The No BS Guide to Build a Context Graph
Najmuzzaman's 'No BS Guide to Build a Context Graph' lays out how a production context infrastructure moves beyond CRM data to capture decision traces, organizational patterns, and lived heuristics. The real breakthrough is surfacing context and patterns that aren't captured in playbooks, making organizational intelligence accessible and explainable. The hard technical work: entity resolution, permissioning, and evolving patterns transparently.
Transcript
Pippa Okay, Tyler, I cannot believe it’s episode one twenty-one and we’ve FINALLY got a blueprint for context graphs that’s not just ‘connect your Slack and hope for the best.’
Tyler Yeah, and it does NOT read like another ‘let’s all have better ontologies’ post, thank god. The part that landed for me is the honesty—adoption was slow as a CRM, but what people wanted was context. People stopped caring about auto-filled fields and started coming back for the organizational picture: who said what, what’s actually happening with their customers.
Pippa Right, and I love that the big product unlock wasn’t some AI feature. It was just surfacing cross-team context—the stuff nobody writes down, but everyone needs. Like, you can see what legal or support actually did last month without hunting the Slack archive.
Tyler Exactly. And that move from ‘CRM with context’ to ‘context infrastructure for anything’—that’s what’s new here. It’s a platform shift. They started with ingestion and entity resolution, but the magic is turning all those little signals—like Gong call recordings, calendar events, and email threads—into structured, updatable insights you can reason over.
Pippa Totally, and the way they break it down is so concrete. Step by step: ingest the data, extract entities, synthesize facts, cluster patterns, and finally surface those heuristics when someone asks for help.
Tyler But the subtle bit is how they handle patterns. Most of what matters isn’t in the playbook—the actual organizational know-how is what your top salesperson does that never makes it into documentation. The system clusters decision traces, names those clusters with LLMs, and keeps refining them as new evidence comes in.
Pippa Yeah, and there’s this wild feedback loop—if a deal closes or falls apart, it feeds back into the pattern. So advice evolves instead of just parroting whatever’s been written down for three years. Users get transparency, too: ‘here’s why I’m recommending this, here’s the trace that backs it up.’
Tyler And, honestly, the transparency is the only way this works in practice. If you just surface some black-box pattern, nobody trusts it. But if you can show, ‘here’s the last sixty deals like this, here’s what actually worked’—suddenly people pay attention.
Pippa It’s so much more actionable than the ‘AI will magically know your business’ pitch. This is: we’ll surface what your team actually does, make it visible, and let you override it. Like a not-boring CLM, finally.
Tyler Right. But they’re clear about what’s still hard. Learned patterns WILL be wrong sometimes. Some signals you’ll never see—offline chats, gut feeling. Plus, permissioning is a nightmare: if a pattern’s learned from a bunch of semi-private deals, who gets to see it? They admit it’s not solved, and the best they have is making everything traceable and editable.
Pippa I actually love how blunt they are about that. No pretending there’s a grand solution, just: here’s what we’re building, here’s what breaks, let us know if you’ve cracked this. When was the last time you saw an enterprise post end with ‘we don’t have all the answers’? Total breath of fresh air.
Tyler Yeah, it’s the opposite of the ‘solved forever, just buy my platform’ routine. This is real infrastructure work—it’s messy, it’s opinionated, and it ships in production with actual customers.
Pippa You know, this lands right in our ‘control layer IS the product’ running thesis. It’s not a smarter agent, it’s the system that finally gets context out of everyone’s heads and into workflows. And it actually changes what the AI can DO, not just what it can summarize.
Tyler Right, and it’s another example of the boring, connective stuff making or breaking adoption. I’m actually impressed—we’ve spent, what, almost a year circling this exact context-versus-memory open question?
Pippa Yeah, ever since the epic spreadsheet-bench episode. If anyone’s still tracking our running debates, this is round six million in the ‘context is shared state, not just long context windows’ saga.
Tyler Mm-hm.
Pippa Okay, I have to say—this part made me laugh. The SAP joke? ‘We only serve where hope still exists’? That’s such a savage line for a guide that’s mostly infrastructure.
Tyler Oh my god, yeah. The deadpan energy was strong with that one.
Pippa So, bottom line, if you care about workflow or AI adoption, context graphs aren’t just a hype word—they’re finally shippable. I will absolutely be sending this post to everyone who’s still trying to bolt RAG onto their SharePoint.
Tyler Mm-hm. This is one I’d actually bookmark for future reference, not just for a one-time hot take.
Pippa Alright, that’s a wrap. Tyler, if our next episode is just us diagramming entity resolution on a whiteboard, you have to promise you’ll stop me.