Attio’s just launched the idea of “Universal Context”, and it’s exactly what I’ve believed, for a while, is the next step in CRM. Here I walk through what exactly this is, why it matters, and how it fits into the next few years of CRM releases and evolution.
A quiet shift: from systems of record to systems of meaning
CRM has always been sold as a system of record. A place where facts about relationships get written down: who said what, what stage the deal is in, when to follow up.
But “record” is a strangely limiting metaphor.
When a human seller opens a CRM, they are not looking for records. They are looking for orientation. They want to feel, in a few seconds, what matters right now. What changed. What’s implied. What’s risky. What’s still unsaid.
For years, we’ve tried to manufacture that feeling through dashboards, required fields, and process discipline. And it mostly worked, as long as the “reader” was human and willing to do the interpretive labor.
AI agents change that premise.
An agent can draft an email, summarize a call, decide which account to focus on, and update the pipeline. But only if it can trust the world it is operating in. Which raises a question that feels philosophical, but is actually architectural: what does it mean for a CRM to be true?
Agents don’t just query data. They live inside it.
When people imagine “AI in CRM,” they often picture a chat box that answers questions.
That’s the shallow version.
The deeper version is agentic workflow: an AI that continuously reads context, takes actions, and leaves the system in a better state than it found it. It triages inbound leads. It notices a deal is at risk and pulls forward a champion call. It asks for missing information at the moment it matters. It opens a task when a key stakeholder changes roles. It follows up with the right tone because it remembers the last three conversations.
This is not a UI feature. It is a new type of participant.
And participants have a basic requirement: consistency.
Humans tolerate inconsistency because we compensate. If the “next step” field is stale, we infer the truth from Slack, email, and the faint memory of last Thursday.
Agents cannot improvise like that. They will do exactly what the system implies. If the CRM says a deal is “Contract Sent” but the last call transcript shows the buyer is still debating budget, which truth wins?
That question becomes urgent once you accept that the agent is not simply answering queries. It is making decisions.
The legacy trap: bolting meaning onto a record
Most legacy CRMs were designed in an era where meaning was computed by humans.
Their core abstraction is straightforward: objects, fields, relationships, permissions, and some workflow automation. You can build an enormous business on that. You can scale it across thousands of users. You can make it extremely configurable.
But when you retrofit AI, you tend to do it the way the broader software world has done it: add a semantic layer.
A vector index here. A “knowledge base” there. A set of embeddings generated from notes, emails, and call transcripts. Maybe a semantic cache. Maybe multiple.
All of those are useful.
They also introduce a new failure mode: drift.
Once you separate “the system of truth” (your structured CRM data) from “the system of meaning” (your semantic index), you’ve created a synchronization problem. If embeddings are generated from an earlier snapshot, your agent may confidently retrieve context that no longer matches the real record.
And the more agentic the system becomes, the more dangerous drift becomes.
A human might read an outdated note and shrug.
An agent might act on it.
So the real critique of bolt-on AI in CRM is not that it’s unhelpful. It’s that it creates two realities: one for reporting, and one for reasoning.
And as soon as you have two realities, you’ve stopped building a CRM and started building a negotiation between data stores.
Universal Context as a data-model idea, not a feature
Attio’s framing of “Universal Context” is interesting because it treats this as a data-model problem, not an AI feature problem.
The simplest way to describe it is this: semantic representations should not be a separate truth. They should be externally consistent with the underlying CRM data.
External consistency is an unusually precise phrase for a CRM launch. It suggests a promise: whatever “meaning” the system derives (embeddings, summaries, retrieval context) is always in sync with what the CRM actually contains right now.
Not eventually.
Not “close enough.”
Perfectly.
Why does that matter? Because it changes what you can safely delegate.
If an agent is going to operate continuously, you want it to be grounded in the same source of truth your humans use. Otherwise, you end up managing the agent the way you manage a new hire with unreliable notes: you double-check everything, you hesitate to delegate, you keep the work small.
Universal Context is a bet that the future CRM is not a database with a chat interface. It’s a unified model where structured fields and unstructured artifacts belong to the same reality.
And once you say that out loud, the implication is hard to ignore: perhaps the “future of CRM” is simply the future of data modeling.
The substrate: why graph-relational starts to feel inevitable
Traditional CRMs are often either relational at heart or they approximate it: tables, joins, and rigid schema constraints.
That works well for known, structured facts.
But sales and GTM are not only structured facts.
They are narratives.
They are changing stakeholders, informal influence paths, shifting priorities, and soft signals that appear in calls and emails. Those signals matter. They are also unstructured.
Attio’s underlying Particle model is positioned as graph-relational: it can represent entities and relationships, while also supporting the kind of querying you want from a CRM. That matters because Universal Context depends on being able to treat “notes,” “recordings,” and “messages” as first-class citizens, not as attachments living off to the side.
If you want a single system that can answer:
- “Which deals are most likely to slip?”
- “What changed since last week?”
- “What is the real objection pattern across the enterprise segment?”
You need to unify how the system queries structured and unstructured data.
In other words, you need the CRM to behave less like a set of forms and more like a knowledge graph that happens to be operational. CRM data is broader than the fields we’ve historically forced humans to fill in.
Universal Context is the next step in that arc: not just storing more kinds of information, but making sure the “meaning layer” remains consistent with the operational layer.
Ask Attio is the interface. Universal Context is the contract.
It’s easy to fixate on the conversational layer.
“Ask Attio” is obviously the kind of feature that demos well. People can feel the value instantly: you ask a question, it answers.
But the interface is not the core innovation.
The core innovation is the contract underneath it.
If a feature is powered by Universal Context, then the question is no longer “can the model answer?” The question becomes: “can the model answer in a way that is operationally safe?”
That’s a different standard.
A generic LLM can answer many questions about your business. It can also hallucinate. It can retrieve outdated context. It can infer the wrong thing from the right data.
A CRM-grade agent needs stronger guarantees.
Universal Context is a move toward those guarantees. It’s a structural claim that the CRM can be queried semantically without creating a shadow copy of the truth.
That may sound like an implementation detail.
But implementation details are where strategy hides.
The cross-platform reality: context should travel, not your exports
There’s another implication in this new direction that’s easy to miss if you only think about CRM as an app.
If agents increasingly live in multiple surfaces, ChatGPT, Claude, internal copilots, then “where the agent runs” becomes less important than “what the agent can reliably access.”
This is where things like MCP servers become strategically important.
The promise is not just that a CRM has an AI feature. It’s that different agents, across different environments, can access the same real-time context and take action on the same source of truth.
That’s the difference between:
- AI as a front-end assistant that answers questions inside one product
- AI as an ecosystem of workers that coordinate around one operational data model
If you’ve ever watched GTM teams fragment across tools, spreadsheets for forecasting, Slack for decision-making, email for relationship management, you can see why this matters.
The future CRM may be less like “the place salespeople log activity” and more like “the place agents synchronize reality.”
And once that’s true, consistency is no longer a nice-to-have. It’s the foundation.
Extensibility is now an AI design decision
Most CRM conversations about extensibility are about developer experience, enterprise requirements, and integration depth.
Those still matter.
But agentic software changes why it matters.
If AI agents will increasingly write and maintain software, then the extension layer of a CRM becomes a compilation target.
That sounds abstract, but it’s practical:
- What language is the platform built around?
- How much of the world’s code (and the world’s training data) looks like that language?
- Can an agent reason about it, generate it, refactor it, test it?
This is where the contrast between Salesforce Apex and Attio’s App SDK becomes more than preference.
Apex, introduced in 2007, is proprietary and Java-like. It has its own ecosystem, its own quirks, and its own institutional gravity. It was designed for a different era of development.
Attio’s SDK approach, centered on TypeScript and React, is aligned with what modern developers already use, and with what modern coding agents have been trained on.
So the question becomes: when your “developer” is increasingly a model, what do you want the platform to look like?
Do you want it to be fluent in the dominant languages of the internet? Or do you want it to require translation into a proprietary dialect?
That is not just a tooling question. It’s a question about how quickly your organization can turn intent into software.
Generative application logic: the end of “CRM customization projects”
There’s a forward-looking claim embedded in this whole direction: that users will request capabilities and receive custom apps without seeing code.
It’s worth pausing on that.
Because today, CRM customization is often a slow ritual:
- requirements gathering
- schema changes
- workflow design
- stakeholder alignment
- implementation
- training
The result is usually a compromise between what the business wants and what the CRM can reasonably accommodate.
Generative application logic proposes a different dynamic.
A RevOps leader asks: “Can we track partners like accounts, but with different stages and different health signals?”
A system responds: “Yes,” and produces the object model, views, workflows, and permissions.
Not as a one-off consulting engagement.
As a normal interaction.
If you believe that world is coming, then Universal Context becomes even more important. Because the more dynamically your system evolves, the more dangerous it is to have a semantic layer drifting away from the operational truth.
In a world of constantly generated workflows, you need the meaning layer to update as fast as the logic.
Otherwise, you’ve built a machine that can change itself, but cannot consistently understand itself.
So what is the future CRM, really?
It might be tempting to summarize all of this as “AI-native CRM.”
But that phrase is starting to blur. Everyone will use it.
A more precise framing is this: the future CRM is a data model designed for delegation.
Delegation to humans, through views and workflows.
Delegation to agents, through consistent context.
A CRM built for delegation must do a few things exceptionally well:
- treat unstructured artifacts as first-class data
- unify retrieval across fields and narrative
- ensure semantic understanding is externally consistent with the operational record
- expose the same source of truth across multiple agent surfaces
- offer an extension layer that is legible to modern developers and modern coding agents
Universal Context is a statement that the heart of CRM is no longer forms and fields. It’s context and consistency.
And Ask Attio is a statement that the primary UI may become language.
Put those together with MCP tooling, and you get a coherent picture: not a retrofitted CRM with an AI assistant, but infrastructure for agentic GTM.
The most interesting part is what it implies about competition.
The next era of CRM might not be won by whoever has the biggest app marketplace or the most dashboards.
It might be won by whoever can make “meaning” as reliable as “record.”
Which leaves us with a closing question that feels almost too simple:
When an agent asks your CRM what is true, will it get one answer, or two?
If you believe agents will become real co-workers, not just features, you already know which future you want.
For more on Attio’s framing of Universal Context and Ask Attio, see the launch post.