New Odyssey
Stop re-keying data across your systems: New Odyssey AI agents
Salesforce Agentforce works inside Salesforce. Here's how it compares with Workday AI agents and cross-platform agents when work spans five systems.
Ademola Afolabi
Founder
If your operations team re-keys the same data between platforms every week, an AI agent built for one vendor's platform won't take that work away. Salesforce Agentforce, for example, is built to act on Salesforce's own data and workflows. It can speed up steps that happen inside Salesforce, but it was not designed to carry a record from Salesforce into your HRIS, accounting or ticketing platform and write the result back. If your team's re-keying happens between platforms, you need agents that span your existing tech stack, with audit trails and workflow approvals your risk team will sign off on.
Your team is the human middleware between your systems
If your operations staff copy the same record from one platform to the next every week, they are acting as human middleware. They connect your ERP, HRIS, CRM, ticketing and document platforms because those platforms don't connect to each other. Every hand-off adds delay, and every re-keyed field is a chance for an error.
Human middleware exists because each platform in a mid-market or enterprise stack was bought to do one job. The HRIS holds people data, the CRM holds accounts and the accounting platform holds invoices. The work between them falls to people: re-keying, document handling, reconciliation and month-end close tasks.
The cost of human middleware grows with volume. When claims, onboarding and invoicing volumes climb while headcount stays flat, the hand-off work climbs too, and backlogs follow. Customers and employees wait longer for decisions.
So the question many operations leaders are now asking is simple: will Salesforce Agentforce, or the AI agent my other software vendor is selling, take this work off my team's plate?
What is Salesforce Agentforce, in plain terms?
Salesforce Agentforce is Salesforce's family of AI agents for tasks such as service, sales, marketing and commerce. Like other vendor-native agents, Agentforce is built around its own platform's data and workflows.
That design is a strength when the whole task lives in Salesforce. An agent that already understands your Salesforce objects, permissions and processes has less to learn. For feature detail, Salesforce's own documentation and Trailhead material are the right reference. This page focuses on a different question: fit, once your work leaves Salesforce.
Where does Salesforce Agentforce stop in a multi-vendor stack?
A vendor-native agent such as Salesforce Agentforce stops at the edge of its own platform, which is exactly where the hand-offs between platforms begin.
Consider a typical chain in a mid-market company. A new hire sits in BambooHR. Their candidate history is in Greenhouse. A related invoice is in Xero. The customer account they'll manage is in HubSpot. No single vendor's agent sees that whole chain, because each one is built around its own data.
The practical result: the steps inside Salesforce may get faster, while the copy-paste between Salesforce and everything else stays with your team. The human middleware doesn't go away; it moves to the edges.
A quick test you can run today:
- Pick one process your team handles every week.
- List every platform that process touches, from first input to final record.
- Count the vendors on that list.
If the answer is more than one vendor, an in-app agent covers only part of the process. If you want to see how Salesforce sits alongside the rest of a stack, our page on connecting Salesforce to the rest of your stack walks through it.
How do Workday AI agents compare with Salesforce Agentforce?
Workday's AI agents and Salesforce Agentforce follow the same model: each is built around its own platform's data and workflows. The difference between them matters less than the model they share.
For a team running Workday for HR and Salesforce for CRM, that model means two sets of agents, each stopping at its own edge. When a change in Workday needs to show up in Salesforce, or the reverse, a person still does the hand-off.
The more useful comparison for an operations leader is not Agentforce vs. Workday. It is in-platform agents vs. agents that span your existing tech stack.
| Approach | Where it works | Who builds and maintains connections | What happens at the hand-off between vendors |
|---|---|---|---|
| In-platform agents (e.g., Salesforce Agentforce, Workday AI agents) | Inside one vendor's platform | The vendor, within its own platform | Stays with your team |
| Build it yourself (frameworks or integration platforms) | Wherever your team builds it | Your team or an implementation partner | Automated if you build and maintain it |
| Cross-platform agents delivered as a service | Across several vendors in one workflow | The provider | Automated, with exceptions sent to a named owner |
What are cross-platform AI agents and how do they work?
Cross-platform AI agents read from and write back to several platforms in one workflow, so the output lands where it belongs and nobody re-keys it. When something doesn't fit the rules, the agent sends it to a named owner instead of a shared inbox.
Cross-platform agents usually connect through unified APIs rather than one-off connectors per vendor. For example, New Odyssey connects its agents to more than 200 HRIS, ATS, accounting, CRM, ticketing and file-storage platforms through Merge's unified APIs, so one workflow can span BambooHR, Greenhouse, Xero and HubSpot. For teams whose stack includes Microsoft 365 and ServiceNow as well as Salesforce, see workflows across Microsoft 365, Salesforce and ServiceNow.
Here is what that looks like in a privacy request workflow, using an illustrative sample from New Odyssey's Compliance Agent page. An agent connected to Salesforce, BambooHR and Workday discovered 1,240 records for a data subject access request. Of those, 1,131 could be released as-is. The remaining 109 went to a person to decide: 73 needed redaction and 36 needed a withhold-or-exempt decision from a privacy officer. The agent did the searching and sorting across three vendors; people made the judgment calls.
Can we build this ourselves with LangGraph or an integration platform?
Yes, building cross-platform agents in-house is possible. Agent frameworks such as LangGraph and integration platforms give your engineers the tooling to do it.
The catch is that integration tooling hands design, build and maintenance back to your team or an implementation partner. The first workflow is rarely the hard part. The hard part is everything that has to keep working after it ships.
The pieces teams tend to underestimate:
- Connectors to every vendor in the stack, and keeping them working when those vendors change their APIs.
- Audit trails that record every action an agent took.
- Versioned deployments, so you know which version of an agent made which decision.
- Workflow approvals and segregation of duties where a person must sign off.
- Exception routing to a named owner, not a queue nobody watches.
Before committing internal engineers, ask two questions. Who maintains the connectors a year from now? And who answers to audit when an agent's decision is questioned?
Why AI agent pilots stall at security and compliance review
AI agent pilots stall because security, compliance and audit teams won't sign off on agents whose actions and decisions they can't trace or explain. A pilot can work perfectly and still never reach production.
What earns a risk team's trust is evidence. Every agent action audited, every deployment versioned, every decision recorded with a human-readable explanation, plus workflow approvals and segregation of duties where the process requires them.
One example of a governance layer is CIP, the control plane under every New Odyssey agent. CIP audits each action, versions each deployment and records each decision with a human-readable explanation. It is open source under the MIT license, so security and compliance teams can inspect the governance code rather than take it on trust.
Apply the same checklist to any agent you evaluate, Salesforce Agentforce included. Can you get reporting on:
- What the agent did?
- When it did it?
- Why it made that decision, in words a reviewer can read?
- Who approved it, and which version of the agent was running?
If any answer is no, expect the pilot to stall at review.
How to decide: in-platform agent, cross-platform agent or neither
The right choice depends on where your process actually lives and whether the numbers justify automating it. Start with one process, not a platform strategy.
- Pick one high-volume process such as claims, onboarding or invoicing, and map every platform and hand-off it touches.
- Baseline the cost. Count the hours of re-keying per month and the errors your team corrects today. Without this number, you can't show leadership whether anything paid off.
- Match the agent to the map. If the whole process lives in Salesforce, Salesforce Agentforce may be the right fit. If it spans vendors, look for agents that span them too.
- Build a working version on real data before scaling. Test against your real platforms and volumes, not a demo environment.
- Accept "don't proceed" as a valid answer. If the measured savings don't cover the cost, stopping is a good outcome, not a failed project.
The goal is the same whichever route you choose: work flows between platforms on its own, output is written back where it belongs, nobody re-keys it, and exceptions reach a named owner.
To go deeper on how AI is changing integration across the platforms you already use, read the free guide What is Enterprise Integration AI? in the Insights section of the New Odyssey homepage.
Frequently Asked Questions
- Does Salesforce Agentforce work with systems outside Salesforce?
- Salesforce Agentforce is built around Salesforce's own data and workflows, so it fits best when the whole task lives inside Salesforce. Hand-offs between Salesforce and other vendors' platforms, such as your HRIS or accounting software, typically stay with your team.
- What is the difference between Workday AI agents and Salesforce Agentforce?
- Both follow the same model: each is built around its own platform's data and workflows. For a team running Workday for HR and Salesforce for CRM, that means two sets of agents, each stopping at its own edge, with the hand-off between them still done by people.
- What is human middleware?
- Human middleware is the work your staff do to connect platforms that don't talk to each other, such as re-keying records, handling documents and reconciling data between ERP, HRIS, CRM, ticketing and document platforms. It grows as process volumes rise while headcount stays flat.
- Can we build cross-platform AI agents in-house with LangGraph?
- Yes, frameworks such as LangGraph and integration platforms give engineers the tooling, but design, build and maintenance stay with your team or an implementation partner. Teams often underestimate the connectors, audit trails, versioned deployments, workflow approvals and exception routing that must keep working after launch.
- What do security and compliance teams need before approving AI agents?
- Risk teams generally need reporting on what each agent did, when, why and who approved it, with every action audited and every deployment versioned. Workflow approvals and segregation of duties help earn their trust before an agent reaches production.
Published
November 17, 2026
Last updated
November 17, 2026
Author
Ademola Afolabi · Founder
Reviewer
New Odyssey Editorial Team · Editorial Review
Methodology
This new odyssey piece is maintained by the New Odyssey editorial and delivery teams using current implementation patterns and operating assumptions.
Which of your workflows is worth automating?
Answer three questions about the process you want to automate and see what it costs you today.
Assess my workflowStay updated
Get weekly integration insights delivered to your inbox.
Related articles
- New OdysseyChoosing an AI Consultancy in London for Enterprise Automation Your Risk Team Will Trust
- New OdysseyEnterprise AI Ontology: Why Your Agents Need Shared Definitions Across Existing Tech Stacks
- New OdysseyEnterprise AI Transformation Stalls Between Your Systems. Here's How to Get Past the Pilot