents will be built by, or deeply integrated with, the system of record from day one.
Otherwise you spend your time fighting access, fees, and latency instead of delivering outcomes.
Curious how others are seeing this play out in their industries.
VERY biased but my answer is the best one
@Gokul & @Nikhil prediction about Systems of Record vs. AI Agents is the one I keep coming back to. But I think there's a missing layer in that battle.
Systems of Record capture what happened. AI Agents execute what should happen. But neither captures why we decided.
That's the gap. When an LP asks "how did you make that call?" — Affinity can't answer it. ChatGPT can't answer it. The reasoning lives in IC discussions, Slack threads, and institutional memory that walks out the door when a senior partner retires.
Jeff Morris and Mercedes are right that memory becomes the moat. But for institutions, it's not just personal memory — it's decision memory. The audit trail of judgment that compounds over decades.
This is exactly what we're building at @Quantum Mosaic — decision infrastructure for institutional capital. We call it the "Judgment OS": encode how decisions get made, enforce consistency across teams, and evidence outcomes for LPs and regulators. The companies that own the decision graph — not just the relationship graph or the deal analytics — will own the infrastructure layer between SoRs and AI agents.
Foundation Capital published their "Context Graphs" thesis in December describing this exact gap. a16z's Big Ideas 2026 called for "systems of coordination" to manage multi-agent interactions. The category is emerging in real-time.
Excited to see how 2026 plays out — and to be building in the middle of it.
— Deepak Jha, Founder & CEO, Quantum Mosaic
unlimited creativity by those who love to build and are curious about using this new ghost for their weird, niche interests. I for myself think about doing a tv series on napoleonic era just because it's so interesting. Also maybe building a mini neural net to see if it can guess european football club games. and personalized games for my nephews based on our family context (one goes to fishing, so a fishing game where he is the player). So much to do and so exciting at the same time.
Regarding Systems of Records vs. AI Agents.
POS systems are classic systems of record.
They’re excellent at capturing data.
They’re terrible at telling operators what to do next.
That gap is where AI agents actually matter.
For restaurants, they need agents that live inside the system of record, watches the data in real time, and behaves like an always on GM. It flags profit leaks, labor drift, and menu issues, then tells teams what to fix before service starts. Not weeks later in a P&L.
What’s been striking is how fast this shows up in practice.
Operators don’t want another dashboard.
They want an agent that understands their business, works inside their existing stack, and drives action.
The idea that systems of record will eventually bundle agents is right. But the winning agents will be built by, or deeply integrated with, the system of record from day one.
Otherwise you spend your time fighting access, fees, and latency instead of delivering outcomes.
Curious how others are seeing this play out in their industries.
Looking forward to seeing what’s in store for AI hardware: https://www.focustools.xyz/p/my-2026-tech-predictions-a-year-of?r=frux
super informative. might tinker around w the ideas that i got from this piece. woud love for you to see it :)))
re: Gokul's SoR point in legal+health, it might be already here for legal. here is a chart: https://x.com/emrahyalaz/status/2008680619992379480