What Is AI Orchestration And Why Is It So Important?
Agents are like "functional experts" and something has to stitch them together
The world of AI agents has exploded. In the domain of HR, we’ve identified more than 100 functional agents in our HR Blueprint (these are explicit “practices”) which could and likely will become independent agents in the future.
How do they work together? Let me give you an example.
Most companies have a compensation and benefits department (or specialist). This team is responsible for setting pay bands by level, benchmarking pay against competitors, deciding how to handle cost of living adjustments, setting guidelines for raises during performance reviews, monitoring pay equity (inequities by gender, race, age, etc.), handling out of range conditions (hiring a specialist or guru with high pay), managing executive compensation, and the huge area of benefits plan design, benefits administration, benefits communication, and administration of things like stock options, bonuses, and special compensation.
And on a strategic level this group looks at budget allocation by business area, the impact of pay on retention, and how hourly rates should be adjusted for minimum wage jurisdictions, overtime, and skills-based pay adjustments.
It’s enormously complicated and many of these policies and decisions require external data on regulatory changes (around the world), benchmarks, and variations in pay by job role which change the “categories” of workers. And none of this includes payroll processing, which in itself is a highly complex global processs.
It’s a perfect opportunity for an AI Agent. Presumably this agent, if it was designed carefully, would have “rules” and workflows for all these use-cases, and each of these policies and guidelines would be programmed (ie. written) into the Agent’s code base. And in today’s world all these rules will be written in clear language (not code), so they’re easy to understand.
The “owner” or “supervisor” of this Agent would be the head of compensation and benefits. His or her team would build the agent, monitor it, and “care for it” by making sure it’s getting good data, it has connections to the HCM and payroll system, and its various rules and workflows are accurately implemented.
And in this case (as with most HR operations), the Agent doesn’t make all the decisions - managers do. So the Agent has to interact with, explain, and teach managers and employees how the policy works and deal with all sorts of special cases.
Theoretically it’s a perfect fit for AI. The AI knowledge in this agent would see the entire company’s pay, find outliers, and point out many options for strategic change. (ie. maybe the benchmark pay for certain jobs or skills has dropped and all of a sudden the company is overpaying people).
Where Does This Agent Fit? Today vs. Future.
Today this body of process and expertise fits into a “Center of Excellence” (COE) of some kind of “department.” And this group of humans does all this work and talks with execs, leaders, and HR business partners to teach them, inform them, and get guidance from them on hundreds of conditions, decisions, and policies. Most companies delegate pay decisions to managers and set guidelines (and reviews) to make sure the process is audited.
This process itself is then institutionalized in the HCM platform (Workday, SuccessFactors, HiBob, Oracle, whatever), and that platform keeps a meticulous history so you can go back and audit, analyze, or find discrepancies or fraud. So the Agent may periodically look at history to analyze trends or find anomolies.
Whew!
How would this work into the future?
Well for this one functional area of HR (and I would call pay and rewards a “Superagent” because benefits, payroll, exec comp, etc. could be sub-agents), the Agent or Superagent has to talk with dozens of things. Let’s just imagine a few.
Hiring: when a manager opens a job requisition, he or she has to “level” the job, allocate budget, and negotiate a salary. They are going to want to know how high or low the salary should be and likely ask the Agent (or the HRBP, which may be an agent as well), how to negotiate for a highly competitive position.
Performance: when a manager does a review, he or she needs to know how much autonomy they have and what guidelines they can use. Believe me this is never a simple process, because sometimes they have to make difficult decisions.
CFO mandates: if the company is doing a layoff or cost-cutting the pay against benchmark may change. So the CFO is likely to work closely with the agent to make sure salaries and benefits are not moving in the wrong direction.
Legal: every now and then a union negotiation comes up or an employee files a discrimination or other lawsuit. The Agent has to justify it’s decisions and possibly support an ongoing investigation or legal filing.
Promotion: when someone is promoted the Agent has a standard raise of some kind, but the manager or HRBP wants to discuss the particular situation with the agent for special cases.
Career ladders and skills: most companies are moving to skills-based pay, jobs, and operations. (I just met with a $20 billion services company last night that spent 18 months building an enterprise skills model which they now use for project cost allocation, employee resource allocation, employee assessment, and development. Pay has to factor into all this). Remember that all skills are constantly changing, so at times the Agent has to understand that highly rare or difficult to hire skills demand pay adjustments. And employees who are “certified” or licensed may as well.
Anyway, you can see that this one “Agent” has a lot of inputs, rules, interfaces, decisions, and external processes and people to support. How do we “orchestrate” all this logic, decision-making, process, approvals, and support?
Orchestration Enables Redesign
Well we need an “Orchestrator” (both within this agent and across agents), and in that “orchestration” layer we both set policies and decide how to re-engineer our company.
IBM, for example, built a compensation and benefits agent many years ago (I’ve talked with IBM’s CHRO about in the past) and that agent not only managed pay levels, it “learned” now to give managers advice during performance reviews. Over the years of it’s life IBM taught it to look at employee work projects (IBM has many automated systems) and they recently reached a point where the agent “tells” the manager what an employee’s annual raise should be.
Building this agent took a decade or so. IBM started this under the Watson regime and today I’m guessing it’s even more integrated into its HR AI architecture with even better data, rules engines, and policy management. I’ve now talked with three companies that built “regulatory monitoring agents” that monitor changes in employment rules around the world (good service for vendors - Galileo does this to a degree now) so that “policy monitoring agent” can keep the pay agent up to date.
And as their agent grew and people trusted it, IBM stripped out more and more old fashioned process. I’m not fully aware of the process today (I’ll reach out and get an update), but I’d imagine many of the old “talent review” steps are gone. So the “Orchestration layer” is an area for process redesign, productivity improvement, and ultimately autonomy.
(At some point IBM likely removed the human oversight of these adjustments, or at least could if they wanted.)
And if you think bigger, imagine this. You’re the recruitment agent and one of your most promising candidates keeps negotiating for a higher pay. Your recruitment Superagent negotiates with the Pay Superagent and they make a decision. Perhaps no human is involved.
Orchestration Is Value Creation
I am by no means an expert on orchestration software (frameworks like Langchain and the many orchestration tools from each AI vendor), but we are very aware of the importance of these layers. So if you’re shopping for AI solutions (from incumbent vendors or new ones), make sure you ask them how their orchestration layer works.
In most cases this is immature, new, and evolving. Various rules engines, rubrics, monitoring tools, and feedback systems need APIs and workflow tools built in. And as you can see from the example above, you’re going to want transparency (why did the agent make a decision? and how did it decide xyz?), monitoring (the agent can do this but not that), and help in visualizing and streamlining the process.
Today AI brings miraculous improvements to process design. We’re now working with a vendor (we’ll explain more soon) who has built a system that can decompose all the workflows you built in Workday or SAP Successfactors. They showed me a visual map of all the decision trees and manager approvals in a given client’s system, all generated by an AI “analysis” of the APIs and externally available interfaces in that system.
Imagine an agent tool (this is what they’re building) that would analyze all theses workflows and streamline or automatically embed them into your chosen AI agent platform. Not only would this give you one of the most amazing re-engineering tools ever, you’d be able to “orchestrate” a better employee experience and reduce HR errors and cost even easier than ever.
Imagine now that the tool is smart enough to “re-engineer” for you. (Galileo can do this today, it’s so aware of HR Processes that is even advises what to change.) See where this is going?
You Are The Consultant Now
This kind of work used to be the domain of consultants, consulting firms, IT, and HR technology product managers. Now, as orchestration tools expand, we are going to be able to do this ourselves.
Bottom Line: Architecture Really Matters
The bottom line is this. Regardless of how advanced you are in your own company’s AI transformation, you need to think of the architecture as early as you can. You can see that some orchestration belongs “in the agent” and some belongs “across the agents.”
As I explained in my keynote at Unleash this week - we’re moving to a much more flexible, intelligent, and ultimately autonomous architecture for operations (and HR). Decisions about how and where the “orchestration” goes are important.
(Here’s a “before and after” for onboarding for example.)
As we all build more and more intelligence into our agents, the way we orchestrate and manage these workflows becomes critical.
We have built out an entire Systemic HR AI Blueprint to help you. Contact us or consult Galileo to help - and our newest Galileo offering “Galileo for Consultants” actually has many of our tools and models for this kind of design built in.
I promise you that “Orchestration” will be a big focus for AI in the future. We’ll stay closely up to date and share all we’re learning. If you’ve built an orchestration solution for your company you’d like to share, please let us know. And join us at Irresistible 2026 (June 8-10 at the beautiful USC campus) to see lots of case studies and get a chance to talk with experts in Los Angeles.
How to Learn More
Our podcast (2-3 episodes a week now) is filled with stories, vendors, and tech discussions on this stuff.
Get Galileo for yourself. We put ALL our research, case studies, and models in there - and we update it every day. It’s like the Bloomberg Terminal, Encyclopedia Brittanica, and HR Expert with new research, vendors, and case studies every day. Our entire 2026 AI Imperatives research is all in Galileo.
Come to our amazing conference (Irresistible 2026) on June 8-10 at the beautiful USC campus to meet other leaders.
Or call us or meet with us at the conferences around the world.
(Here is a clip from this week’s podcast digging into HR tech and AI announcements.)