Localized Orchestrators Will Lock You Out of Enterprise AI

One of the fastest-growing data platform vendors recently announced (again) that customers inside their ecosystem are migrating from Apache Airflow to their proprietary orchestrator. They made the same announcement three years ago, back when their product had a different name.

It’s an argument we’ve all heard before, not just from this vendor, but from every platform that’s ever tried to convince enterprises that data and workflow orchestration belongs inside their walls.

You’ve heard the lock-in arguments. You know they’re right. And yet teams keep making the same choice anyway because it solves the immediate problem, looks good on a performance review, and by the time the bill comes due, it’s someone else’s mess to clean up.

We’ll get back to lock-in. But there’s a newer, more urgent reason why localized orchestrators are a strategic mistake. One that has nothing to do with exit costs or vendor leverage and everything to do with what AI actually needs to work.

The AI Plot Twist

The rise of AI makes platform-neutral orchestration more important, not less. But not for the reasons you might think.

Yes, every major player is racing to ship AI capabilities. AWS has Bedrock. Azure has OpenAI integration. Google has Vertex AI. Databricks has model serving. Snowflake is building AI features. New startups are launching weekly. The models themselves are changing faster than anyone can keep up. Locking your orchestration into any single platform means betting that platform will remain the right choice through all of it.

But there’s a more fundamental issue emerging. What’s becoming consensus is that AI’s real value won’t come from models alone. It will require cross-system context engineering: giving models access to the decision traces, exceptions, approvals, and reasoning that currently live in Slack threads, escalation calls, and people’s heads.

“Context graphs” have emerged as one promising approach, capturing not just the results of actions but why those actions were taken. Jaya Gupta (@JayaGup10) at Foundation Capital recently argued that these accumulated decision traces will become “the single most valuable asset for companies in the era of AI.”

Here’s what matters for orchestration: context graphs can only be built by systems that sit in the cross-system execution path.

“The orchestration layer sees the full picture: what inputs were gathered, what policies applied, what exceptions were granted, and why. Because it’s executing the workflow, it can capture that context at decision time, not after the fact via ETL, but in the moment, as a first-class record.”
Jaya Gupta, Foundation Capital

This is where localized orchestrators hit a wall. By definition, they only see within their walled garden.

Execution paths are local. Context is global.

Consider what happens when a renewal agent proposes a 20% discount. It doesn’t just pull from the CRM. It pulls from:

• PagerDuty for incident history
• Zendesk for escalation threads
• Slack for the VP approval from last quarter
• Salesforce for the deal record
• Snowflake for usage data
• The semantic layer for the definition of “healthy customer”

A single decision. Six different systems. Every enterprise has different combinations of these systems. One department runs Salesforce + Zendesk + Snowflake. Another runs HubSpot + Intercom + Databricks. A third runs a homegrown CRM + ServiceNow + BigQuery.

Prukalpa (@prukalpa), CEO at Atlan, made this point in her recent response to the context graph thesis: “The vertical agent sees the execution path and captures context within their workflow. But enterprises have dozens of agents, across dozens of vendors, each building their own context silo.”

Her broader argument is that vertical agents (and by extension, platform-native tools) can’t solve the heterogeneity problem. The winner, she argues, will be whoever builds the universal context layer that stitches together context across systems.

I’d take this slightly further. She’s right that “in a world of heterogeneity, the integrator always wins.” But I’d argue the orchestration layer is that integrator for operational context. It’s the only layer that sits in the execution path across all systems, at the moment decisions are made. The orchestrator doesn’t just move data. It witnesses the full decision trace.

Heterogeneity is moving up the stack

Prukalpa makes another observation worth highlighting. For the last decade, “heterogeneity” in data meant a mess of point tools and closed warehouses. Iceberg and open table formats are beginning to change that, making storage more open and compute more fungible.

But fragmentation isn’t going away. It’s moving up the stack. Instead of five warehouses, enterprises are moving toward hundreds of agents, copilots, and AI applications. Each with its own partial view of the world, its own embedded definitions, its own “private” context window.

This is where cross-system orchestration becomes essential. It’s no longer just about moving data. It’s becoming a prerequisite for whatever context engineering approach wins out.

Choose a localized orchestrator and you’re not just locking in your workflows. You’re locking yourself out of building the organizational intelligence layer that will define competitive advantage in the AI era.

Oh, and About Lock-In

You’ve heard this part before. In one shape or another, it’s been true for three decades. Consider this your periodic reminder.

Control-M was the enterprise scheduler for mainframes. Powerful and deeply embedded, until enterprises modernized and spent years ripping it out.

Apache Oozie was Hadoop’s native workflow scheduler. Tightly integrated, “native,” and a dead end the moment organizations moved beyond Hadoop.

Informatica Workflow worked beautifully if you stayed within Informatica. Now Salesforce is acquiring them for $8 billion, and if history is any guide, interoperability with external systems will get harder, not easier.

The pattern is always the same: platform creates native orchestrator, it works great inside their ecosystem, innovation happens outside the walled garden, companies find themselves stuck.

Every few years, we collectively forget that lock-in is bad. A new technology wave arrives, and vendors convince us that this time it will be different.

But it’s never different because business incentives are as reliable as physics.

Your orchestrator is the connective tissue of your data operations, which feeds analytics, machine learning, and your future AI initiatives. It’s the system that knows how all your other systems work together. Lock that into a single platform, and you’ve handed that platform enormous leverage over your entire data strategy because orchestration is both the on-ramp and the exit. Control the orchestrator, and you control whether data flows freely across an organization’s stack or stays trapped in one vendor’s ecosystem.

The Bottom Line

When a platform vendor says “customers are migrating from Airflow to our native orchestrator,” they’re describing customers who’ve decided to go all-in on that platform, trading flexibility for simplicity.

The traditional objections still apply: you’re betting your data future on one vendor, making a one-way door decision, and trusting that this platform will remain the right choice for years to come. Those arguments are as true as they’ve ever been.

But the AI era introduces a new cost that’s harder to see and even harder to recover from.

The context AI needs to make good decisions doesn’t live in one platform. It doesn’t live in verticalized workflows. It’s scattered across your CRM, your ticketing system, your communication tools, your data warehouse, and a dozen other systems. The only way to capture that context is to orchestrate across all of them. A localized orchestrator, by definition, can only see fragments.

This isn’t about flexibility anymore. It’s about whether you’ll be able to build the organizational intelligence layer that AI requires. The companies that get this right will have AI systems that learn from precedent, understand exceptions, and compound institutional knowledge over time. The companies that don’t will have AI systems that start from scratch on every decision, forever locked out of the context that would make them useful.

Orchestration is not a feature. It’s infrastructure for all eras.

Like your network. Like your identity provider. Like your observability stack. It needs to work across everything, not just within one vendor’s boundaries.

The platforms that build “native” orchestrators are doing exactly what you’d expect them to do: trying to keep you on their platform. That’s their job. Your job is to build a data architecture that serves your organization’s interests, not theirs.

So the next time someone tells you their platform’s native orchestrator is the future, remember:

• Oozie was the future. Until Hadoop wasn’t.
• Autosys was the future. Until mainframes weren’t.
• Informatica was the future. Until Salesforce decided otherwise.

History doesn’t always repeat itself. But it tends to rhyme.