I spent last week at Salesforce Connections in Chicago. I came back with one nagging thought, and it wasn’t about the agents.
Everyone is racing to put AI agents on top of their data. Almost no one is asking whether the data underneath is ready to be consumed that way.
A few things from the floor. Data Cloud is back to being a “CDP” now rebranded Data 360, and it’s the mandatory foundation under Agentforce. Agents don’t work without it. The framing is explicit: your data is no longer rows in a table, it’s “live context” feeding agents in real time. And that combination, every sales lead, campaign engagement, and support case landing in one place, and agents that can reason and act on it directly, unlocks capability we’ve never seen before: systems that respond in the moment, instead of waiting on next week’s batch job.
The mechanism everyone’s excited about is Zero Copy, leave your data in Snowflake or BigQuery, query it on demand. It’s a genuinely smart idea, and like any architecture choice it moves the performance and cost question to the source layer, where fewer teams are looking. A couple of things worth planning for: you’re metered on both sides (Salesforce credits per million rows, plus your warehouse bill), and “acceleration,” while it does speed things up, still means copying data into Data 360, with the storage that implies. An agent reaching across that connector into millions of unoptimized rows, while the customer waits for their coffee, is a real performance question to design for.
Get that foundation right, though, and the payoff is real. And coincidentally, the best demo I saw featured a café. A cashier enters a loyalty number, and the agent instantly surfaces that this regular spends ~$109/month, has been a member three years, and is trying a mocha latte for the first time, then pushes a follow-up offer to their app. Genuinely slick.
But that magic depends entirely on having fast, clean, well-modeled data to answer at the register. Claude Code can scaffold the app in an afternoon. The agent, as well, including all its topics and actions. At run time the app calls the Data 360 API, and the agent grounds itself in whatever’s behind it. None of that can conjure up trustworthy, discoverable data that isn’t already there.
And here’s the part the demos can’t show you: they all run on curated data. What does the agent do when a 7-day gap appears? When the schema drifts? Even Salesforce’s own CTO is explicit about it: agents need metadata, semantics, types, and relationships, to reason with, not just raw rows.
So the real question I left Chicago with is this: how do I empower my agents through self-describing data? A term getting thrown around for this is AI Readiness, and it’s a higher bar than just having clean, recent, and optimized data. It’s whether the data is legible to machines, self-describing enough to navigate with no human in the loop. Can an agent understand what it’s looking at from table and field comments, data types, and documented structs? Better still, can it follow the crumbs you’ve left in lineage and metadata to discover related customer signals on its own, without anyone pointing the way? That’s the difference between an agent that retrieves and an agent you’ve actually empowered.
It’s what I’m spending my time on at eSage right now: getting data fit for an entirely new kind of consumer, one that can only reason, act, and differentiate when the data underneath it is AI Ready. And that’s where real velocity comes from.
eSage Group specializes in AI readiness. If you’re working on your AI plans and need support, let’s talk. Connect with our team today.
About eSage Group
- 5-star Customer Satisfaction Rating
- An amazing list of long-term, satisfied, Fortune 500 Clients
- Strong technical depth in all major Cloud platforms and Mar-tech tools to provide an Enterprise level integrated solution
- Fast Growing Managed Services Offering providing top-tier on-going support to our Clients
- Experienced team of:
- Data Engineers
- Data Scientists
- Project Managers
- Data Visualization Experts
- Artificial Intelligence Engineers