Salesforce has made its direction clear. With the original launch of Agentforce and the broader shift toward headless architectures, the platform is moving from a system of record to a system where work is executed by agents.
The expectation was straightforward: once AI can understand requests and trigger actions, workflows would naturally become more automated (or autonomous).
But adoption hasn’t followed the original expectations.
Salesforce has actively tried to accelerate adoption, for example by shifting pricing from per-conversation models to more granular Flex Credits to lower the barrier to entry and encourage experimentation.
But lowering the barrier to try AI is not the same as making it work reliably in production.
What We Saw in New York World Tour 2026
At the Agentforce World Tour in New York, we asked a simple question: where are you with agentic AI?
This is clear: most teams are not in production.
Only a small number of teams have moved beyond pilots. Just 11% reported running agents across multiple processes at scale (graph below).
Most are still exploring, testing, or running limited use cases.
When asking about Agentforce specifically, the results are in the same vein, with the majority testing or exploring the technology (graph below).
Overall, it seems that agentic AI usage is starting to move from pure exploration to pilots.
That gap between expectation and reality shows up consistently in conversations.
The blockers are also telling. Data quality comes up first. Complexity and cost follow closely. It seems that cost concerns are emerging before use even scales to production.
With complexity and lack of control comes also security and governance concerns. What if something goes wrong?
So the bottom line is that teams don’t feel in control of how agents behave once they’re live.
That hesitation is not about raw AI capability. It’s about execution reliability.
This aligns with broader market signals, where many Agentforce (and other agentic AI) initiatives are still stuck in testing loops rather than scaling into production environments. There is no broad consensus of what controlled use of agentic AI actually looks like.
What Agentforce Changes and What It Doesn’t
Agentforce is a meaningful step forward.
It improves how work is understood by interpreting requests, extracting information, and triggering actions across systems.
But as we’ve explored in earlier articles, that doesn’t automatically translate into consistent outcomes. You can read more here:
- What the headless 360 actually means – or exposes – in execution
- Why workflows break despite all the automation
- Why triage matters in routing problems
- Why calling Salesforce is easy but what happens after that is not
The underlying challenge remains the same:
Work still needs to be routed, assigned, and progressed through workflows that are often fragmented and hard to control.
Agentforce operates on top of that reality. And according to our event survey:
- 78% of organizations still manually fix or reassign work after routing
- The most common frustrations are:
- too many flows
- constant manual fixes
- systems that are hard to manage
In many ways, Agentforce is acting as a forcing function – no pun intended. It exposes the condition of the underlying Salesforce environment, including data quality, process clarity, and workflow design. It’s not a magical fix to those problems.
Why This Becomes Visible With Agents
As soon as agents start interacting with real workflows, the system behind them becomes visible. Multiple flows. Multiple rules. Multiple integrations.
But no single place defining how work should actually be handled from start to finish.
That’s when teams hit a familiar pattern. It may not look like straight up failure, but more like… stalling?
They’ve proven agents can work, but still struggle to run them reliably and at enterprise scale.
The Maturity Gap
It helps to think about this whole agentic AI adoption as a progression.
At the early stage, AI generates answers. It supports users, but doesn’t execute work.
The next stage is where agents perform some tasks. They interact with systems and start to automate parts of workflows. This seems to be where most Agentforce adoption sits today (note: if it has been adopted).
The final stage is where AI executes business processes as part of normal operations.
The difference between those stages is control. Moving from pilots to production requires a way to define how work should be handled and ensure it executes consistently across systems. Without proper orchestration and execution control, more AI only increases variability.
Governance, Cost, and Control Become the Real Problem
In early agentic AI conversations, teams talk about use cases and capabilities. Later, the discussion changes towards governance and control:
How do we control what the agent is doing?
How do we ensure consistent outcomes?
How do we audit decisions?
How do we manage cost as usage scales?
This is something we heard a lot at the New York World Tour. They are fundamental concerns, because once agents are part of execution, every inconsistency becomes operational. If the workflow in question is a critical one for the business, the errors become critical, too.
And when it comes to cost control, even with the move to more granular pricing, cost predictability remains a concern. It is so especially as workflows become more complex and agent-driven actions scale.
From Execution Gap to Controlled Agentic Orchestration
At this point, the gap is clear. Teams aren’t getting stuck because they can’t use AI. They just can’t run workflows reliably once AI is involved.
As agents start operating across systems, any inconsistency in execution becomes immediately visible and harder to manage.
What’s changing now is the focus. It’s not blindly toward more AI, but toward making execution controllable.
An agentic orchestration layer like Ortoo’s defines how workflows run end to end, so agents, automation, and integrated systems operate within a single, unified structure.
Instead of relying on loosely connected logic and tools, work follows a defined path. Execution becomes consistent, decisions are traceable, and agents operate within clear boundaries.
With Ortoo, AI is applied selectively, only where interpretation is needed, while deterministic rules control execution.
AI can understand intent, extract key details, and classify requests.
But what happens next is not left to the LLM model.
Assignments, priorities, and follow-up actions don’t depend on how a model behaves in that moment. They follow a defined (deterministic) path, step-by-step.
This keeps outcomes consistent, while still allowing AI to handle the parts of the process that benefit from flexibility.
AI becomes a predictable part of the workflow, not the controller of it. Pricing is tied to the work being completed per case, lead, or request, not to how many actions happen along the way. That makes scaling predictable both operationally and financially.
There’s A Practical Way Forward
Salesforce is clear about what it takes to succeed with agentic AI. Its architecture guidance points toward what looks like a full redesign across data, workflows, and governance.
That direction is unboubtedly right. But for most teams, it’s also too much.
Most teams don’t struggle with the vision of an Agentic Enterprise.
They struggle with how to get there without stopping everything else.
In practice, moving to that model often feels like a larger than life transformation effort, so teams hesitate. They run pilots, test isolated workflows, and delay scaling.
This is where Ortoo takes a different approach. Instead of redesigning everything upfront, you start with orchestrating a single workflow.
Define how it should run.
Ensure it executes consistently.
Then expand from there.
There’s no big-bang transformation or need to rebuild the entire architecture before you see value.
That’s what makes agentic AI adoptable in real environments today.
Not just defining the ideal architecture, but providing a way to get agents into production, incrementally and under control.
Ready for the Next Step on Your Agentic AI Journey?
The companies that move forward won’t be the ones using the most AI.
They’ll be the ones that can control how work executes predictably.
We can help you bridge the gap from agentic workflow experimentation to controlled enterprise scale production. The team would love to talk to you!
