If the morning keynote at Informatica World 2026 featured the Joshua Bell metaphor (same violinist, same Stradivarius, totally different reception depending on whether you’re in a Washington DC subway tunnel or the Carnegie Hall stage), this Integration Product Keynote later in the day made the same point a different way. The agents already work. The data underneath them is the difference between a busker and a sold-out concert.
Informatica, President of Product Management, Data and AI Engineering and IDMC Platforms, Pratik Parekh ran the Powering AI at Scale With Modern, Intelligent Data Integration session, and he opened with a framing that’s going to stick in any iTWire reader’s head for the rest of 2026.
You can watch the product keynote below, after which this article continues – please read on!
The intern with the perfect resume
Picture an intern who turns up on day one with the credentials, the theoretical knowledge, and the confidence of someone who’s read every textbook. Now picture them making decisions about your supply chain on hour two. That, in Parekh’s telling, is enterprise AI in 2026.
“It looks like an intern who is very smart, who is coming with the best credentials, who has studied most of the concepts and theoretical understanding of that particular subject he’s working on. However, he’s not fully familiar with an organisation taxonomy, does not know how decisions are made, who is who in the organisation on his first day of job. And now he’s trying to make some decisions. What is he going to do? He’s going to make some very big mistakes, very confidently,” said Parekh.
That paragraph belongs on a slide in every AI steering committee in Australia this month. Confident, credentialed, fluent in the consumer internet, clueless about your business.
Parekh cited the now-familiar MIT and Stanford numbers: 93% of organisations plan to deploy AI agents within the next couple of years, 80% of leaders are accelerating AI investment, and yet pilot-to-production failure rates are stubbornly high.
He says it’s the same line he heard around a customer dinner table the night before: “We like it, but I don’t think our AI is ready yet to make decisions for us.”
If the morning keynote with Salesforce President and CTO, Engineering and Analytical Engineering, Muralidhar Krishnaprasad (MK) said “enterprises just don’t have a data problem, they have a context problem”, this product keynote was the one that showed exactly how Informatica plans to fix it.
Context is the new ETL
Parekh’s definition of context is worth pinning to your monitor. Clean data brought together from every source and system. Semantics, definitions and relationships. Historical events that shape the data. Policy, compliance and access controls. And finally, activation, so the context actually does something.
He pointed at three published customer outcomes Informatica is happy to claim: Telus improving business outcomes by 50%, another customer banking 25 to 30% cost savings, and a full research paper documenting the transformation pattern. The implication for the room: this is not theoretical anymore, and your competitors are quietly cracking the code.
The unifying pitch is Informatica’s Intelligent Data Management Cloud (IDMC), now positioned as three pillars Informatica labels “Unlock”, “Trust” and “Activate”. Modern analytics still sits at the core. AI engineering layers on top. And the headless collaboration surface (announced on Day 1, demonstrated in depth in this product keynote) stretches across the lot.
Modern analytics, 500 features deeper
First up on the demo bench was Informatica’s data engineering product lead Aashoo Saxena, who walked through what’s changed in the modern analytics stack over the past year. Roughly 500 customer-driven features, by his count, with headline upgrades being native open table support for Apache Iceberg and Delta, separation of compute and storage, and data quality baked directly into the same agent interface as data integration.
“Cheap storage and the reliability of a data warehouse, those two put together, you are separating out the compute and the storage as you rightly said, no vendor lock-in, something Informatica truly believes in,” said Saxena.
The live demo was the Copilot inside the IDMC mapping designer. Saxena typed a natural-language prompt, “Extract data from EMP table to EMP using connection ID. W26SCE, type 2,” and the Copilot recognised it needed to build a Slowly Changing Dimension Type 2 mapping. It then auto-detected the join key across source and target, generated the pipeline, and handled change data capture along the way. Typos and all (Saxena’s words: real demo, real typos).
For anyone who tried Copilot six or twelve months ago and walked away unconvinced, Parekh’s pitch was to give it another go.
“This was a journey. There’s a lot of feedback that came in. We were not, you know, we were not ready in our first version, but we are iterating on it at an extreme pace. Probably releasing features, capabilities, enhancements every month,” Parekh added.
AI engineering: ETL plus plus
Then came the part of the keynote that maps directly to Day 1’s “trust by design” message from PepsiCo, Vice President, AI Architecture and Governance, Carla Eid (who at the time was busy talking about PepsiCo’s 1,500-plus production agents).
Informatica, Senior Director, Product Management, Etty Afriat, introduced a fictional 40-billion-dollar company called Nova Corp and asked the room a simple question: what about parental leave for European employees?
Their AI assistant, built on a textbook RAG architecture, confidently answered with the US policy (eight weeks). The room laughed. The Europeans in the room groaned. The correct answer in Europe is between 12 and 26 weeks.
“What about the eight weeks of paternity leave? That’s only for US, by the way,” joked Afriat.
The same question, the same data, the same RAG architecture, but with Informatica’s pipeline sitting between the raw documents and the vector database, returned the correct answer. The difference was not the model. It was the labelling, the document parsing, the sentence-aware chunking, the cleansing, the metadata enrichment, the policy tagging, and the option to bring your own embedding model or use Informatica’s.
“The data is important, and context is the most important thing. That’s what made a change. The problem never was the model. It was the context that we missed,” Afriat added.
That single line is the pull quote that should be on every CIO’s slide deck this quarter. It’s also the same conclusion Day 1’s main keynote landed on. The model is commoditised. The data plumbing is the moat.
The supporting product moves are exactly what you’d expect: unstructured data ingestion across PDFs, Word docs, PowerPoints and emails. Native vectorisation and embedding transformations in the mapping designer. Graph database support alongside vector stores. And an end-to-end control plane that traces an LLM answer back to the source document it pulled from.
Headless data engineering, demonstrated for real
The Day 1 keynote landed the headless announcement. The second keynote was where Informatica, Director, Product Management, Makesh Renganathan, showed what it actually does in anger.
Every IDMC capability (discovery, ingestion, replication, ETL, ELT, data quality, transformations, masking, governance) is now exposed via Model Context Protocol (MCP) and accessible from your developer surface of choice. VS Code. Cursor. Claude. Pick your weapon.
The demo built a customer lifetime value data product from a single natural-language prompt. The discovery agent identified the right sources from enterprise metadata. It generated a multi-step plan covering ingestion, medallion architecture, orchestration and testing. It flagged sensitive data and recommended masking rules. It validated the pipeline in real time, and packaged the whole thing for a pull request. Without ever leaving VS Code.
“We are not just writing code, we are architecting a system as we speak,” said Renganathan.
The kicker is round-tripping. Whatever a pro-code developer writes in VS Code shows up natively in the IDMC graphical interface for their low-code colleagues, and vice versa. Same metadata, same lineage, same data quality rules, same governance. The platform stops being opinionated about your favourite tool.
That’s the natural evolution of the Day 1 demo where Informatica engineer Shivangi Srivastava cut a revenue reconciliation pipeline from six months by hand to less than a day with Claude in VS Code. This keynote’s customer lifetime value demo did the same thing for a different use case, with the same speed lift, but added the round-trip into the graphical UI so a different developer can pick it up where you left off.
Schwab and Cox: the customer reality check
Parekh brought up two long-standing Informatica customers for a fireside chat, and the candour was a refreshing change from the usual conference customer parade.
First, Charles Schwab, Managing Director, Schwab Data, Kamal Khilnani, who runs an enormous data estate. 12 petabytes under management. 4 terabytes ingested daily. 20,000 jobs running nightly. All of it on IDMC and BigQuery, after a five-year migration off PowerCenter and a traditional warehouse, kicked off when Schwab acquired TD Ameritrade and projected a 4x workload increase.
“Now our focus is shifting towards unstructured data,” said Khilnani.
His three-phase AI strategy is the kind of structure most Australian financial services CDOs would do well to copy.
Phase one: passive assistants for productivity and efficiency, with a responsible AI framework underneath.
Phase two: orchestrated intelligence, with LLM-summarised call transcripts plugged into Salesforce CRM.
Phase three: agentic workflows, with humans still in the loop. Measurable wins so far include contact-centre average handle time down significantly, and the financial advisor research assistant cutting meeting prep from eight hours to two.
Then Cox, Vice President, Enterprise Data and Applications, Jake Edwards, who’s six months into his cloud migration off on-premises and into Snowflake and IDMC. Edwards was clear-eyed about what’s hard.
“I think building an agent is the easy part. I think how do you orchestrate it, manage it, monitor it is really the hard part. And that’s sort of that ecosystem that I think Informatica is bring to bear from that perspective,” said Edwards.
His Cox change management strategy is worth borrowing. Every employee has access to Claude. Lunch and learns. Get people comfortable with AI on personal writing and daily tasks before introducing it into data engineering pipelines. The headless tooling then stops feeling like a job-displacing threat and starts feeling like the obvious next step.
Edwards picked the headless announcement as his standout from the conference.
“If you guys know Claude and superpowers, if you use that skill very often, this is essentially data superpowers. It’ll bring more data practitioners to this space. So, classic software engineers love CLI, love IDEs like VS Code, and so it allows them to get into this space without having to go learn a GUI or UI,” Edwards added.
Khilnani’s closing line echoed the same conclusion Day 1 keynote customer PepsiCo’s Carla Eid landed on with “trust by design”:
“Normally the data strategy is an afterthought compared to the AI strategy. So I would say that making the data AI-ready is going to be very important. The foundation for data is going to be what’s going to fuel your AI strategy. So let’s focus on the data foundation first, making your data AI-ready,” Khilnani concluded.
If you take one line from this article into your next board meeting, take that one.
The Salesforce hug gets tighter
Informatica’s acquisition by Salesforce closed on 18 November 2025, after being announced on 29 May the same year. Six months in, Parekh framed the integration roadmap as “Better Together,” with Informatica providing the “Unlock” and “Trust” layers (data plumbing, quality, governance, lineage) and Salesforce Agentforce providing the activation layer (agents acting on harmonised, trusted data).
The concrete moves announced include an agentic harness that puts Agentforce directly on top of Informatica’s Data Management Platform. An end-to-end management layer that traces decisions from agent back to source data. A native high-performance connector for Salesforce Data Cloud (Data 360). And bidirectional integration between Informatica MDM and Data 360.
This pairs naturally with the Day 1 reveal of the Agent Fabric Context Catalog (MuleSoft + Informatica), which already lets you drill from an agent through the MCP server to the underlying API and the actual Snowflake record, with trust status and lineage visible end-to-end. Day 2 widened the same view into the data engineering side.
The pitch to data practitioners: the foundation you’ve been building for two decades is about to feed the entire CRM and CX stack. The pitch to Salesforce-first shops: there’s now an enterprise-grade data fabric underneath your agents, with the governance receipts to back it up.
Customer scoreboard, product keynote edition
The morning keynote had PepsiCo’s 1,500 production agents, Elevance Health’s “millions saved every month”, Paychex’s 512% ROI, ProBrands lifting sales by US$40 million, and Hearst’s security-first take on headless. This second keynote widened the roster.
Parekh ran through patterns from Gilead Sciences (enterprise data readiness for analytics), Subaru(manufacturing quality through connected trusted data), JetBlue (near real-time insights for airport and flight operations), Verisk (Parekh appeared to call them “Viscorn” from stage, but given the early-adopter agent engineering credit, it’s almost certainly Verisk), and Paychex (governed enterprise data for trusted AI).
Same pattern across all of them: data foundation first, agentic on top. Not the other way around.
What Australian iTWire readers should take away
A few things stuck with me after this session.
First, the agent platform conversation is increasingly a data plane conversation. Whether the front end is Claude, Cursor, VS Code, Codex, Slack or Agentforce, the differentiating story is real enterprise data, real governance, real lineage and real evals. The model is commoditised. The plumbing is the moat.
Second, the customer message has shifted from “buy this platform” to “fix your data foundation, and any platform will work.” That’s a more grown-up conversation than the one most CIOs were having 12 months ago, and it’s the right one to be having now.
Third, the pro-code/low-code unification through headless agents and round-tripping back into IDMC is a quiet but significant move. It says Informatica wants to be the data layer for every engineer in your shop, not just the dedicated ETL team. Australian organisations running mixed teams with senior software engineers and traditional data integration developers should take a close look.
Fourth, the partner symmetry is real. The same headless service that runs inside Claude or Cursor also runs inside Snowflake, Databricks, Microsoft Azure, Google Cloud and AWS, with Iceberg governance and zero-copy federation on top. For local customers stuck halfway through a multi-cloud migration, the news is that Informatica is no longer asking you to pick a side.
The intern grows up
Parekh’s intern analogy is going to age well. Enterprise AI today has the credentials and the confidence. What it’s missing is the company memory, the policy fluency, the people knowledge, and the lineage receipts. That’s the work data practitioners have been doing for 20 years. It turns out to be the work that finally turns AI from a confident liability into a trustworthy colleague.
The keynote closed with Parekh’s reminder that this audience, the data practitioners, the integration architects, the governance leads, are the change agents for the agentic enterprise. Not the data scientists. Not the LLM vendors. Not even the application teams.
“There’s nobody better than this group who can actually transform your organisation into an agentic enterprise. The data practitioner is the foundational role in shaping the future of your organisation,” said Parekh.
If you’re an iTWire reader sitting in a data role right now, you’ve just been told you’re the protagonist of the next decade. Act accordingly.
And if you need a single line to take into Monday’s standup: no context, no trust. No trust, no production. No production, no value. Fix the data, and the agents work. Skip the data, and you’re stuck busking in the subway with a Stradivarius and 52 dollars in tips.
Alex Zaharov-Reutt attended Informatica World 2026 in Las Vegas as a guest of Informatica.
